cancers

Review
Application of Artificial Intelligence in Lung Cancer
Hwa-Yen Chiu 1,2,3,4 , Heng-Sheng Chao 1,5, * and Yuh-Min Chen 1,4

1 Department of Chest Medicine, Taipei Veterans General Hospital, Taipei 112, Taiwan;
[email protected] (H.-Y.C.); [email protected] (Y.-M.C.)
2 Institute of Biophotonics, National Yang Ming Chiao Tung University, Taipei 112, Taiwan
3 Division of Internal Medicine, Hsinchu Branch, Taipei Veterans General Hospital, Hsinchu 310, Taiwan
4 School of Medicine, National Yang Ming Chiao Tung University, Taipei 112, Taiwan
5 Institute of Biomedical Informatics, National Yang Ming Chiao Tung University, Taipei 112, Taiwan
* Correspondence: [email protected]; Tel.: +886-2-28712121

Simple Summary: Lung cancer is the leading cause of malignancy-related mortality worldwide.
AI has the potential to help to treat lung cancer from detection, diagnosis and decision making to
prognosis prediction. AI could reduce the labor work of LDCT, CXR, and pathology slides reading.
AI as a second reader in LDCT and CXR reading reduces the effort of radiologists and increases the
accuracy of nodule detection. Introducing AI to WSI in digital pathology increases the Kappa value
of the pathologist and help to predict molecular phenotypes with radiomics and H&E staining. By
extracting radiomics from image data and WSI from the histopathology field, clinicians could use AI
to predict tumor properties such as gene mutation and PD-L1 expression. Furthermore, AI could help
clinicians in decision-making by predicting treatment response, side effects, and prognosis prediction
in medical treatment, surgery, and radiotherapy. Integrating AI in the future clinical workflow would
be promising.

Abstract: Lung cancer is the leading cause of malignancy-related mortality worldwide due to its
heterogeneous features and diagnosis at a late stage. Artificial intelligence (AI) is good at handling



a large volume of computational and repeated labor work and is suitable for assisting doctors in


analyzing image-dominant diseases like lung cancer. Scientists have shown long-standing efforts to
Citation: Chiu, H.-Y.; Chao, H.-S.;
Chen, Y.-M. Application of Artificial
apply AI in lung cancer screening via CXR and chest CT since the 1960s. Several grand challenges
Intelligence in Lung Cancer. Cancers were held to find the best AI model. Currently, the FDA have approved several AI programs in CXR
2022, 14, 1370. https://doi.org/ and chest CT reading, which enables AI systems to take part in lung cancer detection. Following the
10.3390/cancers14061370 success of AI application in the radiology field, AI was applied to digitalized whole slide imaging
(WSI) annotation. Integrating with more information, like demographics and clinical data, the AI
Academic Editor: Francesco Petrella
systems could play a role in decision-making by classifying EGFR mutations and PD-L1 expression.
Received: 17 February 2022 AI systems also help clinicians to estimate the patient’s prognosis by predicting drug response, the
Accepted: 7 March 2022 tumor recurrence rate after surgery, radiotherapy response, and side effects. Though there are still
Published: 8 March 2022
some obstacles, deploying AI systems in the clinical workflow is vital for the foreseeable future.
Publisher’s Note: MDPI stays neutral
with regard to jurisdictional claims in Keywords: artificial intelligence; machine learning; lung cancer; radiomics; whole slide imaging;
published maps and institutional affil- survival prediction
iations.

1. Introduction
Copyright: © 2022 by the authors.
Lung cancer constitutes the largest portion of malignancy-related deaths worldwide [1].
Licensee MDPI, Basel, Switzerland.
This article is an open access article
It is also the leading cause of malignancy-related death in Taiwan [2,3]. The majority of
distributed under the terms and
the patients diagnosed with lung cancer are in the late-stage, and therefore have a poor
conditions of the Creative Commons prognosis. In addition to the late stage at diagnosis, the heterogeneity of imaging features
Attribution (CC BY) license (https:// and histopathology of lung cancer also makes it a challenge for clinicians to choose the best
creativecommons.org/licenses/by/ treatment option.
4.0/).

Cancers 2022, 14, 1370. https://doi.org/10.3390/cancers14061370 https://www.mdpi.com/journal/cancers

Cancers 2022, 14, 1370 2 of 17

The imaging features of lung cancer vary from a single tiny nodule to ground-glass
opacity, multiple nodules, pleural effusion, lung collapse, and multiple opacities [4]; simple
and small lesions are extremely difficult to detect [5]. Histopathological features include
adenocarcinoma, squamous cell carcinoma, small cell carcinoma, and many other rare
histological types. The histology subtypes vary even more. For example, at least six
common subtypes and a total of eleven subtypes of adenocarcinoma were reported in the
2015 World Health Organization classification of lung tumors [6], with more subtypes added
to the 2021 version [7]. Treatment options are heavily dependent on the clinical staging,
histopathology, and genomic features of the lung cancer. In the era of precision medicine,
clinicians need to collect all the features and make a decision to administer chemotherapy,
targeted therapy, immunotherapy, and/or combined with surgery or radiotherapy.
Whether to treat or not to treat the disease is always a question in daily practice. Clini-
cians would like to know the true relationship between the observations and interventions
(inputs) and the results (outputs). In other words, to find a model for disease detection,
classification, or prediction. Currently, this knowledge is based on clinical trials and the
experience of doctors. This exhausts the doctors in reading images and/or pathology
slides repeatedly to make an accurate diagnosis. Reviewing charts to determine the best
treatment options for patients also consumes a considerable amount of time. A good predic-
tion/classification model would simplify the entire process. Here, artificial intelligence(AI)
is introduced.
AI is a general term that does not have a strict definition. AI is an algorithm driven by
existing data to predict or classify objects [8]. The main components include the dataset
used for training, pretreatment method, an algorithm used to generate the prediction
model, and the pre-trained model to accelerate the speed of building models and inherit
previous experience. Machine learning (ML) is a subclass of AI, and is the science of
obtaining algorithms to solve problems without being explicitly programmed, including
decision trees (DTs), support vector machines (SVMs), and Bayesian networks (BNs). Deep
learning is a further subclass of ML, featured with multiple layered ML, achieving feature
selection and model fitting at the same time [9]. The hierarchical relationship between
those definitions is displayed in Figure 1.
However, to develop such a model, a large amount of computation is required. In the
past, building a multidimensional algorithm for image analysis has taken hours, or even
days, for the human brain. The large computational power required becomes a significant
obstacle in creating a sophisticated prediction model. The booming computational power
of chip technology and software optimization makes large and sophisticated calculations
easier to achieve [10,11]. When a large matrix can be computed in a short time, it is
possible to develop models that are much more complex than linear regression or logistic
regression. DTs and SVMs were used to build models in the ML era around the year
2000 [12]. By estimating the probability, BNs were used to select treatments by predicting
survival [13]. In recent years, deep learning models, including artificial neural networks
(ANNs), convolutional networks (CNNs), recurrent neural networks (RCNNs), long-term
and short-term memory (LSTM) [14], and generative adversarial networks (GANs) [15],
have outperformed most old models and are thus widely used in research and commercial
fields [16].
In the 21st century, human life has been largely integrated with AI, and this trend
also extends to the medical field. The heterogeneity of lung cancer makes it the best
field for AI application. A large number of studies have reported the application in
lung nodule detection, diagnostic application in histopathology, disease risk stratification,
drug development, and even prognosis prediction. In this article, we present a narrative
review of AI applications in lung cancer by introducing AI models first and then reported
applications according to the clinical workflow: screening, diagnosis, decision making, and
prognosis prediction. Table 1 listed the potential AI application fields in lung cancer.
 previous experience. Machine learning (ML) is a subclass of AI, and is the science of ob-
taining algorithms to solve problems without being explicitly programmed, including de-
cision trees (DTs), support vector machines (SVMs), and Bayesian networks (BNs). Deep
learning is a further subclass of ML, featured with multiple layered ML, achieving feature
Cancers 2022, 14, 1370 selection and model fitting at the same time [9]. The hierarchical relationship between3 of 17
those definitions is displayed in Figure 1.

Venndiagram
Figure1.1.Venn
Figure diagramofof artificial
artificial intelligence
intelligence (AI),
(AI), machine
machine learning
learning (ML),(ML),
neuralneural network,
network, deep deep
learning, and further algorithms in each category. AI is a general term for a program that predicts
learning, and further algorithms in each category. AI is a general term for a program that predicts an
an answer
answer to to a certain
a certain problem,
problem, where
where one
one ofofthe
the conventionalmethods
conventional methodsisislogistic
logisticregression.
regression.ML
MLlearns
learns the algorithm
the algorithm through
through inputinput
data data without
without explicit
explicit programming.
programming. MLML includes
includes algorithms
algorithms such as
decision trees (DTs), support vector machines (SVMs), and Bayesian networks (BNs). By using each
ML algorithm as a neuron with multiple inputs and a single output, a neural network is a structure
that mimics the human brain. Deep learning is formed with multiple layers of neural networks, and
convolutional neural network (CNN) is one of the elements of the famous architecture.

Table 1. Summary of AI application fields.

Screening Diagnosis Treatment

Radiology:
Tumor property classification:
CXR [17–21]
Drug selection [44–46,50,55–57]
CXR [17–21] Risk prediction:
Prognosis prediction:
LDCT [22–33] Radiomics [40–46]
Drug treatment response [58–60]
Novel tools: WSI [47–53]
Post-Surgery recurrence [54,61,62]
Genomics [34] WSI [47–53]
Radiotherapy response [63,64]
Genomics [34] Genomics [50,54]
Side effect estimation:
Proteomics [35,36]
Radiation pneumonitis [65,66]
Exhaled breath [37–39]
CXR: Chest X-ray, LDCT: low-dose computed tomography, WSI: whole slide imaging.

2. AI Models
Cancers 2022, 14, x FOR PEER REVIEW Numerous AI models are constructed with different algorithms are published 4nowa-
of 18
days. Generally, the AI models can be divided into: supervised learning, unsupervised
learning, semi-supervised learning [9], and reinforcement learning (Figure 2).

Figure 2. The concept map of supervised learning, unsupervised learning and reinforcement learning.
Figure 2. The concept map of supervised learning, unsupervised learning and reinforcement learn-
ing.

2.1. Supervised Learning

In supervised learning, researchers need to prepare the labeled dataset with both in-
Cancers 2022, 14, 1370 4 of 17

2.1. Supervised Learning

In supervised learning, researchers need to prepare the labeled dataset with both
inputs and desired outputs (answers) to train the algorithm. It is suitable to solve prediction
problems, such as classification and regression. The architecture of the algorithms varies.
Researchers can use multiple binary nodes to create DTs as a classifier, or find a plane
in a multidimensional space as a SVM classifier. Bayesian classifiers used input data to
calculate the probability of correct classification. With the probability calculated from the
above-mentioned algorithm, researchers can turn the answer into a continuous variable to
solve regression problems and vice versa. Most AI applications predicting survival [13,59],
cancer risk [34–39], nodule detection [22,23], and nodule characteristics [33] are based on
supervised learning.

