Redefining brain tumor segmentation: a cutting-edge convolutional neural networks-transfer learning approach

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 14, No. 3, June 2024, pp. 2583∼2591
ISSN: 2088-8708, DOI: 10.11591/ijece.v14i3.pp2583-2591 ❒ 2583
Redefining brain tumor segmentation: a cutting-edge
convolutional neural networks-transfer learning approach
Shoffan Saifullah1,2
, Rafał Dreżewski1
1Faculty of Computer Science, AGH University of Krakow, Krakow, Poland
2Department of Informatics, Universitas Pembangunan Nasional Veteran Yogyakarta, Yogyakarta, Indonesia
Article Info
Article history:
Received Oct 23, 2023
Revised Dec 30, 2023
Accepted Jan 5, 2024
Keywords:
Brain tumor segmentation
Convolutional neural networks
-transfer learning
Deep learning
Magnetic resonance imaging
Medical image analysis
ABSTRACT
Medical image analysis has witnessed significant advancements with deep learn-
ing techniques. In the domain of brain tumor segmentation, the ability to
precisely delineate tumor boundaries from magnetic resonance imaging (MRI)
scans holds profound implications for diagnosis. This study presents an en-
semble convolutional neural network (CNN) with transfer learning, integrating
the state-of-the-art Deeplabv3+ architecture with the ResNet18 backbone. The
model is rigorously trained and evaluated, exhibiting remarkable performance
metrics, including an impressive global accuracy of 99.286%, a high-class accu-
racy of 82.191%, a mean intersection over union (IoU) of 79.900%, a weighted
IoU of 98.620%, and a Boundary F1 (BF) score of 83.303%. Notably, a de-
tailed comparative analysis with existing methods showcases the superiority of
our proposed model. These findings underscore the model’s competence in pre-
cise brain tumor localization, underscoring its potential to revolutionize medical
image analysis and enhance healthcare outcomes. This research paves the way
for future exploration and optimization of advanced CNN models in medical
imaging, emphasizing addressing false positives and resource efficiency.
This is an open access article under the CC BY-SA license.
Corresponding Author:
Shoffan Saifullah
Faculty of Computer Science, AGH University of Krakow
Krakow, Poland
Department of Informatics, Universitas Pembangunan Nasional Veteran Yogyakarta
Yogyakarta, Indonesia
Email: saifulla@agh.edu.pl, shoffans@upnyk.ac.id
1. INTRODUCTION
Brain tumors present a complex medical challenge that demands accuracy and efficiency in diagno-
sis [1]. This challenge is further compounded by the diverse morphology of brain tumors, spanning variations
in shape, size, and intensity. With advancements in medical imaging technologies, particularly magnetic reso-
nance imaging (MRI), there is an increasing opportunity to improve the precision of brain tumor detection. The
accurate segmentation of brain tumors from MRI scans plays a pivotal role in early diagnosis [2]. However,
manual segmentation methods are often time-consuming and prone to error [3], making the development of
automated and precise segmentation techniques essential [4], [5].
Traditional methods relied on handcrafted features and classical machine learning algorithms [6],
paving the way for early endeavors in deep learning for MRI detection. These techniques utilized texture
and shape features like gabor filters, gray level co-occurrence matrices (GLCM), zernike moments, region,
circularity, and wavelet transformations [7], [8]. Classifiers such as markov random field (MRF), artificial
Journal homepage: http://ijece.iaescore.com

