ESJ Natural/Life/Medical Sciences Peer-reviewed
Neural Network Approaches for Early Breast Cancer
Detection
Ala’a R. Al-Shamasneh
Najla Althuniyan
Nadia Alabdulkarim
Maya Ahmad
Department of Computer Science, Prince Sultan University,
Kingdom of Saudi Arabia, Riyadh
Doi:10.19044/esj.2024.v20n33p42
Submitted: 22 February 2024 Copyright 2024 Author(s)
Accepted: 05 November 2024 Under Creative Commons CC-BY 4.0
Published: 30 November 2024 OPEN ACCESS
Cite As:
Al-Shamasneh A.R.,Althuniyan N., Alabdulkarim N. & Ahmad M. (2024). Neural Network
Approaches for Early Breast Cancer Detection. European Scientific Journal, ESJ, 20 (33), 42.
https://doi.org/10.19044/esj.2024.v20n33p42
Abstract
Breast cancer research remains essential due to its high mortality rates
and the critical need for improved diagnostic accuracy. This study investigates
the application of neural network techniques to enhance early breast cancer
detection, utilizing Artificial Neural Networks (ANN) implemented in
MATLAB. By analyzing datasets from the UCI Machine Learning
Repository, specifically the Coimbra and Wisconsin Breast Cancer datasets,
this research demonstrates the accuracy and efficiency of neural networks in
classifying malignant and benign cases. Methodologically, the study involves
data preprocessing, ANN pattern recognition modeling, and testing across
multiple metrics including confusion matrices and ROC plots to validate the
model's predictive performance. The findings underscore the potential of
neural networks as a diagnostic support tool, facilitating faster and more
accurate cancer detection, thereby contributing to improved patient outcomes
and supporting global health initiatives for early diagnosis.
Keywords: Breast Cancer, Neural Networks, Early Detection
www.eujournal.org 42
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Introduction
Detecting cancer at late stages remains a pervasive challenge in the
medical field worldwide, often leading to critical health conditions and limited
treatment options for patients. The urgency for timely diagnosis to prevent
life-threatening situations underscores the importance of leveraging
technology to save lives. This paper explores the potential of neural
networking, an artificial intelligence technique, in revolutionizing breast
cancer detection and improving patient outcomes.
Neural networking, a key component of artificial intelligence, mimics
the human brain's ability to process data through interconnected neurons or
nodes arranged in layered structures. By utilizing deep learning principles,
neural networks can continuously learn and enhance their ability to interpret
data accurately over time. This approach holds promise for transforming
cancer detection by providing faster, more accurate, and reliable results.
While mammogram screening has been a conventional method for
breast cancer detection, recent advancements in technology have opened new
avenues for leveraging artificial intelligence and neural networking. The
integration of these modern techniques aims to augment existing diagnostic
capabilities, enabling early detection and intervention for women worldwide.
By harnessing the power of artificial intelligence, medical professionals can
more effectively identify cancerous lesions and prioritize high-risk patients.
Although artificial intelligence models, such as those used in prostate
cancer detection, offer promising results in automating cancer detection
processes, they are not intended to replace human expertise. Instead, these
technologies serve as valuable tools to assist oncologists and radiologists in
identifying cancerous abnormalities and distinguishing them from benign
masses. The inherent limitations of current detection methods, including false
positives and missed diagnoses, highlight the need for more advanced and
accurate approaches.
The application of neural networking in breast cancer detection holds
immense potential for improving imaging speed, accuracy, and prognostic
capabilities. This paper aims to present a comprehensive overview of the
neural networking method, utilizing MATLAB as a powerful computational
tool. MATLAB's specialized functionalities and toolboxes facilitate the
analysis of large datasets, making it an ideal platform for integrating medical
expertise with modern artificial neural network technology.
By combining insights from the medical field with cutting-edge
artificial intelligence techniques, this research endeavors to contribute to the
early detection of breast cancer. Ultimately, the goal is to empower healthcare
professionals with innovative tools and methodologies to aid in the timely
diagnosis and treatment of this life-threatening disease.
www.eujournal.org 43
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Literature Review
Breast cancer remains a significant global health concern, prompting
ongoing research into more effective diagnostic methods. Recent studies have
emphasized the importance of early detection in improving patient outcomes
and reducing mortality rates (Duffy et al., 2020). Mammogram screening has
traditionally been the primary method for breast cancer detection, but its
limitations, including false positives and missed diagnoses, have led to the
exploration of alternative approaches (Yala et al., 2019).