2.2. Unsupervised Learning

In unsupervised learning, the algorithm divides the samples according to the inputs
by itself. Labeled data are not necessary. It is suitable to do clustering, to find associations
between samples, and to do dimensionality reduction. For example, cluster analysis was
used to find oncogenes in lung cancer [67,68].

2.3. Semi-Supervised Learning

Though supervised learning provides a more accurate algorithm, the labeled data are
relatively rare, and the labeling process is labor intensive. Unsupervised learning can adopt
unlabeled data but the algorithm is less accurate. Therefore, semi-supervised learning
could have both of the advantages when using supervised learning to generate a labeling
tool and use supervised learning to generate a large scaled labeled dataset for further
training [52].

2.4. Reinforcement Learning

Reinforcement learning is a reward-based system. The algorithm evolves as it interacts
with the environment (dataset). A reward function is used to adjust the algorithm or
the network. This type of AI is famous for playing chess, shogi, and Go through self-
play [69] or generating data with GANs [70]. With this technique, researchers can develop
a self-evolving AI for nodule hunting on CT images and achieve better accuracy [15,71,72].
In conclusion, there is no best method to build AI models for all. The best method
should be tailored according to the clinical question and the dataset used for training.

3. Screening
Approximately 7% of patients diagnosed with lung cancer are asymptomatic [73], and
more than half of the patients who underwent lung cancer resection were asymptomatic [74].
Several attempts at screening have been made, including imaging, sputum cytology [75,76],
blood test screening [77,78], and breath test [79,80]. However, only image screening is able
to provide the relevant clues. Although chest X-rays (CXRs) are widely used clinically, low-
dose computed tomography (LDCT) is the only method that has been proven to diagnose
lung cancer earlier and extend the survival of lung cancer patients [81,82].
The reading workflow of repetitive imaging provides room for AI to participate,
because human eyes become sore and images start to blur after reading images for a long
time. Furthermore, mistakes in reading CXR or LDCT images occur, and it constitutes a
large number of malpractice law suits [83]. Though experts were shown to detect more
pulmonary nodules on CXRs [84], approximately 20% of lung nodules <3 cm are missed
by radiologists [85]. In the 21st century, the prediction accuracy of pulmonary nodules on
CXRs has improved with the computer-aided diagnosis systems or AI-based programs. The
sensitivity of radiologists improves from 65.1% to 70.3% with the assistance of AI and the
false negative rate decreases from 0.2 to 0.18, changing the diagnosis in 6.7% of the cases [17].
In CT images, the sensitivity of lung nodules were more than 90% spotted by AI-based
programs [23]. Integrating AI into lung cancer screening protocol is an ongoing event.
Cancers 2022, 14, 1370 5 of 17

3.1. DICOM Format

Digital imaging and communications in medicine (DICOM) is the standard format
for image restoration and transfer to enable communication between different servers,
manufacturers, and hospitals [86]. The DICOM not only carries pixel data of the image
file but also a patient identification number, image type, machine-related parameters, and
other information in a format managed by the Medical Imaging and Technology Alliance, a
division of the National Electrical Manufacturers Association. After its first publication in
1993, DICOM changed the workflow of radiology, allowing image data to be transmitted
quickly and to be analyzed by computers. Later, huge datasets were established for data
sharing, model training, or as a benchmark for model testing, and are shown in Table 2.

Table 2. Summary of frequently used datasets for model training.

Database Year Material Volume Features

Contains 100 CXRs with malignant nodule, 54 CXRs with benigh
JSRT [87] 1998 CXR 154
nodule, and 93 normal CXRs
Contains 326 normal CXRs, and 336 CXRs with tuberculosis. Ribs
Shenzhen CXR set [88] 2012 CXR 662
were labeled.
Contains 80 normal CXRs, and 58 CXRs with tuberculosis. Ribs
Montgomery CXR set [88] 2014 CXR 138
were labeled.
Classified into 8 features: atelectasis, cardiomegaly, effusion,
ChestXray8 [89] 1992–2015 CXR 108,948
infiltration, mass, nodule, normal, pneumonia, and pneumothorax
Classified into 14 features: atelectasis, cardiomegaly, consolidation,
ChestXray14 [89] 1992–2015 CXR edema, effusion, emphysema, fibrosis, hernia, infiltration, mass,
nodule, pleural thickening, pneumonia, pneumothorax.
Labeled with 174 different radiographic findings, 19 differential
PadChest [90] 2009–2017 CXR >160,000
diagnoses and 104 anatomic locations
LIDC [91] 2011 LDCT 1018 Nodules were annotated and labeled with nodule sizes
Adapted from LIDC, with additional nodules found during model
LUNA16 [23] 2016 LDCT 888 training.
1186 lung nodules annotated in 888 CT scans
Classified into 14 labels derived from two natural language
MIMIC-CXR [92] 2011–2016 CXR 377,110
processing tools.
Labeled with 14 features: no finding, enlarged cardiom,
cardiomegaly, lung opacity, lung lesion, edema, consolidation,
ChestXpert [93] 2019 CXR 224,316
pneumonia, atelectasis, pneumothorax, pleural effusion, pleural
other, fracture, support devices
VinDr-RibCXR [94] 2020 CXR 18,000 Rib suppression images
RadGraph [95] 2021 CXR 500 Inference dataset of MMIC-CXR and reports
Labeled by 5 radiologists and synchronized sets of eye-tracking
REFLACX [96] 2021 CXR 3032
data and timestamped report transcriptions
CXR: chest CX-ray set, JSRT: Japanese Society of Radiological Technology, LIDC: Lung Image Database Consortium,
LUNA: LUng Nodule Analysis, REFLACX: Reports and Eye-Tracking Data for Localization of Abnormalities in
Chest X-rays.

3.2. CXR
CXRs are the most frequently used imaging modality in the medical field. With
0.1 mSv radiation exposure, similar to 10 days of natural background radiation, CXR
provides a good examination of the patient’s thorax. Far before digital imaging, the
computer-aided diagnosis(CAD) system for CXR has been developed since the 1960s [97].
Image features, such as shape, size, intensity, and texture, must be manually labeled before
being sent for further analysis. In the digital era, computers can directly analyze images.
By computing the image pixel-by-pixel, radiomics expands the definition of image features
from a computer perspective. By computing the image texture and density using different
mathematical techniques, the region of interest area can be converted to higher dimension
data and expressed as a huge matrix. Because the principle of radiomics is math, the various
image qualities of CXR give the computer another task. To obtain accurate radiomics data,
Cancers 2022, 14, 1370 6 of 17

image augmentation is an important procedure before nodule detection [98], including

pre-processing, lung segmentation [88], and rib suppression [94].
Further malignancy/benign classification was performed using a different algorithm.
DT-based algorithms were widely used to analyze these features before 2011. Later, deep-
learning-based algorithms demonstrated their power in image analysis. CheXNet, a
radiologist-level deep learning algorithm trained on Chest-Xray14, one of the largest
CXR databases in the world, exceeds radiologist performance in the detection of 14 pul-
monary diseases, including lung nodules and lung masses with an area under the receiver
operating curve (AUROC, AUC) 0.78 and 0.87, respectively [18]. Further deep learning
models pushed the sensitivity to 0.83 at a false-positive rate of 0.2 per CXR [19]. Currently,
several software programs have been approved by the FDA [17,20,21].

3.3. Chest CT
CT technology provides a noninvasive method to explore the 3-dimensional structure
of the thorax. As the technology advanced, the radiation exposure has reduced from 7 mSv
(conventional chest CT) to 1.6 mSv (LDCT). Screening with LDCT showed an approxi-
mately 20% mortality reduction in two large randomized control trials: the National Lung
Screening Trial (NLST) [81] and the Dutch-Belgian Randomized Lung Cancer Screening
Trial (Dutch acronym: NELSON study) [82]. The Multicentric Italian Lung Detection
(MILD) trial showed that prolonged LDCT screening for more than five years reduced
lung cancer mortality and overall mortality at ten years. These trials boosted the demand
for image reading. The application of AI in LDCT reading can help radiologists reduce
laborious work, minimize reader variability, and improve screening efficiency [99,100]. The
main task for AI application in LDCT reading is the same as in CXR: nodule detection and
classification/malignancy prediction. However, unlike CXR, the radiodensity of LDCT is
based on an international standard scale with a Hounsfield unit (HU) and fixed resolution.
The preprocessing of the CT images focused on denoising, resizing, and lung segmentation.
Many studies have used AI algorithms to detect lung nodules in chest CT
images [101,102]. Because they used different models on different datasets and evaluated
the models with different benchmarks, such as sensitivity, specificity, AUC, and accuracy, it
was difficult to evaluate the models scientifically. A series of grand challenges, such as the
Automated Nodule Detection 2009 (ANODE09) study [22] and the Lung Nodule Analysis
2016 (LUNA16) challenge [23] were conducted to find the benchmark model of nodule
detection on CT. The best algorithm achieved a sensitivity of 97.2% at one false-positive
rate per scan. AI has also been proven to help radiologists increase the sensitivity of nodule
detection [24–26] and reduce interpretation time. AI has proven to be a good concurrent
reader or a second reader. It was noticeable that consumer AI was not approved as the
first reader, in case that the radiologists may not have the chance to access to the AI-missed
nodules.
Lung nodule classification and malignancy prediction are important tasks in nodule
detection. Nodules are classified according to their texture as solid, part-solid, or non-solid,
and their size. An AI model trained on the MILD trial [103] dataset and externally validated
on the Danish Lung Cancer Screening Trial (DLCST) [104] showed that AI performed equiv-
alently to a human expert on differential six textures (sold, part-solid, non-solid, calcified,
perifissural, and speculated) [27]. The classification was then used to predict the malig-
nancy probability as recommended by the Lung CT Screening Reporting and Data System
(LUNG-RADS) [105] and the Fleischner guideline [106]. Traditionally, researchers have
used AI to classify lung nodules as a feature extraction step, and then go for malignancy
prediction. Later, researchers substituted the classification step with radiomics’ feature
extraction to increase prediction accuracy [28–30].
Similar to nodule detection on CT, challenges were conducted to compare the predic-
tion models. In 2015, the LUNGx Challenge for computerized lung nodule classification
was established. Ten teams sent their reports with AUC between 0.50 and 0.68, and only
three of them performed statistically better than a random guess, while radiologists per-
Cancers 2022, 14, 1370 7 of 17

formed with AUC between 0.70–0.85. The technology advanced rapidly in the ISBI 2018
Lung Nodule Malignancy Prediction Challenge, and 11 participants completed the chal-
lenge with an AUC between 0.70–0.91. The top five participants used deep learning models
with AUC between 0.87–0.91 without significant differences from each other [31]. The
accuracy was 93% with a sensitivity of 82% and precision of 84% based on the weighted
voting method of the autoencoder, ResNet, and handicraft features [32], and 96% with deep
convolutional network learning (DCN) [33].