2584 ❒ ISSN: 2088-8708
neural network (ANN), and support vector machine (SVM) achieved accuracy rates ranging from 75% to 98%,
playing a vital role in tissue categorization [9]. Advanced features and techniques like combining Zernike
moments with ANN-Gabor wavelets with SVM classifiers were explored, alongside experiments evaluating
texture and shape features with naı̈ve Bayes (NB) classifiers [10].
The advent of deep learning, particularly convolutional neural networks (CNNs), transformed MRI
classification in brain tumor detection. However, early attempts with CNNs faced challenges due to limited
sample sizes and overfitting risks. The nuances of MRI detection, including the diverse nature of brain tumors
and dataset imbalances, added complexity to the quest for automated detection [11]. Proposing transfer learn-
ing from pre-trained CNNs addressed these issues, showcasing an initial architecture achieving an 84.19%
accuracy in classification [12]. MRI detection encountered challenges in brain tumor variability and dataset
imbalances. Researchers aimed to automate detection without manual segmentation, incorporating additional
metrics (precision, sensitivity, and specificity) for accurate detection assessment.
Recent advances in deep learning, marked by innovative methods like capsule networks (CapsNets),
deep residual networks (ResNets), and inception models, have reshaped brain tumor detection. Integration of
multiple architectures, novel approaches, and ensemble techniques addressed spatial boundary complexities
in segmentation [13]. However, the journey towards optimal brain tumor segmentation persists, with the ap-
plication of asymmetric and symmetric network architectures, novel loss functions, and knowledge exchange
strategies [14], [15]. These developments highlight the continuous evolution of medical image analysis, steer-
ing towards enhanced accuracy and efficiency.
In response to these challenges, our study introduces ensemble CNNs with transfer learning, integrat-
ing the Deeplabv3+ architecture with the ResNet18 backbone to redefine the landscape of brain tumor segmen-
tation. Deep learning has shown remarkable potential in automatically learning intricate patterns in complex
data, and the concept of transfer learning, which adapts pre-trained CNN models [16], has emerged as a critical
factor in enhancing their performance. Our research focuses on developing and implementing a CNN-transfer
learning framework tailored explicitly to brain tumor segmentation. By harnessing the knowledge embedded
in pre-trained models and fine-tuning them for tumor detection, we aim to significantly improve the accuracy
and efficiency of brain tumor segmentation in medical practice.
This article unfolds as follows: section 2 presents our CNN-transfer learning framework’s method-
ology, section 3 unveils experimental results, and section 4 provides a thoughtful conclusion with insights into
future research directions. Our article aims to underscore the transformative potential of the CNN-transfer
learning framework, promising a revolution in brain tumor detection and, by extension, the broader landscape
of medical image analysis
2. METHOD
Our brain tumor prediction model relies on a robust deep learning architecture to harness the predictive
power of CNNs and the knowledge transfer capabilities of transfer learning. We have tailored this architecture
to excel in medical image segmentation, specifically for brain tumor localization. The core elements of this
architecture include:
2.1. Data collection and preprocessing
Data quality and preprocessing are critical pillars in our brain tumor prediction and segmentation
methodology. For this task, we sourced a dataset from Kaggle, curated by Nikhil Tomar [17]. This dataset
consists of 3,064 MRI images, each paired with its corresponding ground truth image as shown in Figure 1.
A subset of MRI images as shown in Figure 1(a) was randomly selected for visual inspection to ensure our
data’s uniformity and high quality. These images were overlaid with their corresponding ground truth masks
as shown in Figure 1(b), a crucial step to verify proper alignment between the MRI and ground truth masks –
a prerequisite for practically training our prediction
2.2. Base CNN model: ResNet18
Our ensemble CNN-transfer learning architecture [18] relies on the DeepLabV3+ with ResNet18
model, forming the backbone of our brain tumor prediction system. ResNet18, renowned for its deep ar-
chitecture as shown in Figure 2 and residual connections, facilitates the direct transfer of information between
layers, mitigating the vanishing gradient problem during training. With 18 layers, ResNet18 strikes an optimal
balance between depth and capacity, enabling it to discern intricate patterns within medical images. Leveraging
Int J Elec & Comp Eng, Vol. 14, No. 3, June 2024: 2583-2591

Int J Elec & Comp Eng ISSN: 2088-8708 ❒ 2585
pre-trained knowledge from the ImageNet dataset, the model efficiently identifies pertinent features in medical
images. Fine-tuning tailors the model to brain tumor segmentation, refining its capacity to make precise predic-
tions. ResNet18’s deep structure, residual connections, and pre-trained foundation make it a powerful choice
for accurately identifying and segmenting brain tumors in medical images.
(a) (b)
Figure 1. Dataset of (a) MRI brain images and (b) the ground truth
Figure 2. ResNet18 architecture for brain tumor segmentation
2.3. DeepLabv3+ layers and ensemble approach
The efficiency of our ensemble CNN-transfer learning system relies on the innovative architecture
of DeepLabv3+, as shown in Figure 3. This model excels in semantic segmentation, emphasizing precise
object boundary delineation crucial for medical image analysis [19]. Atrous or dilated convolution expands the
receptive field without increasing parameters, ensuring accurate segmentation by capturing features from fine-
grained to high-level details. Atrous spatial pyramid pooling (ASPP) and feature refinement modules enhance
the model’s proficiency in recognizing large and small tumor regions.
Our ensemble leverages multiple CNN outputs to enhance accuracy, particularly in complex tasks like
medical image segmentation [20]. Blending ResNet18’s feature extraction with DeepLabv3+’s architecture
allows the ensemble to capture diverse features at different scales and resolutions. This strategic fusion en-
Redefining brain tumor segmentation: a cutting-edge convolutional ... (Shoffan Saifullah)