Artificial intelligence (AI) and machine learning (ML) techniques have
emerged as promising tools for enhancing cancer detection and diagnosis.
Neural networking, a subset of AI, utilizes deep learning algorithms to analyze
complex data patterns, mimicking the human brain's processing capabilities
(Esteva et al., 2019). Neural networks have shown remarkable success in
various medical imaging tasks, including the detection of breast cancer from
mammograms (Rodriguez-Ruiz et al., 2019).
Recent advancements in AI-based breast cancer detection have
demonstrated significant improvements in accuracy and efficiency. For
example, deep learning models trained on large-scale datasets have achieved
impressive results in identifying malignant lesions and reducing false positives
(McKinney et al., 2020). These models leverage convolutional neural
networks (CNNs) to extract relevant features from mammogram images and
classify them with high precision (Wang et al., 2021).
Moreover, the integration of AI technologies into clinical practice has
shown promise in streamlining diagnostic workflows and improving
radiologists' productivity. AI-assisted interpretation tools, such as computer-
aided detection (CAD) systems, provide radiologists with real-time feedback
and decision support, enhancing their diagnostic accuracy and confidence
(Rodriguez-Ruiz et al., 2020).
Despite the advancements, challenges remain in deploying AI-based
breast cancer detection systems in real-world clinical settings. Issues such as
data privacy, model interpretability, and regulatory compliance pose
significant barriers to adoption (Liu et al., 2021). Additionally, ensuring the
equitable distribution and accessibility of AI technologies across diverse
patient populations is essential to mitigate healthcare disparities (Ward et al.,
2020).
In conclusion, recent literature underscores the transformative
potential of AI and neural networking in revolutionizing breast cancer
detection. By leveraging advanced computational techniques and large-scale
datasets, researchers and clinicians are poised to enhance early diagnosis and
improve patient outcomes in the fight against breast cancer.
www.eujournal.org 44
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Methods
Below, we delve into the neural network approach facilitated by
MATLAB, offering specialized toolboxes tailored for neural networks,
leading to expedited, precise, and efficient outcomes.
A neural network comprises an input layer, one or more hidden layers,
and an output layer, with numerous nodes or neurons in each layer. These
nodes receive inputs from every node in the preceding layer, establishing
interconnectedness throughout the network. Modeled after the human brain,
neural networks learn through this layered structure, enabling pattern
recognition, data classification, and predictive analysis based on learned data.
Similar to human learning, neural networks can discern patterns in
various data types, such as voice or images, through iterative training. This
learning process involves adjusting connection weights between neurons
based on a predefined learning rule until the network achieves desired
performance. MATLAB simplifies neural network development, allowing
users to create models and implement pattern recognition with minimal
expertise.
MATLAB facilitates rapid neural network model creation,
visualization, integration into existing applications, and deployment across
diverse platforms. When developing artificial intelligence applications,
particularly neural networks, common steps include data preparation, AI
modeling, simulation, testing, and deployment.
A. Data Preparation:
In this stage, we gather a suitable dataset with labels and ensure its
quality through human examination. The neural network then undergoes
training using this dataset, with simulations generating the training process.
Additional data may be incorporated into the training set to capture a broader
range of patterns and variations.
B. AI Modeling:
When comparing the effectiveness of shallow neural networks with
traditional machine learning methods like decision trees or SVMs, or when
dealing with limited labeled training data, it's advisable to utilize the
command-line functions available in the Classification and Regression
Learner tool from the Statistics and Machine Learning Toolbox™. This
toolbox is well-suited for neural networks, offering comprehensive support for
specifying, training, and customizing neural network models. Additionally, it
facilitates interoperability, allowing seamless sharing and utilization of
information across different computer systems or applications.
www.eujournal.org 45
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
C. Simulation and Testing:
This stage facilitates the integration of intricate systems, such as neural
networks, and includes testing, simulation for larger systems, and validation
and verification of the system's functionality.