3.4. Novel Screening Tests

Genomics [107], proteomics, microbiomes [108], and exhaled breath [109] are novel
screening tools for lung cancer [110]. Although these screening methods yield a large set of
signals for each patient, an advanced algorithm would elevate the diagnostic yield.
Genomics is one of the most popular topics in oncology. With the polymerase chain
reaction (PCR) amplification method and related technology, scientists can now analyze the
whole genome [111], exome, transcriptome [112], and epigenome of cancers and produce
large sets of information about patients and their tumors. By analyzing whole-genome and
whole-transcriptome sequencing data from treatment-naïve patients in The Cancer Genome
Atlas [113] (TCGA), the machine learning model successfully discriminated cancer-free
healthy controls from patients with cancer [34].
Proteins and other metabolites acquired from plasma and urine samples are relatively
easy to obtain and have been studied as a screening tool for lung cancer for decades. To
handle the extremely large number of variables produced by proteomics, researchers have
used machine learning methods to reduce dimensionality and feature selection [35,36]. In
2003, the machine learning methods were applied to analyze 1676 original and 124 pre-
screened mass spectra data from 24 diseased and 17 healthy specimens, and researchers
successfully built predictive models to discriminate lung cancer specimens from healthy
specimens. However, the most accurate predictions were obtained using less interpretable
models [35]. Following this idea, the urine proteome combined with machine learning
analysis successfully established models that can discriminate lung cancer samples not
only from healthy ones, but also from samples from other cancers [36].
Exhaled breath is composed of volatile organic compounds (VOCs) and exhaled
breath condensates [79,109]. To date, more than 3000 VOCs have been identified to be
related to lung cancer [114], however, not a single VOC could be accurate enough for
diagnosis. Therefore, a composite prediction model is a way to solve this problem, in
addition to escalating sensor technology. In 2018, a logistic regression model that was able
to discriminate patients with lung cancer in both smokers (sensitivity, 95.8%; specificity,
92.3%), and non-smokers (sensitivity, 96.2%; specificity, 90.6%) [37]. Using this, an SVM
model [38] and an ANN model [39] were created.

4. Diagnosis
When a nodule is detected, clinicians must know the properties of the lung nodule. The
gold standard is to acquire tissue samples via either biopsy or surgery. The image features
provide a way to guess the properties of the lung nodule by radiomics as mentioned in
the previous section. Aside from imaging features, the histopathological features also
affect further treatment. Following the path of digital radiology, whole slide imaging
(WSI) has opened the trend of digital histopathology. With digitalized WSI data, AI can
help pathologists with daily tasks and beyond, ranging from tumor cell recognition and
segmentation [47], histological subtype classification [48–51], PD-L1 scoring [52], to tumor-
infiltrating lymphocyte (TIL) count [53].

4.1. Radiomics
Following the idea of radiomics in nodule detection and malignancy risk stratification,
radiomics was applied to predict the histopathological features of lung nodules/masses [40].
Researchers used logistic regression of radiomics and clinical features to distinguish small
Cancers 2022, 14, 1370 8 of 17

cell lung cancer from non-small cell lung cancer with an AUC of 0.94 and an accuracy of
86.2% [41]. The LASSO logistic regression model was used to classify adenocarcinomas
and squamous cell carcinomas in the NSCLC group [42]. Further molecular features such
as Ki-67 [43], epidermal growth factor receptor (EGFR) [44], anaplastic lymphoma kinase
(ALK) [45], and programmed cell death 1 ligand, (PD-L1) [46] were also shown to be
predictable with AI-analyzed radiomics, a non-invasive and simple method.

4.2. WSI
The emergence of WSI is a landmark in modern digital pathology. The WSI depends
on a slide scanner that can transform glass slides into digital images with the desired
resolution. Once the images are stored on the server, pathologists can view them on their
personal computers or handheld devices. Similar to DICOM in diagnostic radiology, in
2017, the FDA approved two vendors for the WSI system for primary diagnosis [115,116].
Meanwhile, the DICOM also planned support for WSI in the PACS systems to facilitate
the adaption of digital pathology in hospitals and further information exchange [117,118].
These features enable the building of a digital pathology network to share expertise for
consultations and make education across the country possible [119].
Each WSI digital slide is a large image. It may contain more than 4 billion pixels and
may exceed 15 GB when scanned with a resolution of 0.25 micrometers/pixel, referred to
as 40× magnification [118,120]. With recent advances in AI and DL in image classification,
segmentation, and transformation, digitalized WSI provides another broad field to play.
There are many applications for deep learning in cytopathology.

4.3. Histopathology
Detecting cancerous regions is the most basic and essential task of deep learning in
pathology. Some models combine the detection, segmentation, and histological subtyping
together [47–49]. Accuracy depends on the data quality, quantity, and abundance of the
malignant cell differentiation status. It is difficult to perform histological subtyping of lung
cancer without special immunohistochemistry (IHC) staining. This causes inter-observer
disagreement when reading H&E staining. While the agreement between pathologists
came to a Kappa value of 0.485, a trained AI model can achieve a Kappa value of up to
0.525 when compared with a pathologist [48]. In the detection of lymph node metastasis,
a well-trained AI model can help reduce human workload and prevent errors [121]. It
obviously performs better than a pathologist in a limited time and has a greater detection
rate of single-cell metastasis or micro-metastasis [121].
Although WSI with H&E-stained slides is designed to view the morphology of tissues,
with the aid of AI, researchers have designed methods to predict specific gene mutations,
PD-L1 expression level, treatment response, and even the prognosis of patients. Focusing
on lung adenocarcinoma, Coudray et al. developed an AI application using Inception-V3
for the prediction of frequently appearing gene mutations including STK11, EGFR, FAT1,
SETBP1, KRAS, and TP53 [50]. The AUC of this prediction reached 0.754 for EGFR and
0.814 for KRAS which can be treated with effective targeted agents. Sha et al., used ResNet-
18 as the backbone to predict the PD-L1 status in NSCLC [55]. Their model showed an
AUC between 0.67 and 0.81, while different PD-L1 cutoff levels were chosen. They believed
that the morphological features may be related to PD-L1 expression level.
Next-generation sequencing (NGS) plays an important role in modern lung cancer
treatment [122]. Successful NGS testing depends on a sufficient number of tumor cells
and tumor DNA. AI can assist in determining tumor cellularity [123,124]. In addition, a
trained AI can help count the immune cells, while the tissue specimen is adequately stained
for special surface markers [53]. Since the PD-L1 expression level is the key predictor
for immunotherapy in lung cancer, AI has been trained to count the proportion score for
PD-L1 expression [52,125]. When properly stained, computer-aided PD-L1 scoring and
quantitative tumor microenvironment analysis may meet the requests of pathologists, and
eliminate inter-observer variations and achieve precise lung cancer treatment [126].
Cancers 2022, 14, 1370 9 of 17

However, there are several barriers to the translation of AI applications into clinical
services. First, AI applications may not work well when applied to other pathology
laboratories, scanners, or diverse protocols [127]. Second, most AIs are designed for their
own unique functions. Users are requested to launch several applications for different
purposes and spend a lot of time transferring the data. Medical devices powered by
AI applications require approval by regulations. Most articles and works were in-house
studies and laboratory-developed tests. All of these barriers may restrict the deployment
of trained AI models in daily clinical practice [119].

4.4. Cytology
The WSI for cytology differs from pathology. Cytology slides are not even sliced flat
layers. Instead, they have an entire cell on the glass and would be multiple cell layers.
Cytologists tend to use the focus function and look into the cells. While digitalizing
the cytology glass slide, the focus function was simulated through the Z-stack function
and multiple layers of different focus [128,129]. This method yields a larger WSI file,
approximately 10 times that of a typical histological case. Multiple image layers also
increase complexity and pose challenges to AI applications.
Few articles have discussed cytology, especially those focusing on lung cancer. For thy-
roid cancer, Lin et al. proposed a DL method for thyroid fine-needle aspiration (FNA) sam-
ples and ThinPrep (TP) cytological slides for detecting papillary thyroid carcinoma [130].
The authors did not claim the ability to detect other cell types of thyroid cancer using their
method. AI can be performed for various cytology samples from lung cancer patients,
including pleural effusion, lymph node aspiration, tissue aspiration samples, and endo-
bronchial ultrasound-guided fine-needle aspiration (EBUS-TBNA) of mediastinal lymph
nodes.

5. Decision Making and Prognosis Prediction

Oncologists prefer to deploy this technique to its limits. There are many exciting
possibilities for the use of the AI technique. By predicting treatment response, including
survival and adverse events, AI was proven to have the potential to play a role in clinical
decision making [13], to help surgeons choose the specific groups of patients to receive
surgery, and to aid radiotherapists in planning the radiation zone.

5.1. Medication Selection

In late-stage lung cancer, the identification of driver mutations, PD-L1 expression, and
tumor oncogenes affects most the treatment of choice. Using WSI and radiomics, AI could
help to identify EGFR mutations [44,50], ALK [45], and PD-L1 expression [46,55,56]. EGFR
mutation subtypes have also been classified using radiomic features [57].
Another research point is the use of radiomics, WSI, and clinical data to directly predict
cancer treatment response or survival [131]. Dercle et al. retrospectively analyzed the data
from prospective clinical trials and found that the AI model based on the random forest
algorithm and CT-based radiomic features predicted the treatment sensitivity of nivolumab
with an AUC of 0.77, docetaxel with an AUC of 0.67, and gefitinib with an AUC of 0.82 [58].
CT-based radiomics models have also been reported to predict the overall survival of lung
cancer [59,60].
One patent application publication declared that using radiomics features of seg-
mented cell nuclei of lung cancer can predict responses to immunotherapy with an AUC
up to 0.65 in the validation dataset [132]. Although there is no specific survival prediction
model for lung cancer, Ellery et al. developed a risk prediction model using the TCGA
Pan-Cancer WSI database including lung cancer [133]. However, the DL algorithm did
not provide acceptable prediction power for lung adenocarcinoma or lung squamous cell
carcinoma.
Cancers 2022, 14, 1370 10 of 17

5.2. Surgery
The gold standard for the treatment of early-stage lung cancer is surgical resection.
AI was applied to pre-surgical evaluation [61,62], and prognosis prediction after surgery,
and could help identify patients who are suitable to receive adjuvant chemotherapy after
surgery [54].
In pre-surgical evaluation, radiologist-level AI could help predict visceral pleural
invasion [62], and identify early stage lung adenocarcinomas suitable for sub-lobar resec-
tion [61]. After surgery, AI could play a role in predicting prognosis. The model based
on radiomic feature nomograms could identify high-risk groups whose postsurgical tu-
mor recurrence risk is 16-fold higher than that of low-risk group [134]. The CNN model
pre-trained with the radiotherapy dataset successfully predict a 2-year overall survival
after surgery [135]. The model integrating genomic and clinicopathological features was
able to identify patients at risk for recurrence and who were suitable to receive adjuvant
therapy [54].