2586 ❒ ISSN: 2088-8708
sures robust performance, mitigating overfitting risks and promoting generalization to new data. The ensemble
achieves high accuracy, showcasing the adaptability of deep learning in medical image analysis.
2.4. Training, validation, and parameter configuration of segmentation
The training and validation phase is crucial for developing our brain tumor segmentation model, in-
volving the meticulous partitioning of the dataset into training, validation, and testing subsets. The training
dataset, which contains annotated brain MRI scans, is the foundation for instructing the model to identify tu-
mor regions. Simultaneously, the validation dataset, which is kept separate during training, plays a pivotal role
in performance monitoring, overfitting detection, and hyperparameter refinement.
Figure 3. Ensemble CNN-Resnet18 architecture using DeepLabV3+ for brain tumor segmentation
The configuration of training parameters is vital for achieving optimal model performance, preventing
overfitting, and ensuring efficient convergence [6]. Leveraging stochastic gradient descent with momentum
(SGD) as the optimizer, we dynamically adjust model weights to minimize the loss function. Key parameters,
including the learning rate and L2 regularization, are carefully tuned to prevent overfitting. Batch processing
enhances training efficiency, and periodic evaluations on the validation dataset facilitate progress tracking.
Early stopping ensures prompt conclusion if performance stagnates.
These meticulously adjusted parameters collectively contribute to a model achieving accuracy and
robust generalization [6]. Post-training, the inference and segmentation phase marks the practical application
of our trained model to previously unseen brain MRI scans. Pixel-wise segmentation maps are generated, aiding
accurate diagnosis and treatment planning. This transformative capability showcases the substantial impact of
deep learning in advancing medical imaging and healthcare.
2.5. Performance evaluation metrics
Our semantic segmentation model is assessed using key metrics [21]. Accuracy gauges overall classifi-
cation performance by calculating the ratio of correctly classified samples to the total number
(Accuracy = (T P +T N)
(T P +F P +T N+F N) ). Precision focuses on correctly classifying positive samples, consider-
ing true positives and false positives (Precision = T P
(T P +F P ) ). Recall evaluates the model’s effectiveness in
identifying relevant instances, using true positives and false negatives (Recall = T P
(T P +F N) ). The F-measure
(F1 score), of dice coefficient, a balance of precision and recall, is computed as F1 = 2x P recisionxRecall
(P recision+Recall) .
In addition, we utilize global accuracy for overall pixel correctness and mean accuracy for class-
specific pixel accuracy, addressing imbalances. Finally, intersection over union (IoU) measures semantic seg-
mentation by assessing the overlap between correctly classified pixels and the ratio of ground truth to predicted
pixels. IoU values, ranging from 0 to 1, indicate the similarity between ground truth and model predictions.