D. Deployment:
In this report, pattern recognition will be employed for implementation
instead of traditional coding methods.Overall, the methodology outlined in
this report will encompass the following steps, which will be elaborated upon
further:
A. Data preprocessing and classification.
B. Application of Artificial Neural Network (ANN) pattern
recognition using MATLAB.
C. Results and discussion.
The tool utilized in this study to enable a neural network to identify
signs and patterns of breast cancer is MATLAB. Within MATLAB's
algorithm, there are several tools available, including the neural network
pattern recognition (ANN classification pattern recognition) application. This
application assists in detecting malignant tumors, utilizing data sets sourced
from The UCI Machine Learning Repository.
According to MATLAB, the Artificial Neural Network (ANN) pattern
recognition process involves training the neural network to classify data sets
into specific categories, thus addressing classification challenges. Users can
import data sets from various applications such as Excel or any other
compatible data file recognized by MATLAB. Subsequently, the ANN is
trained, validated, and tested, with results displaying the error margin and
accuracy of the application.
The following subsections of the methodology detail the steps and
procedures involved in utilizing pattern recognition for classifying two data
sets: distinguishing between malignant and benign tumors, and identifying
blood diagnosis data from breast cancer patients versus healthy individuals.
A. Data Pre-processing and Classification:
The initial phase involved gathering the necessary data for utilizing
Artificial Neural Networks (ANN) in breast cancer detection. Data sets were
acquired from the UCI Machine Learning Repository, with all preprocessing
and organization documented and attached.
The primary dataset chosen from the repository is the "Breast Cancer
Coimbra Data Set." This dataset comprises 10 attributes and a total of 116
instances of individuals' laboratory test results. Among these instances, 64
www.eujournal.org 46
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
corresponded to breast cancer patients, while 52 were from individuals
without breast cancer. The data was presented in an Excel table format, with
column headers indicating attributes such as Age, BMI, Glucose, Insulin,
HOMA, Leptin, Adiponectin, Resistin, MCP-1, and a label (1 for healthy
controls, 2 for patients).
Figure 1: Breast Cancer Coimbra Data Set before preprocessing
To prepare the data for use in a neural network pattern recognition
application, it needs to be structured into inputs and target outputs. This
involves creating two new Excel sheets, which will be utilized in MATLAB.
The first sheet, named "Input-BreastCancer," includes the laboratory
results of individuals, excluding age and label information. Subsequently, the
rows of the dataset are shuffled to separate patients from healthy individuals,
enabling further testing of the pattern recognition capabilities of the ANN in
classification tasks.
Following this, the second Excel sheet, named "Output-
TargetBreastcancer," is created to contain the labels. However, the labels are
modified from 1 (indicating a healthy individual) and 2 (indicating a breast
cancer patient) to 0 (representing a healthy individual) and 1 (representing a
breast cancer patient). This adjustment makes the data readable to the pattern
recognition program's target output, allowing it to recognize these numbers as
a pattern. This completes the processing of the Coimbra dataset.
www.eujournal.org 47
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Figure 2: Breast Cancer Coimbra Data Set after processing, divided into two datasets of
input and output
The second dataset chosen from the repository is the "Breast Cancer
Wisconsin (Original) Data Set." This dataset comprises 11 attributes and a
total of 699 instances of tumor test results. Among these tumors, 458 were
classified as benign and 241 as malignant.The data is provided in the form of
a .DATA file, which can be imported into Excel for further processing. Once
imported, the data is formatted to default text with no formatted cells.
Subsequently, attention is turned to the column headers, which include
attributes such as Sample code number, Clump Thickness, Uniformity of Cell
Size, Uniformity of Cell Shape, Marginal Adhesion, Single Epithelial Cell
Size, Bare Nuclei, Bland Chromatin, Normal Nucleoli, Mitoses, and Class (2
for benign and 4 for malignant).
Figure 3: Breast Cancer Wisconsin Data Set before pre processing
To prepare this data for import into the neural network pattern
recognition application, it needs to be structured into inputs and target outputs.