5.3. Radiotherapy
SBRT is currently the standard of care to treat early-stage lung cancer and/or provide
local control for patients who are medically inoperable or refuse surgery. Radiomics-based
models have been reported to successfully predict 1-year tumor recurrence via CT scans
performed after 3 and 6 months of SBRT [63]. Lewis and Kemp also developed a model
trained on TCGA dataset to predict cancer resistance to radiation [64]. As a well-known
side effect of radiotherapy, radiation pneumonitis can be lethal, and clinicians would like
to prevent this situation. The AI model based on pretreatment CT radiomics was superior
to the traditional model using dosimetric and clinical predictors in predicting radiation
pneumonitis [65]. Another ANN algorithm trained with radiomics extracted from a 3D
dose map of radiotherapy has been shown to predict the acute and late pulmonary toxicities
with an accuracy of 0.69 [66]. A well-designed prediction model for radiation pneumonitis
may help to prevent radiation pneumonitis in the future.

6. Future Development
The future of AI applications in lung cancer could focus on integration and applica-
tions. First, because AI is a data-driven technology, scientist can integrate small datasets to
create large data sets for training. However, regulations regarding data sharing are a huge
obstacle for researchers. Federated learning, a method that shares the trained parameters
rather than sharing the data, is a simple solution [136,137]. In federated learning, the
models were trained at each different hospitals separately and only the trained models
were sent to the main server, so that the main server does not touch the raw data directly.
The final model was then reported back to individual hospitals (Figure 3).
Second, most previous researches were conducted by separate specialists and focused
on separated fields such as radiology, pathology, surgery, or clinical oncology. However,
integrating all aspects such as radiology, pathology, demographics and clinical data, and
both old and new technologies could better reflect reality. The combination of different
features also helps researchers build predictive models [138,139]. This brings about the
idea of multi-omics [140,141] or “Medomics” [40]. Similar to multidisciplinary teams in
clinical lung cancer treatment [142,143], the combination of different domain knowledge
and multidisciplinary integration is worth pursuing in the future.
Apart from improvement in model accuracy by increasing the training sample size and
multidisciplinary integration, another issue is the application of AI programs. Although
the studies above all showed the promising results of applying AI in lung cancer and
some products were approved by the FDA [17,20,21,116,144], real implementation of the
clinical workflow is rare. The user interface, speed of data analysis, expanse of the AI
program, internet bandwidth, and resources consumed by the AI program are all barriers
to real-world applications. More infrastructure needs to be constructed before we can enter
the AI-assisted world.
 to create large data sets for training. However, regulations regarding data sharing are a
huge obstacle for researchers. Federated learning, a method that shares the trained pa-
rameters rather than sharing the data, is a simple solution [136,137]. In federated learning,
the models were trained at each different hospitals separately and only the trained models
were sent to the main server, so that the main server does not touch the raw data directly.
Cancers 2022, 14, 1370 The final model was then reported back to individual hospitals (Figure 3). 11 of 17

Figure
Figure3.3.The
Thecomparison
comparisonofoftraditional
traditionalAI
AIserver
serverarchitecture
architectureand
andfederated
federatedlearning
learningserver
serverarchi-
archi-
tecture.
tecture.(a)
(a)In
Intraditional
traditionalserver
serverarchitecture,
architecture,the
themain
mainserver
serverprocesses
processesall
allthe
theraw
rawdata
dataatatthe
thesame
same
site,
site,leading
leadingto toconcerns
concernsabout
aboutprivacy;
privacy;(b)
(b)In
Infederated
federatedlearning,
learning,the
thedatasets
datasetsareareprocessed
processedatateach
each
individual
individualsitesiteand
andonly
onlythe
thetrained
trainedmodels
modelsareareshared
sharedwith
withthe
themain
mainserver.
server.Privacy
Privacyof
ofeach
eachdataset
dataset
isisprotected.
protected.

Second, most previous researches were conducted by separate specialists and fo-
Author Contributions: Conceptualization, H.-Y.C.; methodology, H.-Y.C. and H.-S.C.; writing—
cused on separated fields such as radiology, pathology, surgery, or clinical oncology.
original draft preparation, H.-Y.C. and H.-S.C.; writing—review and editing, H.-S.C. and Y.-M.C.;
However,
supervision,integrating
Y.-M.C. Allall aspects
authors such
have readasand
radiology,
agreed topathology,
the publisheddemographics
version of the and clinical
manuscript.
data, and both old and new technologies could better reflect reality. The combination of
different
Funding: features alsowas
This research helps researchers
funded buildofpredictive
by the Ministry Science andmodels [138,139].
Technology, Taiwan, This
grant brings
number
MOST 110-2321-B-075-001.
about the idea of multi-omics [140,141] or “Medomics” [40]. Similar to multidisciplinary
teams in clinical lung
Acknowledgments: cancer
Thanks treatment
to the Ministry of[142,143], theTechnology,
Science and combination of different
Taiwan, domain
for funding.
knowledge and multidisciplinary integration is worth pursuing in the future.
Conflicts of Interest: The authors declare no conflict of interest. The funders had no role in the design
Apart
of the study;from
in theimprovement in model
collection, analyses, accuracy by
or interpretation increasing
of data; the training
in the writing sample sizeor
of the manuscript,
and multidisciplinary
in the decision to publishintegration,
the results. another issue is the application of AI programs. Alt-
hough the studies above all showed the promising results of applying AI in lung cancer
References and some products were approved by the FDA [17,20,21,116,144], real implementation of
1. Cancer. Available online: https://www.who.int/news-room/fact-sheets/detail/cancer (accessed on 29 November 2021).
2. Luo, Y.H.; Chiu, C.H.; Scott Kuo, C.H.; Chou, T.Y.; Yeh, Y.C.; Hsu, H.S.; Yen, S.H.; Wu, Y.H.; Yang, J.C.; Liao, B.C.; et al. Lung
Cancer in Republic of China. J. Thorac. Oncol. 2021, 16, 519–527. [CrossRef] [PubMed]
3. Cause of Death Statistics. Available online: https://www.mohw.gov.tw/lp-4650-2.html (accessed on 1 October 2021).
4. Panunzio, A.; Sartori, P. Lung Cancer and Radiological Imaging. Curr. Radiopharm. 2020, 13, 238–242. [CrossRef] [PubMed]
5. Migliore, M.; Palmucci, S.; Nardini, M.; Basile, A. Imaging patterns of early stage lung cancer for the thoracic surgeon. J. Thorac.
Dis. 2020, 12, 3349–3356. [CrossRef] [PubMed]
6. Travis, W.D.; Brambilla, E.; Nicholson, A.G.; Yatabe, Y.; Austin, J.H.M.; Beasley, M.B.; Chirieac, L.R.; Dacic, S.; Duhig, E.; Flieder,
D.B.; et al. The 2015 World Health Organization Classification of Lung Tumors: Impact of Genetic, Clinical and Radiologic
Advances Since the 2004 Classification. J. Thorac. Oncol. 2015, 10, 1243–1260. [CrossRef]
7. Nicholson, A.G.; Tsao, M.S.; Beasley, M.B.; Borczuk, A.C.; Brambilla, E.; Cooper, W.A.; Dacic, S.; Jain, D.; Kerr, K.M.; Lantuejoul, S.;
et al. The 2021 WHO Classification of Lung Tumors: Impact of advances since 2015. J. Thorac. Oncol. 2021, 17, 362–387. [CrossRef]
8. Klang, E. Deep learning and medical imaging. J. Thorac. Dis. 2018, 10, 1325–1328. [CrossRef]
9. Lawson, C.E.; Marti, J.M.; Radivojevic, T.; Jonnalagadda, S.V.R.; Gentz, R.; Hillson, N.J.; Peisert, S.; Kim, J.; Simmons, B.A.;
Petzold, C.J.; et al. Machine learning for metabolic engineering: A review. Metab. Eng. 2021, 63, 34–60. [CrossRef]
10. Leiserson, C.E.; Thompson, N.C.; Emer, J.S.; Kuszmaul, B.C.; Lampson, B.W.; Sanchez, D.; Schardl, T.B. There’s plenty of room at
the Top: What will drive computer performance after Moore’s law? Science 2020, 368, eaam9744. [CrossRef]
11. Shalf, J. The future of computing beyond Moore’s Law. Philos. Trans. A Math. Phys. Eng. Sci. 2020, 378, 20190061. [CrossRef]
Cancers 2022, 14, 1370 12 of 17