3. RESULT AND DISCUSSION
This section outlines the methods employed in the proposed CNN-transfer learning framework for
improved brain tumor detection and segmentation. This framework combines the power of convolutional neural
networks (CNNs) with transfer learning to enhance the accuracy of MRI-based brain tumor classification and
segmentation. The following subsections delve into the critical components of this approach.
3.1. Marker dataset creation and tumor detection (DeepLabv3+ with ResNet18)
This section outlines the creation of marker datasets and brain tumor detection using our ensemble
CNN-transfer learning with Deeplabv3+ architecture with ResNet18 backbone. Figure 4 showcases foun- da-
tional MRI datasets with green-marked tumor boundaries for training and testing. Following the modeling
process, experiments with separate datasets as shown in Figure 4(a) reveal accurate tumor detection but preci-
sion variations against the ground truth as shown in Figure 4(b). In Figure 4(b), i) closely matches the ground
truth, ii) and iii) show close approximations, and iv) exhibit perceptible deviations. Green lines represent the
ground truth, and red lines signify model predictions.
(a)
(b)
Figure 4. Samples of (a) dataset with ground truth annotations and (b) segmentation prediction with red
(prediction) and green (ground truth) comparisons
This nuanced analysis illuminates the method’s overall effectiveness in brain tumor detection and
provides valuable insights into specific areas that could benefit from refinement. The detailed examination of
Figure 4(b) reveals the model’s successes and underscores the imperative for ongoing fine-tuning to enhance
segmentation precision. This emphasis on continuous improvement is particularly crucial when addressing the
intricate challenges of certain tumor complexities. By recognizing and addressing these nuances, the model can
evolve further, ensuring a more robust and accurate approach to detecting brain tumors in diverse scenarios.
3.2. Performance analysis based on model training and testing
This section comprehensively analyzes our brain tumor prediction model based on CNN-ResNet18.
Rigorous training and testing procedures were implemented to ensure the model’s accuracy and loss as shown
in Figures 5(a) and 5(b). Throughout the training phase, the model exhibited consistent improvement over
ten epochs, starting with a modest 15.38% accuracy during the initial epoch and achieving an impressive
99.72% accuracy by the tenth epoch as shown in Figure 6. This significant enhancement in training accuracy
underscores the model’s effectiveness in learning from the dataset, showcasing its proficiency in accurately
detecting brain tumors.
The evolution of accuracy over the training epochs is graphically depicted in Figure 5(a). This vi-
sualization showcases the remarkable growth in accuracy, highlighting the model’s learning capability as it
becomes increasingly adept at identifying and classifying brain tumors. Additionally, Figure 5(b) provides

2588 ❒ ISSN: 2088-8708
insights into the loss graphs of the model during training. These loss graphs reveal how the model’s error
decreases as training progresses, emphasizing its ability to refine its predictions over time.
We utilized a normalized confusion matrix to assess the model’s classification performance. This
matrix provides valuable insights into the model’s true positive and false positive rates for brain tumors and
background regions. The confusion matrix as shown in Table 1 illustrates the percentages of predicted brain
tumors correctly identified (64.5%) and the correctly classified background regions (99.69%). It also indicates
that false positives are minimal (0.1134%), signifying the model’s precision in classifying non-tumor regions.
The confusion matrix suggests several vital observations: high true positive rate, low false negative rate, low
false positive rate, and high true negative rate. These observations collectively affirm the model’s effectiveness
in classifying tumor and non-tumor regions exactly.
(a)
(b)
Figure 5. Training progress (a) accuracy graph and (b) loss graphs
The semantic segmentation results as shown in Table 2 offer crucial insights into the model’s accurate
prediction of brain tumor regions across two experiments, emphasizing its exceptional proficiency. Metrics, in-
cluding global accuracy, Mean Accuracy, Mean IoU, Weighted IoU, and Mean BF-Score, showcase the model’s
consistent and accurate predictions, highlighting its precision in delineating tumor regions and preserving fine-
grained details. The remarkable global accuracy of 0.99286 in the first experiment and 0.97480 in the second,
along with Mean Accuracy scores of 0.82191 and 0.95860, reflect the model’s pixel-wise precision. Mean
IoU scores of 0.79900 and 0.93403 demonstrate significant overlap with ground truth regions, while Weighted
IoU scores of 0.98620 and 0.95089 highlight the model’s versatility in handling class imbalances. Notably,
Mean BF-Score values of 0.83303 and 0.91239 underscore the model’s exceptional capability in preserving
fine-grained tumor details crucial for medical image segmentation, providing valuable evidence of the model’s
effectiveness.
Table 1. Confusion matrix of segmentation results
Predicted brain tumor Predicted Background
True brain tumor 64.5 33.5
True background 0.1134 99.89
Table 2. Performance of semantic segmentation results
Experiment Global accuracy Mean accuracy Mean IoU Weight IoU Mean BF-Score
1st 0.99286 0.82191 0.79900 0.98620 0.83303
2nd 0.97480 0.95860 0.93403 0.95089 0.91239