This requires the creation of two new Excel sheets. The first sheet, named
"WisconsinBreastCancerIN," includes all the tumor's specification attributes
except for Sample code number and Class.
www.eujournal.org 48
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Following this, a second sheet called "WisconsinBreastCancerOUT" was
generated. This sheet contains the class information, but it was transformed
from 2 for benign and 4 for malignant to 0 for benign and 1 for malignant.
This adjustment ensures compatibility with the pattern recognition program's
target output. Thus, the processing and organization of the Wisconsin dataset
are concluded.
Figure 4: Breast Cancer Wisconsin Data Set after processing, divided into two datasets of
input and output
Following the data preprocessing phase, the next step involves
classifying the processed data using the neural network pattern recognition
algorithm, as outlined in the forthcoming subsection.
B. ANN Pattern Recognition Application in MATLAB
As previously explained in the methodology introduction, the ANN
pattern recognition process entails categorizing datasets into distinct
categories.
To elaborate further, the processed data, now labeled as input data and
output data, are imported into MATLAB. The input data corresponds to the
input, while the target output corresponds to the desired output. The
illustration below demonstrates how the Coimbra breast cancer datasets are
entered into MATLAB as an example:
For this research, it's important to note that all target outputs in the
datasets denote 1 for breast cancer patients or malignant tumors, and 0 for
healthy individuals or benign tumors.Subsequently, the datasets are
randomized to train the ANN, as well as for validation and testing purposes.
The user selects the percentages for each accordingly.
www.eujournal.org 49
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Afterwards, the hidden layers for the artificial neural network are entered
as well:
Lastly, the neural network training commences, yielding results
including data, graphs, error percentages, and more upon completion. If
the training results appear inadequate, adjusting the number of hidden
layers can improve classification accuracy and reduce errors.
Additionally, modifying the distribution of training, validation, and testing
datasets can impact results positively. In the final segment of the
methodology, the outcomes of the neural network training for classifying
the Coimbra and Wisconsin datasets will be presented and scrutinized.
The aim is to assess whether the ANN's pattern recognition successfully
identifies whether a tumor is benign or malignant in the case of the
Wisconsin dataset, or discerns a breast cancer patient from a healthy
individual in the Coimbra dataset.
Results And Discussion
The Coimbra Breast cancer dataset was partitioned into 55% for
training, 20% for validation, and the remaining 25% for testing purposes.
Additionally, the number of hidden layers was adjusted to 15 to optimize the
program's classification performance.
The result showed the following:
www.eujournal.org 50
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Before delving into the analysis, it's important to reiterate the ultimate
objective of this pattern recognition endeavor: to distinguish between
laboratory results belonging to breast cancer patients and those of healthy
individuals within the Coimbra dataset. As depicted above, the dataset
comprising 116 samples has been divided into 64 for training, 29 for
validation, and 23 for testing. To gain deeper insights into the classification
performance, it's crucial to examine the results of CE (cross-entropy) and %E
(percentage error). Below, we will discuss the analysis of training
performance, training state, error histogram, confusion matrix, and ROC plot.
Figure 5: NN training performance plot
The plot above reveals that the validation achieved its peak
performance at epoch 6, with a value of approximately 0.6. Additionally, a
gradual decline in training errors suggests a reduction in errors over time.
Moreover, the consistent decrease in the test line reaffirms the accuracy of the
program's pattern recognition, indicating successful detection of patterns.
www.eujournal.org 51
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Figure 6: Error Histogram
With 20 bars or bins displayed in the histogram, it becomes evident
that a zero error is observed around the -0.006 mark. These error histograms
provide insight into the discrepancies between our target and predicted digits
post-training of our neural network. The remarkably low error percentage of
0.006 underscores the high accuracy of our pattern recognition in detection,
indicating minimal errors in the process.
Figure 7: Confusion Matrix
The presented confusion matrix illustrates the breakdown of
classifications for training, validation, test, and a combined matrix,
respectively. Focusing on the training matrix, it reveals that the pattern
recognition accurately identified 28 attributes as 0, signifying healthy
individuals, and 16 attributes as breast cancer patients. The training's accuracy
stands at approximately 70%, with the error rate around 30%.