12. Somvanshi, M.; Chavan, P.; Tambade, S.; Shinde, S. A review of machine learning techniques using decision tree and support
vector machine. In Proceedings of the 2016 International Conference on Computing Communication Control and Automation
(ICCUBEA), Pune, India, 12–13 August 2016; pp. 1–7.
13. Sesen, M.B.; Nicholson, A.E.; Banares-Alcantara, R.; Kadir, T.; Brady, M. Bayesian networks for clinical decision support in lung
cancer care. PLoS ONE 2013, 8, e82349. [CrossRef]
14. Gao, R.; Huo, Y.; Bao, S.; Tang, Y.; Antic, S.L.; Epstein, E.S.; Balar, A.B.; Deppen, S.; Paulson, A.B.; Sandler, K.L. Distanced LSTM:
Time-distanced gates in long short-term memory models for lung cancer detection. In International Workshop on Machine Learning
in Medical Imaging; Springer: New York, NY, USA, 2019; pp. 310–318.
15. Onishi, Y.; Teramoto, A.; Tsujimoto, M.; Tsukamoto, T.; Saito, K.; Toyama, H.; Imaizumi, K.; Fujita, H. Automated pulmonary
nodule classification in computed tomography images using a deep convolutional neural network trained by generative
adversarial networks. BioMed Res. Int. 2019, 2019, 6051939. [CrossRef] [PubMed]
16. Kourou, K.; Exarchos, T.P.; Exarchos, K.P.; Karamouzis, M.V.; Fotiadis, D.I. Machine learning applications in cancer prognosis and
prediction. Comput. Struct. Biotechnol. J. 2015, 13, 8–17. [CrossRef] [PubMed]
17. Sim, Y.; Chung, M.J.; Kotter, E.; Yune, S.; Kim, M.; Do, S.; Han, K.; Kim, H.; Yang, S.; Lee, D.J.; et al. Deep Convolutional Neural
Network-based Software Improves Radiologist Detection of Malignant Lung Nodules on Chest Radiographs. Radiology 2020, 294,
199–209. [CrossRef] [PubMed]
18. Rajpurkar, P.; Irvin, J.; Zhu, K.; Yang, B.; Mehta, H.; Duan, T.; Ding, D.; Bagul, A.; Langlotz, C.; Shpanskaya, K. Chexnet:
Radiologist-level pneumonia detection on chest x-rays with deep learning. arXiv 2017, arXiv:1711.05225.
19. Kim, Y.-G.; Cho, Y.; Wu, C.-J.; Park, S.; Jung, K.-H.; Seo, J.B.; Lee, H.J.; Hwang, H.J.; Lee, S.M.; Kim, N. Short-term reproducibility of
pulmonary nodule and mass detection in chest radiographs: Comparison among radiologists and four different computer-aided
detections with convolutional neural net. Sci. Rep. 2019, 9, 18738. [CrossRef] [PubMed]
20. Tam, M.; Dyer, T.; Dissez, G.; Morgan, T.N.; Hughes, M.; Illes, J.; Rasalingham, R.; Rasalingham, S. Augmenting lung cancer
diagnosis on chest radiographs: Positioning artificial intelligence to improve radiologist performance. Clin. Radiol. 2021, 76,
607–614. [CrossRef] [PubMed]
21. Kim, J.H.; Han, S.G.; Cho, A.; Shin, H.J.; Baek, S.-E. Effect of deep learning-based assistive technology use on chest radiograph
interpretation by emergency department physicians: A prospective interventional simulation-based study. BMC Med. Inform.
Decis. Mak. 2021, 21, 311. [CrossRef]
22. Van Ginneken, B.; Armato, S.G., III; de Hoop, B.; van Amelsvoort-van de Vorst, S.; Duindam, T.; Niemeijer, M.; Murphy, K.;
Schilham, A.; Retico, A.; Fantacci, M.E. Comparing and combining algorithms for computer-aided detection of pulmonary
nodules in computed tomography scans: The ANODE09 study. Med. Image Anal. 2010, 14, 707–722. [CrossRef]
23. Setio, A.A.A.; Traverso, A.; De Bel, T.; Berens, M.S.; Van Den Bogaard, C.; Cerello, P.; Chen, H.; Dou, Q.; Fantacci, M.E.; Geurts, B.
Validation, comparison, and combination of algorithms for automatic detection of pulmonary nodules in computed tomography
images: The LUNA16 challenge. Med. Image Anal. 2017, 42, 1–13. [CrossRef]
24. Roos, J.E.; Paik, D.; Olsen, D.; Liu, E.G.; Chow, L.C.; Leung, A.N.; Mindelzun, R.; Choudhury, K.R.; Naidich, D.P.; Napel, S.
Computer-aided detection (CAD) of lung nodules in CT scans: Radiologist performance and reading time with incremental CAD
assistance. Eur. Radiol. 2010, 20, 549–557. [CrossRef]
25. Lo, S.B.; Freedman, M.T.; Gillis, L.B.; White, C.S.; Mun, S.K. JOURNAL CLUB: Computer-aided detection of lung nodules on CT
with a computerized pulmonary vessel suppressed function. Am. J. Roentgenol. 2018, 210, 480–488. [CrossRef] [PubMed]
26. Liang, M.; Tang, W.; Xu, D.M.; Jirapatnakul, A.C.; Reeves, A.P.; Henschke, C.I.; Yankelevitz, D. Low-dose CT screening for lung
cancer: Computer-aided detection of missed lung cancers. Radiology 2016, 281, 279–288. [CrossRef] [PubMed]
27. Ciompi, F.; Chung, K.; Van Riel, S.J.; Setio, A.A.A.; Gerke, P.K.; Jacobs, C.; Scholten, E.T.; Schaefer-Prokop, C.; Wille, M.M.;
Marchiano, A. Towards automatic pulmonary nodule management in lung cancer screening with deep learning. Sci. Rep. 2017, 7,
46479. [CrossRef] [PubMed]
28. Sun, Y.; Li, C.; Jin, L.; Gao, P.; Zhao, W.; Ma, W.; Tan, M.; Wu, W.; Duan, S.; Shan, Y. Radiomics for lung adenocarcinoma
manifesting as pure ground-glass nodules: Invasive prediction. Eur. Radiol. 2020, 30, 3650. [CrossRef] [PubMed]
29. Hawkins, S.; Wang, H.; Liu, Y.; Garcia, A.; Stringfield, O.; Krewer, H.; Li, Q.; Cherezov, D.; Gatenby, R.A.; Balagurunathan, Y.
Predicting malignant nodules from screening CT scans. J. Thorac. Oncol. 2016, 11, 2120–2128. [CrossRef]
30. Tu, S.-J.; Wang, C.-W.; Pan, K.-T.; Wu, Y.-C.; Wu, C.-T. Localized thin-section CT with radiomics feature extraction and machine
learning to classify early-detected pulmonary nodules from lung cancer screening. Phys. Med. Biol. 2018, 63, 065005. [CrossRef]
31. Balagurunathan, Y.; Beers, A.; Mcnitt-Gray, M.; Hadjiiski, L.; Napel, S.; Goldgof, D.; Perez, G.; Arbelaez, P.; Mehrtash, A.; Kapur,
T. Lung Nodule Malignancy Prediction in Sequential CT Scans: Summary of ISBI 2018 Challenge. IEEE Trans. Med. Imaging 2021,
40, 3748–3761. [CrossRef]
32. Xiao, N.; Qiang, Y.; Bilal Zia, M.; Wang, S.; Lian, J. Ensemble classification for predicting the malignancy level of pulmonary
nodules on chest computed tomography images. Oncol. Lett. 2020, 20, 401–408. [CrossRef]
33. Lv, E.; Liu, W.; Wen, P.; Kang, X. Classification of Benign and Malignant Lung Nodules Based on Deep Convolutional Network
Feature Extraction. J. Healthc. Eng. 2021, 2021, 8769652. [CrossRef]
34. Poore, G.D.; Kopylova, E.; Zhu, Q.; Carpenter, C.; Fraraccio, S.; Wandro, S.; Kosciolek, T.; Janssen, S.; Metcalf, J.; Song, S.J.
Microbiome analyses of blood and tissues suggest cancer diagnostic approach. Nature 2020, 579, 567–574. [CrossRef]
Cancers 2022, 14, 1370 13 of 17