3.3. Comparison with other methods
The performance of our ensemble CNN-transfer learning model in brain tumor segmentation is thor-
oughly examined in this section. Performance metrics, including dice coefficient (0.91239), mean IoU (0.93403),
and accuracy (0.97480), highlight the model’s exceptional accuracy and proficiency in tumor segmentation.
Our model demonstrates superior performance compared to alternative methods in Table 3, such as cascaded
dual-scale LinkNet and segnet-VGG-16. The proposed method has a dice coefficient of 0.91239, indicating
precise spatial overlap between predicted and ground truth. Its high Mean IoU of 0.93403 reflects a substan-
tial alignment between predicted and ground truth, highlighting the model’s proficiency in delineating tumor
boundaries accurately. Moreover, the accuracy score of 0.97480 emphasizes the model’s effectiveness in over-
all classification, showcasing its ability to distinguish between tumor and non-tumor regions with reliability.
Table 3. Comparison of proposed methods with others
No Methods Dice coefficient Mean IoU Accuracy
1 Proposed method 0.91239 0.93403 0.97480
2 Cascaded Dual-Scale LinkNet [22] 0.8003 0.9074 -
3 Segnet-VGG-16 [23] 0.9314 0.914 0.9340
4 2D-UNet [24] 0.8120 - 92.16
5 CNN with LinkNet [25] 0.73 - -
6 U-Net with adaptive thresholding [26] 0.6239 - 0.9907
7 O2U-Net [27] 0.8083 - 0.9934
8 CNN U-Net [28] - 0.8196 0.9854
While competing approaches, including cascaded dual-scale LinkNet and 2D-UNet, demonstrate re-
spectable metrics, the proposed method consistently outperforms both in terms of the Dice coefficient and mean
IoU, showcasing its advanced precision in tumor segmentation. Specifically, our method competes closely with
Segnet-VGG-16, achieving comparable results in dice coefficient and mean IoU, underscoring its suitability for
accurate tumor segmentation. The model’s high global accuracy, substantial mean accuracy, and remarkable
mean IoU underscore its precision in pixel segmentation and tumor region delineation. Complementary met-
rics, such as weighted IoU and mean BF-Score, further affirm the model’s ability to preserve fine-grained tumor
details. These outcomes position the model as a powerful tool in neuroradiology, promising enhanced precision
in brain tumor detection, particularly in cases with intricate nuances that challenge human assessment.
4. CONCLUSION
In this study, we have developed and rigorously assessed an ensemble CNN-transfer learning frame-
work, leveraging Deeplabv3+ architecture with ResNet18 backbone, for the intricate task of brain tumor seg-
mentation in medical images. The detailed comparison with various existing methods reinforces the superior
performance of our proposed model, demonstrating consistently higher dice coefficient and mean IoU. The
research outcomes affirm the model’s robustness and accuracy, as evidenced by remarkable global accuracy,
class accuracy, intersection over union (IoU), weighted IoU, and Boundary F1 (BF) score—critical metrics in
medical imaging. Our model’s demonstrated capabilities underscore its potential as a valuable tool for precise
brain tumor localization, a crucial aspect of medical diagnosis.
Integrating cutting-edge deep learning techniques into medical image segmentation signifies a signifi-
cant advancement in the healthcare sector. Beyond reducing subjectivity, these innovations can vastly improve
diagnostic precision and enhance the overall quality of patient care. As we look ahead, the research presented
here sets the stage for future endeavors to address the pertinent challenges and limitations, such as mitigat-
ing false positives and optimizing resource usage. By overcoming these obstacles, we can further refine and
elevate the model’s performance, solidifying the role of advanced CNN models in various medical imaging
applications.
ACKNOWLEDGEMENT
We extend our heartfelt gratitude to AGH University of Krakow and the Ministry of Education and
Science of Poland for their invaluable support, both in collaboration and financial backing, which has greatly
contributed to this research. Their unwavering assistance has been instrumental in the success of this study.