When considering the combined confusion matrix encompassing
training, validation, and testing, the neural network demonstrated a notable
accuracy level. Specifically, it correctly identified 43 attributes representing
www.eujournal.org 52
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
healthy individuals while misclassifying 30 laboratory attributes belonging to
breast cancer patients as healthy. Additionally, it accurately identified 34
laboratory attributes as indicative of breast cancer patients, yet erroneously
categorized 9 healthy individuals as having breast cancer. Overall, the neural
network achieved an accuracy rate of approximately 66%, with an error
percentage of 33%. This suggests a reliable performance, as it surpasses the
50% threshold.
Figure 8: Receiver Operating Characteristic (ROC) Plot
From the ROC curves presented above, it's evident that the neural
network's detection accuracy is steadily improving. The positive incline of
the ROC curve, particularly noticeable when it surpasses the linear line in the
center, indicates an enhancement in accuracy. Notably, when the ROC curve
consistently lies above the linear line across all four graphs, especially in the
combined ALL ROC, it underscores the neural network's proficiency in
distinguishing between healthy individuals and breast cancer patients based
on laboratory results.
In conclusion, the analysis of the Coimbra dataset results confirms
the neural network's capability to accurately detect relevant patterns in the
provided information.
Result and Discussion of Data and Graphs for Wisconsin:
Similar to the Coimbra dataset, the Wisconsin Breast Cancer dataset
was divided into training, validation, and testing sets, with 55% allocated for
training, 20% for validation, and 25% for testing. The neural network
architecture included 15 hidden layers to optimize its performance in
classification tasks, mirroring the setup used for the Coimbra dataset.
www.eujournal.org 53
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
The result showed the following:
As done in the Coimbra Dataset, a quick reminder of the following
before discussing the analysis, the end goal of this pattern recognition is to
detect which of the tumor results in the Wisconsin dataset is a malignant tumor
and which is a benign tumor. From the above results, the sample of 699 has
been separated into 384 for training, 175 for validation, and 140 for testing.
Again, to further understand the show results of CE and %E, An
analysis of the training performance, training state, error histogram, confusion
matrix, and ROC plot will be discussed below.
Figure 9: NN training performance plot
The graph above shows that the best performance of the validation is
at epoch 17 with the value of about 0.11, as well as it can be shown that the
training is also at a decrease, meaning that less errors are occurring. Finally,
to further harden that the detection of the program is majorly correct and
accurate, we can see that the test line, just like in the Coimbra dataset result,
has also been at a steady
www.eujournal.org 54
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
decrease line, which further indicates that yes, the pattern is being recognized
correctly.
Figure 10: Error Histogram
The histogram contains 20 bars or bins, its results signify that there is
a zero error laying at about - 0.01 since it's between -0.05 and 0.05 and it leans
a bit to the left. As stated before , the error histograms aids in allowing the user
to witness visually the errors between our target and our predicted digits after
training the neural network, so with our error percentage being as low as 0.01,
indicates that our pattern recognition in the detection has a high accuracy due
to its low error rate.
Figure 11: Confusion Matrix
Reading one of the matrices of the Wisconsin Breast cancer data set
that is training, we can witness that the neural network recognition has
correctly detected 257 attributes to be 0 , indicating benign tumors, and 127
www.eujournal.org 55
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
attributes indicating to malignant tumors, with the training’s accuracy is at an
extremely high 97% with the error being as low as 3% , indicating high and
accurate training that solidifies our ANN’s training capabilities of detecting
tumors. When we see the total confusion matrix, which again includes the
validation test and training combined together, we can see that the neural
network has correctly detected 445 tumor attributes that show to be benign
tumors whilst mis detecting 8 tumor attributes that belong to malignant tumors
as benign tumors, and has correctly detected 233 tumor attributes that show
malignant tumors while mis detecting 13 of those to be malignant even though
it is a benign tumor. It comes down to show that the total accuracy of this
neural network is at a correct accuracy of 97% , and has an extremely low
error percentage matrix of 3%, which shows that our ANN was more than
capable of accurately distinguishing malignant tumor patterns from benign
without having too many mistakes.