35. Hilario, M.; Kalousis, A.; Müller, M.; Pellegrini, C. Machine learning approaches to lung cancer prediction from mass spectra.
Proteomics 2003, 3, 1716–1719. [CrossRef] [PubMed]
36. Zhang, C.; Leng, W.; Sun, C.; Lu, T.; Chen, Z.; Men, X.; Wang, Y.; Wang, G.; Zhen, B.; Qin, J. Urine proteome profiling predicts
lung cancer from control cases and other tumors. EBioMedicine 2018, 30, 120–128. [CrossRef] [PubMed]
37. Tirzïte, M.; Bukovskis, M.; Strazda, G.; Jurka, N.; Taivans, I. Detection of lung cancer with electronic nose and logistic regression
analysis. J. Breath Res. 2018, 13, 016006. [CrossRef] [PubMed]
38. Huang, C.-H.; Zeng, C.; Wang, Y.-C.; Peng, H.-Y.; Lin, C.-S.; Chang, C.-J.; Yang, H.-Y. A study of diagnostic accuracy using a
chemical sensor array and a machine learning technique to detect lung cancer. Sensors 2018, 18, 2845. [CrossRef] [PubMed]
39. Kort, S.; Brusse-Keizer, M.; Gerritsen, J.W.; Schouwink, H.; Citgez, E.; de Jongh, F.; van der Maten, J.; Samii, S.; van den Bogart, M.;
van der Palen, J. Improving lung cancer diagnosis by combining exhaled-breath data and clinical parameters. ERJ Open Res. 2020,
6, 00221–02019. [CrossRef]
40. Wu, G.; Jochems, A.; Ibrahim, A.; Yan, C.; Sanduleanu, S.; Woodruff, H.C.; Lambin, P. Structural and functional radiomics for
lung cancer. Eur. J. Nucl. Med. Mol. Imaging 2021, 48, 3961–3974. [CrossRef]
41. Liu, S.; Liu, S.; Zhang, C.; Yu, H.; Liu, X.; Hu, Y.; Xu, W.; Tang, X.; Fu, Q. Exploratory study of a CT Radiomics model for the
classification of small cell lung cancer and non-small-cell lung cancer. Front. Oncol. 2020, 10, 1268. [CrossRef]
42. Zhu, X.; Dong, D.; Chen, Z.; Fang, M.; Zhang, L.; Song, J.; Yu, D.; Zang, Y.; Liu, Z.; Shi, J. Radiomic signature as a diagnostic factor
for histologic subtype classification of non-small cell lung cancer. Eur. Radiol. 2018, 28, 2772–2778. [CrossRef]
43. Gu, Q.; Feng, Z.; Liang, Q.; Li, M.; Deng, J.; Ma, M.; Wang, W.; Liu, J.; Liu, P.; Rong, P. Machine learning-based radiomics strategy
for prediction of cell proliferation in non-small cell lung cancer. Eur. J. Radiol. 2019, 118, 32–37. [CrossRef]
44. Wang, S.; Shi, J.; Ye, Z.; Dong, D.; Yu, D.; Zhou, M.; Liu, Y.; Gevaert, O.; Wang, K.; Zhu, Y. Predicting EGFR mutation status in
lung adenocarcinoma on computed tomography image using deep learning. Eur. Respir. J. 2019, 53, 1800986. [CrossRef]
45. Song, L.; Zhu, Z.; Mao, L.; Li, X.; Han, W.; Du, H.; Wu, H.; Song, W.; Jin, Z. Clinical, conventional CT and radiomic feature-based
machine learning models for predicting ALK rearrangement status in lung adenocarcinoma patients. Front. Oncol. 2020, 10, 369.
[CrossRef]
46. Zhu, Y.; Liu, Y.-L.; Feng, Y.; Yang, X.-Y.; Zhang, J.; Chang, D.-D.; Wu, X.; Tian, X.; Tang, K.-J.; Xie, C.-M. A CT-derived deep neural
network predicts for programmed death ligand-1 expression status in advanced lung adenocarcinomas. Ann. Transl. Med. 2020, 8,
930. [CrossRef]
47. Šarić, M.; Russo, M.; Stella, M.; Sikora, M. CNN-based method for lung cancer detection in whole slide histopathology images. In
Proceedings of the 2019 4th International Conference on Smart and Sustainable Technologies (SpliTech), Split, Croatia, 18–21 June
2019; pp. 1–4.
48. Wei, J.W.; Tafe, L.J.; Linnik, Y.A.; Vaickus, L.J.; Tomita, N.; Hassanpour, S. Pathologist-level classification of histologic patterns on
resected lung adenocarcinoma slides with deep neural networks. Sci. Rep. 2019, 9, 3358. [CrossRef]
49. Gertych, A.; Swiderska-Chadaj, Z.; Ma, Z.; Ing, N.; Markiewicz, T.; Cierniak, S.; Salemi, H.; Guzman, S.; Walts, A.E.; Knudsen, B.S.
Convolutional neural networks can accurately distinguish four histologic growth patterns of lung adenocarcinoma in digital
slides. Sci. Rep. 2019, 9, 1483. [CrossRef]
50. Coudray, N.; Ocampo, P.S.; Sakellaropoulos, T.; Narula, N.; Snuderl, M.; Fenyö, D.; Moreira, A.L.; Razavian, N.; Tsirigos, A.
Classification and mutation prediction from non–small cell lung cancer histopathology images using deep learning. Nat. Med.
2018, 24, 1559–1567. [CrossRef]
51. Wang, X.; Chen, H.; Gan, C.; Lin, H.; Dou, Q.; Tsougenis, E.; Huang, Q.; Cai, M.; Heng, P.-A. Weakly supervised deep learning for
whole slide lung cancer image analysis. IEEE Trans. Cybern. 2019, 50, 3950–3962. [CrossRef]
52. Kapil, A.; Meier, A.; Zuraw, A.; Steele, K.E.; Rebelatto, M.C.; Schmidt, G.; Brieu, N. Deep semi supervised generative learning for
automated tumor proportion scoring on NSCLC tissue needle biopsies. Sci. Rep. 2018, 8, 17343. [CrossRef]
53. Aprupe, L.; Litjens, G.; Brinker, T.J.; van der Laak, J.; Grabe, N. Robust and accurate quantification of biomarkers of immune cells
in lung cancer micro-environment using deep convolutional neural networks. PeerJ 2019, 7, e6335. [CrossRef]
54. Jones, G.D.; Brandt, W.S.; Shen, R.; Sanchez-Vega, F.; Tan, K.S.; Martin, A.; Zhou, J.; Berger, M.; Solit, D.B.; Schultz, N. A
genomic-pathologic annotated risk model to predict recurrence in early-stage lung adenocarcinoma. JAMA Surg. 2021, 156,
e205601. [CrossRef]
55. Sha, L.; Osinski, B.L.; Ho, I.Y.; Tan, T.L.; Willis, C.; Weiss, H.; Beaubier, N.; Mahon, B.M.; Taxter, T.J.; Yip, S.S. Multi-field-of-view
deep learning model predicts nonsmall cell lung cancer programmed death-ligand 1 status from whole-slide hematoxylin and
eosin images. J. Pathol. Inform. 2019, 10, 24.
56. Jiang, M.; Sun, D.; Guo, Y.; Guo, Y.; Xiao, J.; Wang, L.; Yao, X. Assessing PD-L1 expression level by radiomic features from PET/CT
in nonsmall cell lung cancer patients: An initial result. Acad. Radiol. 2020, 27, 171–179. [CrossRef]
57. Li, S.; Ding, C.; Zhang, H.; Song, J.; Wu, L. Radiomics for the prediction of EGFR mutation subtypes in non-small cell lung cancer.
Med. Phys. 2019, 46, 4545–4552. [CrossRef]
58. Dercle, L.; Fronheiser, M.; Lu, L.; Du, S.; Hayes, W.; Leung, D.K.; Roy, A.; Wilkerson, J.; Guo, P.; Fojo, A.T. Identification of
non–small cell lung cancer sensitive to systemic cancer therapies using radiomics. Clin. Cancer Res. 2020, 26, 2151–2162. [CrossRef]
59. Le, V.-H.; Kha, Q.-H.; Hung, T.N.K.; Le, N.Q.K. Risk score generated from CT-based radiomics signatures for overall survival
prediction in non-small cell lung cancer. Cancers 2021, 13, 3616. [CrossRef]
Cancers 2022, 14, 1370 14 of 17

60. Sun, F.; Chen, Y.; Chen, X.; Sun, X.; Xing, L. CT-based radiomics for predicting brain metastases as the first failure in patients with
curatively resected locally advanced non-small cell lung cancer. Eur. J. Radiol. 2021, 134, 109411. [CrossRef]
61. Yoshiyasu, N.; Kojima, F.; Hayashi, K.; Bando, T. Radiomics technology for identifying early-stage lung adenocarcinomas suitable
for sublobar resection. J. Thorac. Cardiovasc. Surg. 2021, 162, 477–485.e1. [CrossRef]
62. Choi, H.; Kim, H.; Hong, W.; Park, J.; Hwang, E.J.; Park, C.M.; Kim, Y.T.; Goo, J.M. Prediction of visceral pleural invasion in lung
cancer on CT: Deep learning model achieves a radiologist-level performance with adaptive sensitivity and specificity to clinical
needs. Eur. Radiol. 2021, 31, 2866–2876. [CrossRef]
63. Mattonen, S.A.; Palma, D.A.; Haasbeek, C.J.; Senan, S.; Ward, A.D. Early prediction of tumor recurrence based on CT texture
changes after stereotactic ablative radiotherapy (SABR) for lung cancer. Med. Phys. 2014, 41, 033502. [CrossRef]
64. Lewis, J.E.; Kemp, M.L. Integration of machine learning and genome-scale metabolic modeling identifies multi-omics biomarkers
for radiation resistance. Nat. Commun. 2021, 12, 2700. [CrossRef]
65. Krafft, S.P.; Rao, A.; Stingo, F.; Briere, T.M.; Court, L.E.; Liao, Z.; Martel, M.K. The utility of quantitative CT radiomics features for
improved prediction of radiation pneumonitis. Med. Phys. 2018, 45, 5317–5324. [CrossRef]
66. Bourbonne, V.; Da-Ano, R.; Jaouen, V.; Lucia, F.; Dissaux, G.; Bert, J.; Pradier, O.; Visvikis, D.; Hatt, M.; Schick, U. Radiomics
analysis of 3D dose distributions to predict toxicity of radiotherapy for lung cancer. Radiother. Oncol. 2021, 155, 144–150.
[CrossRef]
67. Girard, L.; Zochbauer-Muller, S.; Virmani, A.K.; Gazdar, A.F.; Minna, J.D. Genome-wide allelotyping of lung cancer identifies new
regions of allelic loss, differences between small cell lung cancer and non-small cell lung cancer, and loci clustering. Cancer Res.
2000, 60, 4894–4906.
68. Shen, R.; Olshen, A.B.; Ladanyi, M. Integrative clustering of multiple genomic data types using a joint latent variable model with
application to breast and lung cancer subtype analysis. Bioinformatics 2009, 25, 2906–2912. [CrossRef]
69. Silver, D.; Hubert, T.; Schrittwieser, J.; Antonoglou, I.; Lai, M.; Guez, A.; Lanctot, M.; Sifre, L.; Kumaran, D.; Graepel, T. A general
reinforcement learning algorithm that masters chess, shogi, and Go through self-play. Science 2018, 362, 1140–1144. [CrossRef]
70. Shi, H.; Lu, J.; Zhou, Q. A novel data augmentation method using style-based GAN for robust pulmonary nodule segmentation.
In Proceedings of the 2020 Chinese Control and Decision Conference (CCDC), Hefei, China, 22–24 August 2020; pp. 2486–2491.
71. Ali, I.; Hart, G.R.; Gunabushanam, G.; Liang, Y.; Muhammad, W.; Nartowt, B.; Kane, M.; Ma, X.; Deng, J. Lung Nodule Detection
via Deep Reinforcement Learning. Front. Oncol. 2018, 8, 108. [CrossRef]
72. Capizzi, G.; Sciuto, G.L.; Napoli, C.; Połap, D.; Woźniak, M. Small lung nodules detection based on fuzzy-logic and probabilistic
neural network with bioinspired reinforcement learning. IEEE Trans. Fuzzy Syst. 2019, 28, 1178–1189. [CrossRef]
73. In, K.-H.; Kwon, Y.-S.; Oh, I.-J.; Kim, K.-S.; Jung, M.-H.; Lee, K.-H.; Kim, S.-Y.; Ryu, J.-S.; Lee, S.-Y.; Jeong, E.-T. Lung cancer
patients who are asymptomatic at diagnosis show favorable prognosis: A Korean Lung Cancer Registry Study. Lung. Cancer 2009,
64, 232–237. [CrossRef]
74. Quadrelli, S.; Lyons, G.; Colt, H.; Chimondeguy, D.; Buero, A. Clinical characteristics and prognosis of incidentally detected lung
cancers. Int. J. Surg. Oncol. 2015, 2015, 287604. [CrossRef]
75. Melamed, M.R.; Flehinger, B.J.; Zaman, M.B.; Heelan, R.T.; Perchick, W.A.; Martini, N. Screening for early lung cancer. Results of
the Memorial Sloan-Kettering study in New York. Chest 1984, 86, 44–53. [CrossRef]
76. Hocking, W.G.; Hu, P.; Oken, M.M.; Winslow, S.D.; Kvale, P.A.; Prorok, P.C.; Ragard, L.R.; Commins, J.; Lynch, D.A.; Andriole,
G.L.; et al. Lung cancer screening in the randomized Prostate, Lung, Colorectal, and Ovarian (PLCO) Cancer Screening Trial. J.
Natl. Cancer Inst. 2010, 102, 722–731. [CrossRef]
77. Chu, G.C.W.; Lazare, K.; Sullivan, F. Serum and blood based biomarkers for lung cancer screening: A systematic review. BMC
Cancer 2018, 18, 181. [CrossRef]
78. Montani, F.; Marzi, M.J.; Dezi, F.; Dama, E.; Carletti, R.M.; Bonizzi, G.; Bertolotti, R.; Bellomi, M.; Rampinelli, C.; Maisonneuve, P.;
et al. miR-Test: A blood test for lung cancer early detection. J. Natl. Cancer Inst. 2015, 107, djv063. [CrossRef]
79. Campanella, A.; De Summa, S.; Tommasi, S. Exhaled breath condensate biomarkers for lung cancer. J. Breath Res. 2019, 13, 044002.
[CrossRef]
80. Lopez-Sanchez, L.M.; Jurado-Gamez, B.; Feu-Collado, N.; Valverde, A.; Canas, A.; Fernandez-Rueda, J.L.; Aranda, E.; Rodriguez-
Ariza, A. Exhaled breath condensate biomarkers for the early diagnosis of lung cancer using proteomics. Am. J. Physiol. Lung Cell
Mol. Physiol. 2017, 313, L664–L676. [CrossRef]
81. National Lung Screening Trial Research, T.; Aberle, D.R.; Adams, A.M.; Berg, C.D.; Black, W.C.; Clapp, J.D.; Fagerstrom, R.M.;
Gareen, I.F.; Gatsonis, C.; Marcus, P.M.; et al. Reduced lung-cancer mortality with low-dose computed tomographic screening. N.
Engl. J. Med. 2011, 365, 395–409. [CrossRef]
82. de Koning, H.J.; van der Aalst, C.M.; de Jong, P.A.; Scholten, E.T.; Nackaerts, K.; Heuvelmans, M.A.; Lammers, J.J.; Weenink, C.;
Yousaf-Khan, U.; Horeweg, N.; et al. Reduced Lung-Cancer Mortality with Volume CT Screening in a Randomized Trial. N. Engl.
J. Med. 2020, 382, 503–513. [CrossRef]
83. Baker, S.R.; Patel, R.H.; Yang, L.; Lelkes, V.M.; Castro, A., 3rd. Malpractice suits in chest radiology: An evaluation of the histories
of 8265 radiologists. J. Thorac. Imaging 2013, 28, 388–391. [CrossRef]
84. Sakai, M.; Kato, A.; Kobayashi, N.; Nakamura, R.; Okawa, S.; Sato, Y. Improved Lung Cancer Detection in Cardiovascular
Outpatients by the Pulmonologist-based Interpretation of Chest Radiographs. Intern. Med. 2015, 54, 2991–2997. [CrossRef]
Cancers 2022, 14, 1370 15 of 17