2590 ❒ ISSN: 2088-8708
REFERENCES
[1] S. Alsubai, H. U. Khan, A. Alqahtani, M. Sha, S. Abbas, and U. G. Mohammad, “Ensemble deep learning for brain
tumor detection,” Frontiers in Computational Neuroscience, vol. 16, 2022, doi: 10.3389/fncom.2022.1005617.
[2] A. B. Abdusalomov, M. Mukhiddinov, and T. K. Whangbo, “Brain tumor detection based on deep learning approaches
and magnetic resonance imaging,” Cancers, vol. 15, no. 16, 2023, doi: 10.3390/cancers15164172.
[3] H. Jiang et al., “A review of deep learning-based multiple-lesion recognition from medical images:
classification, detection and segmentation,” Computers in Biology and Medicine, vol. 157, 2023, doi:
10.1016/j.compbiomed.2023.106726.
[4] S. Saifullah and R. Dreżewski, “Enhanced medical image segmentation using CNN based on histogram equalization,”
in 2023 2nd
International Conference on Applied Artificial Intelligence and Computing (ICAAIC), May 2023, pp.
121–126, doi: 10.1109/ICAAIC56838.2023.10141065.
[5] S. Saifullah and R. Dreżewski, “Modified histogram equalization for improved CNN medical image segmentation,”
Procedia Computer Science, vol. 225, pp. 3021–3030, 2023, doi: 10.1016/j.procs.2023.10.295.
[6] Z. Liu et al., “Deep learning based brain tumor segmentation: a survey,” Complex and Intelligent Systems, vol. 9, no.
1, pp. 1001–1026, 2023, doi: 10.1007/s40747-022-00815-5.
[7] P. K. Ramtekkar, A. Pandey, and M. K. Pawar, “A comprehensive review of brain tumour detection mechanisms,”
The Computer Journal, 2023, doi: 10.1093/comjnl/bxad047.
[8] S. Kaur, Y. Kumar, A. Koul, and S. Kumar Kamboj, “A systematic review on metaheuristic optimization techniques
for feature selections in disease diagnosis: open issues and challenges,” Archives of Computational Methods in Engi-
neering, vol. 30, no. 3, pp. 1863–1895, 2023, doi: 10.1007/s11831-022-09853-1.
[9] S. Krishnapriya and Y. Karuna, “A survey of deep learning for MRI brain tumor segmentation methods: trends,
challenges, and future directions,” Health and Technology, vol. 13, no. 2, pp. 181–201, 2023, doi: 10.1007/s12553-
023-00737-3.
[10] S. Maurya, S. Tiwari, M. C. Mothukuri, C. M. Tangeda, R. N. S. Nandigam, and D. C. Addagiri, “A review on recent
developments in cancer detection using machine learning and deep learning models,” Biomedical Signal Processing
and Control, vol. 80, 2023, doi: 10.1016/j.bspc.2022.104398.
[11] S. Ali, J. Li, Y. Pei, R. Khurram, K. ur Rehman, and T. Mahmood, “A comprehensive survey on brain tumor diag-
nosis using deep learning and emerging hybrid techniques with multi-modal MR image,” Archives of Computational
Methods in Engineering, vol. 29, no. 7, pp. 4871–4896, 2022, doi: 10.1007/s11831-022-09758-z.
[12] E. Irmak, “Multi-classification of brain tumor MRI images using deep convolutional neural network with fully opti-
mized framework,” Iranian Journal of Science and Technology - Transactions of Electrical Engineering, vol. 45, no.
3, pp. 1015–1036, 2021, doi: 10.1007/s40998-021-00426-9.
[13] E. Ghafourian, F. Samadifam, H. Fadavian, P. Jerfi Canatalay, A. R. Tajally, and S. Channumsin, “An ensemble model
for the diagnosis of brain tumors through MRIs,” Diagnostics, vol. 13, no. 3, 2023, doi: 10.3390/diagnostics1303056
[14] H. Abbasi, M. Orouskhani, S. Asgari, and S. S. Zadeh, “Automatic brain ischemic stroke segmentation with deep
learning: a review,” Neuroscience Informatics, vol. 3, no. 4, 2023, doi: 10.1016/j.neuri.2023.100145.
[15] S. Saifullah, B. Yuwono, H. C. Rustamaji, B. Saputra, F. A. Dwiyanto, and R. Drezewski, “Detection of chest X-ray
abnormalities using CNN based on hyperparameters optimization,” Engineering Proceedings, vol. 52, pp. 1–7, 2023.
[16] S. Iqbal, A. N. Qureshi, J. Li, and T. Mahmood, “On the analyses of medical images using traditional machine
learning techniques and convolutional neural networks,” Archives of Computational Methods in Engineering, vol. 30,
no. 5, pp. 3173–3233, 2023, doi: 10.1007/s11831-023-09899-9.
[17] N. Tomar, “Brain tumor segmentation (A 2D brain tumor segmentation dataset),” Kaggle, 2022.
https://www.kaggle.com/datasets/nikhilroxtomar/brain-tumor-segmentation/data (accessed Sep. 15, 2023).
[18] S. Saifullah et al., “Nondestructive chicken egg fertility detection using CNN-transfer learning algorithms,” Jurnal
Ilmiah Teknik Elektro Komputer dan Informatika (JITEKI), vol. 9, no. 3, pp. 854–871, 2023.
[19] O. Akcay, A. C. Kinaci, E. O. Avsar, and U. Aydar, “Semantic segmentation of high-resolution airborne images
with dual-stream DeepLabV3+,” ISPRS International Journal of Geo-Information, vol. 11, no. 1, Dec. 2021, doi:
10.3390/ijgi11010023.
[20] R. Karthik, R. Menaka, M. Hariharan, and D. Won, “Ischemic lesion segmentation using ensemble of
multi-scale region aligned CNN,” Computer Methods and Programs in Biomedicine, vol. 200, 2021, doi:
10.1016/j.cmpb.2020.105831.
[21] A. Kumar, “Study and analysis of different segmentation methods for brain tumor MRI application,” Multimedia
Tools and Applications, vol. 82, no. 5, pp. 7117–7139, 2023, doi: 10.1007/s11042-022-13636-y.
[22] Z. Sobhaninia, S. Rezaei, N. Karimi, A. Emami, and S. Samavi, “Brain tumor segmentation by cascaded deep neural
networks using multiple image scales,” in 2020 28th Iranian Conference on Electrical Engineering (ICEE), 2020,
doi: 10.1109/ICEE50131.2020.9260876.
[23] A. Rehman, S. Naz, U. Naseem, I. Razzak, and I. A. Hameed, “Deep autoencoder-decoder framework for semantic
segmentation of brain tumor,” Australian Journal of Intelligent Information Processing Systems, vol. 15, no. 4, pp.
54–60, 2019.