Figure 12: Receiver Operating Characteristic (ROC) Plot
Finally in terms of data analysis of Wisconsin, we can see from the
ROC above, the curve is , again shown to be a positive incline above that of
the linear line in the center, just like the Coimbra dataset , then it shows that
the neural network is yet again beginning to increase in terms of detection
accuracy. Since the curve is a positive incline above the line at the combined
ALL ROC, this solidifies that the artificial neural network pattern recognition
application is detecting the malignant tumors from the benign accurately,
without much fault nor errors.
Thus, this concludes the analysis of the Wisconsin dataset results and
its discussion. We finalize that yes, just like the Coimbra analysis, neural
network pattern recognition can also detect quite accurately when it comes to
the given information of the tumor attributes, thus concluding our results and
analysis, as well as our methodology.
www.eujournal.org 56
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
Conclusions
In summary, MATLAB has demonstrated its effectiveness in early
breast cancer detection through neural network-based pattern recognition,
effectively addressing classification challenges. The analysis conducted on the
Coimbra breast cancer dataset showcased the program's ability to minimize
errors while accurately recognizing patterns. Various metrics, including
training performance, error histograms, confusion matrices, and ROC plots,
confirmed its reliability in detecting breast cancer accurately.
Similarly, when applied to the Wisconsin Breast Cancer dataset,
MATLAB exhibited high accuracy in pattern recognition and detection due to
its low error rate. These findings underscore the significance of employing
artificial neural networks and pattern recognition techniques in aiding
healthcare professionals, such as doctors and radiologists, in diagnosing breast
cancer at early stages. MATLAB serves as a testament to how modern
technology continues to support and enhance breast cancer detection efforts,
benefiting the broader breast cancer community.
Conflict of Interest: The authors reported no conflict of interest.
Data Availability: All data are included in the content of the paper.
Funding Statement: The authors did not obtain any funding for this research.
References:
1. Duffy, S. W., Parmar, D., Jatoi, I., & Miller, A. B. (2020).
Fundamental Issues in Breast Cancer Screening. Journal of Clinical
Oncology, 38(15), 1637–1643.
2. Esteva, A., Kuprel, B., Novoa, R. A., Ko, J., Swetter, S. M., Blau, H.
M., & Thrun, S. (2019). Dermatologist-level classification of skin
cancer with deep neural networks. Nature, 542(7639), 115–118.
3. McKinney, S. M., Sieniek, M., Godbole, V., Godwin, J., Antropova,
N., Ashrafian, H., & Toma, M. (2020). International evaluation of an
AI system for breast cancer screening. Nature, 577(7788), 89–94.
4. Rodriguez-Ruiz, A., Krupinski, E., Mordang, J. J., Schilling, K.,
Heywang-Köbrunner, S. H., & Sechopoulos, I. (2019). Detection of
breast cancer with mammography: Effect of an artificial intelligence
support system. Radiology, 290(2), 305–314.
5. Rodriguez-Ruiz, A., Lång, K., Gubern-Merida, A., Broeders, M.,
Gennaro, G., Clauser, P., & Mann, R. M. (2020). Stand-alone artificial
intelligence for breast cancer detection in mammography: Comparison
with 101 radiologists. Journal of the National Cancer Institute, 112(9),
916–922.
www.eujournal.org 57
European Scientific Journal, ESJ ISSN: 1857-7881 (Print) e - ISSN 1857-7431
November 2024 edition Vol.20, No.33
6. Wang, T., Jing, C., Ma, Y., Lai, M., Zhang, Y., & He, Z. (2021).
Convolutional neural networks in breast cancer screening: A
comprehensive review. Frontiers in Medicine, 8, 664720.
7. Liu, Y., Chen, P.-H. C., Krause, J., Peng, L., LeCun, Y., & Deng, J.
(2021). How to Read Articles That Use Machine Learning: Users’
Guides to the Medical Literature. JAMA, 325(6), 519–529.
8. Ward, R., Deaton, A., El-Sayed, A. M., & Gray, J. (2020). Data
Science, Artificial Intelligence, and Machine Learning: Opportunities
for Advancement in Breast Cancer Research. JAMA Oncology, 6(10),
1521–1522.
www.eujournal.org 58