85. White, C.S.; Salis, A.I.; Meyer, C.A. Missed lung cancer on chest radiography and computed tomography: Imaging and
medicolegal issues. J. Thorac. Imaging 1999, 14, 63–68. [CrossRef]
86. About DICOM: Overview. Available online: https://www.dicomstandard.org/about (accessed on 3 December 2021).
87. Shiraishi, J.; Katsuragawa, S.; Ikezoe, J.; Matsumoto, T.; Kobayashi, T.; Komatsu, K.-I.; Matsui, M.; Fujita, H.; Kodera, Y.; Doi, K.
Development of a digital image database for chest radiographs with and without a lung nodule: Receiver operating characteristic
analysis of radiologists’ detection of pulmonary nodules. Am. J. Roentgenol. 2000, 174, 71–74. [CrossRef]
88. Jaeger, S.; Candemir, S.; Antani, S.; Wang, Y.X.; Lu, P.X.; Thoma, G. Two public chest X-ray datasets for computer-aided screening
of pulmonary diseases. Quant. Imaging Med. Surg. 2014, 4, 475–477. [CrossRef] [PubMed]
89. Wang, X.; Peng, Y.; Lu, L.; Lu, Z.; Bagheri, M.; Summers, R.M. Chestx-ray8: Hospital-scale chest x-ray database and benchmarks
on weakly-supervised classification and localization of common thorax diseases. In Proceedings of the IEEE Conference on
Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017; pp. 2097–2106.
90. Bustos, A.; Pertusa, A.; Salinas, J.-M.; de la Iglesia-Vayá, M. Padchest: A large chest x-ray image dataset with multi-label annotated
reports. Med. Image Anal. 2020, 66, 101797. [CrossRef] [PubMed]
91. Armato, S.G., III; McLennan, G.; Bidaut, L.; McNitt-Gray, M.F.; Meyer, C.R.; Reeves, A.P.; Zhao, B.; Aberle, D.R.; Henschke, C.I.;
Hoffman, E.A. The lung image database consortium (LIDC) and image database resource initiative (IDRI): A completed reference
database of lung nodules on CT scans. Med. Phys. 2011, 38, 915–931. [CrossRef] [PubMed]
92. Johnson, A.E.; Pollard, T.J.; Greenbaum, N.R.; Lungren, M.P.; Deng, C.-Y.; Peng, Y.; Lu, Z.; Mark, R.G.; Berkowitz, S.J.; Horng, S.
MIMIC-CXR-JPG, a large publicly available database of labeled chest radiographs. arXiv 2019, arXiv:1901.07042.
93. Irvin, J.; Rajpurkar, P.; Ko, M.; Yu, Y.; Ciurea-Ilcus, S.; Chute, C.; Marklund, H.; Haghgoo, B.; Ball, R.; Shpanskaya, K. Chexpert:
A large chest radiograph dataset with uncertainty labels and expert comparison. In Proceedings of the AAAI Conference on
Artificial Intelligence, Honolulu, HI, USA, 27 January–1 February 2019; pp. 590–597.
94. Nguyen, H.C.; Le, T.T.; Pham, H.; Nguyen, H.Q. VinDr-RibCXR: A Benchmark Dataset for Automatic Segmentation and Labeling
of Individual Ribs on Chest X-rays. arXiv 2021, arXiv:2107.01327.
95. Jain, S.; Agrawal, A.; Saporta, A.; Truong, S.Q.; Bui, T.; Chambon, P.; Zhang, Y.; Lungren, M.P.; Ng, A.Y.; Langlotz, C. RadGraph:
Extracting Clinical Entities and Relations from Radiology Reports. arXiv 2021, arXiv:2106.14463.
96. Lanfredi, R.B.; Zhang, M.; Auffermann, W.F.; Chan, J.; Duong, P.-A.T.; Srikumar, V.; Drew, T.; Schroeder, J.D.; Tasdizen, T.
REFLACX, a dataset of reports and eye-tracking data for localization of abnormalities in chest X-rays. arXiv 2021, arXiv:2109.14187.
97. Lodwick, G.S.; Keats, T.E.; Dorst, J.P. The Coding of Roentgen Images for Computer Analysis as Applied to Lung Cancer. Radiology
1963, 81, 185–200. [CrossRef]
98. Munir, K.; Elahi, H.; Ayub, A.; Frezza, F.; Rizzi, A. Cancer diagnosis using deep learning: A bibliographic review. Cancers 2019, 11,
1235. [CrossRef]
99. Van Riel, S.J.; Jacobs, C.; Scholten, E.T.; Wittenberg, R.; Wille, M.M.W.; de Hoop, B.; Sprengers, R.; Mets, O.M.; Geurts, B.; Prokop,
M. Observer variability for Lung-RADS categorisation of lung cancer screening CTs: Impact on patient management. Eur. Radiol.
2019, 29, 924–931. [CrossRef]
100. Schreuder, A.; Scholten, E.T.; van Ginneken, B.; Jacobs, C. Artificial intelligence for detection and characterization of pulmonary
nodules in lung cancer CT screening: Ready for practice? Transl. Lung Cancer Res. 2021, 10, 2378. [CrossRef] [PubMed]
101. Liu, B.; Chi, W.; Li, X.; Li, P.; Liang, W.; Liu, H.; Wang, W.; He, J. Evolving the pulmonary nodules diagnosis from classical
approaches to deep learning-aided decision support: Three decades’ development course and future prospect. J. Cancer Res. Clin.
Oncol. 2020, 146, 153–185. [CrossRef] [PubMed]
102. Li, D.; Mikela Vilmun, B.; Frederik Carlsen, J.; Albrecht-Beste, E.; Ammitzbøl Lauridsen, C.; Bachmann Nielsen, M.; Lindskov
Hansen, K. The performance of deep learning algorithms on automatic pulmonary nodule detection and classification tested on
different datasets that are not derived from LIDC-IDRI: A systematic review. Diagnostics 2019, 9, 207. [CrossRef]
103. Pastorino, U.; Rossi, M.; Rosato, V.; Marchianò, A.; Sverzellati, N.; Morosi, C.; Fabbri, A.; Galeone, C.; Negri, E.; Sozzi, G. Annual
or biennial CT screening versus observation in heavy smokers: 5-year results of the MILD trial. Eur. J. Cancer Prev. 2012, 21,
308–315. [CrossRef] [PubMed]
104. Pedersen, J.H.; Ashraf, H.; Dirksen, A.; Bach, K.; Hansen, H.; Toennesen, P.; Thorsen, H.; Brodersen, J.; Skov, B.G.; Døssing, M.
The Danish randomized lung cancer CT screening trial—Overall design and results of the prevalence round. J. Thorac. Oncol.
2009, 4, 608–614. [CrossRef] [PubMed]
105. Martin, M.D.; Kanne, J.P.; Broderick, L.S.; Kazerooni, E.A.; Meyer, C.A. Lung-RADS: Pushing the limits. Radiographics 2017, 37,
1975–1993. [CrossRef] [PubMed]
106. MacMahon, H.; Naidich, D.P.; Goo, J.M.; Lee, K.S.; Leung, A.N.; Mayo, J.R.; Mehta, A.C.; Ohno, Y.; Powell, C.A.; Prokop, M.
Guidelines for management of incidental pulmonary nodules detected on CT images: From the Fleischner Society 2017. Radiology
2017, 284, 228–243. [CrossRef]
107. Chabon, J.J.; Hamilton, E.G.; Kurtz, D.M.; Esfahani, M.S.; Moding, E.J.; Stehr, H.; Schroers-Martin, J.; Nabet, B.Y.; Chen, B.;
Chaudhuri, A.A. Integrating genomic features for non-invasive early lung cancer detection. Nature 2020, 580, 245–251. [CrossRef]
108. Cammarota, G.; Ianiro, G.; Ahern, A.; Carbone, C.; Temko, A.; Claesson, M.J.; Gasbarrini, A.; Tortora, G. Gut microbiome, big data
and machine learning to promote precision medicine for cancer. Nat. Rev. Gastroenterol. Hepatol. 2020, 17, 635–648. [CrossRef]
109. Peled, N.; Fuchs, V.; Kestenbaum, E.H.; Oscar, E.; Bitran, R. An Update on the Use of Exhaled Breath Analysis for the Early
Detection of Lung Cancer. Lung Cancer Targets Ther. 2021, 12, 81. [CrossRef]
Cancers 2022, 14, 1370 16 of 17