[24] K. Sailunaz, D. Bestepe, S. Alhajj, T. Özyer, J. Rokne, and Reda Alhajj, “Brain tumor detection and segmentation:
Interactive framework with a visual interface and feedback facility for dynamically improved accuracy and trust,”
PLoS ONE, vol. 18, 2023, doi: 10.1371/journal.pone.0284418.
[25] Z. Sobhaninia et al., “Brain tumor segmentation using deep learning by type specific sorting of images,”
arXiv:1809.07786, Sep. 2018.
[26] B. V. Isunuri and J. Kakarla, “Fast brain tumour segmentation using optimized U-Net and adaptive thresholding,”
Automatika, vol. 61, no. 3, pp. 352–360, 2020, doi: 10.1080/00051144.2020.1760590.
[27] S. A. Zargari, Z. S. Kia, A. M. Nickfarjam, D. Hieber, and F. Holl, “Brain tumor classification and segmentation using
dual-outputs for u-Net architecture: O2U-Net,” Studies in Health Technology and Informatics, vol. 305, pp. 93–96,
2023, doi: 10.3233/SHTI230432.
[28] C. B. Ruiz, “Classification and segmentation of brain tumor MRI images using convolutional neu-
ral networks,” in 2023 IEEE International Conference on Engineering Veracruz (ICEV), 2023, doi:
10.1109/ICEV59168.2023.10329651.
BIOGRAPHIES OF AUTHORS
Shoffan Saifullah received a bachelor’s degree in informatics engineering from Univer-
sitas Teknologi Yogyakarta, Indonesia, in 2015 and a Master’s Degree in computer science from
Universitas Ahmad Dahlan, Yogyakarta, Indonesia, in 2018. He is a lecturer at Universitas Pemban-
gunan Nasional “Veteran” Yogyakarta, Indonesia. His research interests include image processing,
computer vision, and artificial intelligence. He is currently a Ph.D. student at AGH University of
Krakow, Poland, with a concentration in the field of artificial intelligence (bio-inspired algorithms),
image processing, and medical image analysis. He can be contacted at email: shoffans@upnyk.ac.id
and saifulla@agh.edu.pl.
Rafał Dreżewski received the M.Sc., Ph.D., and D.Sc. (Habilitation) degrees in computer
science from the AGH University of Krakow, Poland in 1998, 2005, and 2019, respectively. Since
2019, he has been an associate professor with the Institute of Computer Science, AGH University of
Krakow, Poland. He is the author of more than 80 papers. His research interests include bio-inspired
artificial intelligence algorithms and agent-based modeling and simulation of complex and emergent
phenomena. He can be contacted at email: drezew@agh.edu.pl.

Redefining brain tumor segmentation: a cutting-edge convolutional neural networks-transfer learning approach

More Related Content

Similar to Redefining brain tumor segmentation: a cutting-edge convolutional neural networks-transfer learning approach (20)

More from IJECEIAES (20)

Recently uploaded (20)

Redefining brain tumor segmentation: a cutting-edge convolutional neural networks-transfer learning approach