110. Xiang, D.; Zhang, B.; Doll, D.; Shen, K.; Kloecker, G.; Freter, C. Lung cancer screening: From imaging to biomarker. Biomark. Res.
2013, 1, 4. [CrossRef] [PubMed]
111. Daniels, M.; Goh, F.; Wright, C.M.; Sriram, K.B.; Relan, V.; Clarke, B.E.; Duhig, E.E.; Bowman, R.V.; Yang, I.A.; Fong, K.M. Whole
genome sequencing for lung cancer. J. Thorac. Dis. 2012, 4, 155. [PubMed]
112. Choi, Y.; Qu, J.; Wu, S.; Hao, Y.; Zhang, J.; Ning, J.; Yang, X.; Lofaro, L.; Pankratz, D.G.; Babiarz, J. Improving lung cancer risk
stratification leveraging whole transcriptome RNA sequencing and machine learning across multiple cohorts. BMC Med. Genom.
2020, 13, 151. [CrossRef] [PubMed]
113. Chang, K.; Creighton, C.; Davis, C.; Donehower, L. The cancer genome atlas pan-cancer analysis project. Nat. Genet. 2013, 45,
1113–1120.
114. Phillips, M.; Gleeson, K.; Hughes, J.M.B.; Greenberg, J.; Cataneo, R.N.; Baker, L.; McVay, W.P. Volatile organic compounds in
breath as markers of lung cancer: A cross-sectional study. Lancet 1999, 353, 1930–1933. [CrossRef]
115. Evans, A.J.; Bauer, T.W.; Bui, M.M.; Cornish, T.C.; Duncan, H.; Glassy, E.F.; Hipp, J.; McGee, R.S.; Murphy, D.; Myers, C. US Food
and Drug Administration approval of whole slide imaging for primary diagnosis: A key milestone is reached and new questions
are raised. Arch. Pathol. Lab. Med. 2018, 142, 1383–1387. [CrossRef]
116. Abels, E.; Pantanowitz, L. Current state of the regulatory trajectory for whole slide imaging devices in the USA. J. Pathol. Inform.
2017, 8, 23. [CrossRef]
117. Niazi, M.K.K.; Parwani, A.V.; Gurcan, M.N. Digital pathology and artificial intelligence. Lancet Oncol. 2019, 20, e253–e261.
[CrossRef]
118. DICOM Whole Slide Imaging (WSI). Available online: https://dicom.nema.org/Dicom/DICOMWSI/ (accessed on 29 November 2021).
119. Sakamoto, T.; Furukawa, T.; Lami, K.; Pham, H.H.N.; Uegami, W.; Kuroda, K.; Kawai, M.; Sakanashi, H.; Cooper, L.A.D.; Bychkov,
A. A narrative review of digital pathology and artificial intelligence: Focusing on lung cancer. Transl. Lung Cancer Res. 2020, 9,
2255. [CrossRef]
120. Giovagnoli, M.R.; Giansanti, D. Artificial Intelligence in Digital Pathology: What Is the Future? Part 1: From the Digital Slide
Onwards. Healthc. Multidiscip. Digit. Publ. Inst. 2021, 9, 858. [CrossRef]
121. Bejnordi, B.E.; Veta, M.; Van Diest, P.J.; Van Ginneken, B.; Karssemeijer, N.; Litjens, G.; Van Der Laak, J.A.; Hermsen, M.; Manson,
Q.F.; Balkenhol, M. Diagnostic assessment of deep learning algorithms for detection of lymph node metastases in women with
breast cancer. Jama 2017, 318, 2199–2210. [CrossRef] [PubMed]
122. Biermann, J.; Adkins, D.; Agulnik, M.; Benjamin, R.; Brigman, B.; Butrynski, J.; Cheong, D.; Chow, W.; Curry, W.; Frassica, D.
National comprehensive cancer network. Bone cancer. J. Natl. Compr. Cancer Netw. 2013, 11, 688–723. [CrossRef] [PubMed]
123. Furukawa, T.; Kuroda, K.; Bychkov, A.; Pham, H.; Kashima, Y.; Fukuoka, J. Verification of Deep Learning Model to Measure Tumor
Cellularity in Transbronchial Biopsies of Lung Adenocarcinoma; Laboratory Investigation, Nature Publishing Group: New York, NY,
USA, 2019.
124. Sakamoto, T.; Furukawa, T.; Pham, H.H.; Kuroda, K.; Tabata, K.; Kashima, Y.; Okoshi, E.N.; Morimoto, S.; Bychkov, A.; Fukuoka, J.
Collaborative workflow between pathologists and deep learning for evaluation of tumor cellularity in lung adenocarcinoma.
bioRxiv 2022. [CrossRef]
125. Hondelink, L.M.; Hüyük, M.; Postmus, P.E.; Smit, V.T.; Blom, S.; von der Thüsen, J.H.; Cohen, D. Development and validation of a
supervised deep learning algorithm for automated whole-slide programmed death-ligand 1 tumour proportion score assessment
in non-small cell lung cancer. Histopathology 2021, 80, 635–647. [CrossRef] [PubMed]
126. Wu, J.; Lin, D. A Review of Artificial Intelligence in Precise Assessment of Programmed Cell Death-ligand 1 and Tumor-infiltrating
Lymphocytes in Non− Small Cell Lung Cancer. Adv. Anat. Pathol. 2021, 28, 439–445. [CrossRef]
127. Campanella, G.; Hanna, M.G.; Geneslaw, L.; Miraflor, A.; Silva, V.W.K.; Busam, K.J.; Brogi, E.; Reuter, V.E.; Klimstra, D.S.; Fuchs,
T.J. Clinical-grade computational pathology using weakly supervised deep learning on whole slide images. Nat. Med. 2019, 25,
1301–1309. [CrossRef]
128. Giansanti, D.; Grigioni, M.; D’Avenio, G.; Morelli, S.; Maccioni, G.; Bondi, A.; Giovagnoli, M.R. Virtual microscopy and digital
cytology: State of the art. Ann. Dell’istituto Super. Di Sanità 2010, 46, 115–122.
129. Boschetto, A.; Pochini, M.; Bottini, L.; Giovagnoli, M.R.; Giansanti, D. The focus emulation and image enhancement in digital
cytology: An experience using the software Mathematica. Comput. Methods Biomech. Biomed. Eng. Imaging Vis. 2015, 3, 110–116.
[CrossRef]
130. Lin, Y.-J.; Chao, T.-K.; Khalil, M.-A.; Lee, Y.-C.; Hong, D.-Z.; Wu, J.-J.; Wang, C.-W. Deep Learning Fast Screening Approach on
Cytological Whole Slides for Thyroid Cancer Diagnosis. Cancers 2021, 13, 3891. [CrossRef]
131. Echle, A.; Rindtorff, N.T.; Brinker, T.J.; Luedde, T.; Pearson, A.T.; Kather, J.N. Deep learning in cancer pathology: A new generation
of clinical biomarkers. Br. J. Cancer 2021, 124, 686–696. [CrossRef]
132. Predicting Response to Immunotherapy Using Computer Extracted Featuresof Cancer Nuclei from Hematoxylin and Eosin
(H&E) Stained Images of Non-Small Cell Lung Cancer (NSCLC). Available online: https://patft.uspto.gov/netacgi/nph-Parser?
Sect1=PTO2&Sect2=HITOFF&p=1&u=%2Fnetahtml%2FPTO%2Fsearchbool.html&r=1&f=G&l=50&co1=AND&d=PTXT&s1=1
1,055,844.PN.&OS=PN/11,055,844&RS=PN/11,055,844 (accessed on 6 March 2022).
133. Wulczyn, E.; Steiner, D.F.; Xu, Z.; Sadhwani, A.; Wang, H.; Flament-Auvigne, I.; Mermel, C.H.; Chen, P.-H.C.; Liu, Y.; Stumpe, M.C.
Deep learning-based survival prediction for multiple cancer types using histopathology images. PLoS ONE 2020, 15, e0233678.
[CrossRef] [PubMed]
Cancers 2022, 14, 1370 17 of 17

134. D’Antonoli, T.A.; Farchione, A.; Lenkowicz, J.; Chiappetta, M.; Cicchetti, G.; Martino, A.; Ottavianelli, A.; Manfredi, R.;
Margaritora, S.; Bonomo, L. CT radiomics signature of tumor and peritumoral lung parenchyma to predict nonsmall cell lung
cancer postsurgical recurrence risk. Acad. Radiol. 2020, 27, 497–507.
135. Hosny, A.; Parmar, C.; Coroller, T.P.; Grossmann, P.; Zeleznik, R.; Kumar, A.; Bussink, J.; Gillies, R.J.; Mak, R.H.; Aerts, H.J. Deep
learning for lung cancer prognostication: A retrospective multi-cohort radiomics study. PLoS Med. 2018, 15, e1002711. [CrossRef]
[PubMed]
136. Jochems, A.; Deist, T.M.; Van Soest, J.; Eble, M.; Bulens, P.; Coucke, P.; Dries, W.; Lambin, P.; Dekker, A. Distributed learning:
Developing a predictive model based on data from multiple hospitals without data leaving the hospital–a real life proof of
concept. Radiother. Oncol. 2016, 121, 459–467. [CrossRef] [PubMed]
137. Jochems, A.; Deist, T.M.; El Naqa, I.; Kessler, M.; Mayo, C.; Reeves, J.; Jolly, S.; Matuszak, M.; Ten Haken, R.; van Soest, J.
Developing and validating a survival prediction model for NSCLC patients through distributed learning across 3 countries. Int. J.
Radiat. Oncol. Biol. Phys. 2017, 99, 344–352. [CrossRef]
138. Wang, D.D.; Zhou, W.; Yan, H.; Wong, M.; Lee, V. Personalized prediction of EGFR mutation-induced drug resistance in lung
cancer. Sci. Rep. 2013, 3, 2855. [CrossRef]
139. Giang, T.-T.; Nguyen, T.-P.; Tran, D.-H. Stratifying patients using fast multiple kernel learning framework: Case studies of
Alzheimer’s disease and cancers. BMC Med. Inform. Decis. Mak. 2020, 20, 108. [CrossRef]
140. Gao, Y.; Zhou, R.; Lyu, Q. Multiomics and machine learning in lung cancer prognosis. J. Thorac. Dis. 2020, 12, 4531. [CrossRef]
141. Wissel, D.; Rowson, D.; Boeva, V. Hierarchical autoencoder-based integration improves performance in multi-omics cancer
survival models through soft modality selection. bioRxiv 2022. [CrossRef]
142. Coory, M.; Gkolia, P.; Yang, I.A.; Bowman, R.V.; Fong, K.M. Systematic review of multidisciplinary teams in the management of
lung cancer. Lung Cancer 2008, 60, 14–21. [CrossRef]
143. Denton, E.; Conron, M. Improving outcomes in lung cancer: The value of the multidisciplinary health care team. J. Multidiscip.
Healthc. 2016, 9, 137–144. [CrossRef] [PubMed]
144. Wichmann, J.L.; Willemink, M.J.; De Cecco, C.N. Artificial Intelligence and Machine Learning in Radiology: Current State and
Considerations for Routine Clinical Implementation. Investig. Radiol. 2020, 55, 619–627. [CrossRef] [PubMed]