Iranian Journal of Medical Physics

Iranian Journal of Medical Physics

Shuffle Attention–Enhanced ResNet50 for Robust Skin Lesion Classification on Dermoscopic Images

Document Type : Original Paper

Authors
Seyed Vahab Shojaedini 1
Amir Nooraliei 2
Mahsa Monajemi 3
1 Electrical Engineering Department, Iranian Research Organization for Science and Technology, Tehran, Iran
2 Department of Computer and IT, Qa .c., Isalamic Azad university , Qazvin , Iran
3 Department of Biomedical Engineering, Qa .c., Isalamic Azad university , Qazvin , Iran
10.22038/ijmp.2026.96772.2718
Abstract
Introduction: Skin cancer is a prevalent global malignancy, and despite CNN advances in automated diagnosis, challenges in feature discrimination, robustness, and performance stability remain.
Moreover, most previous studies primarily report single-run performance metrics, providing limited insight into the reproducibility and stability of deep learning models in clinically relevant settings.
Material and Methods: In this study, a Shuffle Attention–enhanced ResNet50 (SA-ResNet50) architecture was developed for automated skin lesion classification. The proposed model was systematically compared with baseline ResNet50 and ResNet50 integrated with the Convolutional Block Attention Module (CBAM). Experiments were conducted on two publicly available dermoscopic datasets, PH2 and HAM10000, using two data-partitioning strategies. To evaluate model robustness, independent-repeated ten times and K-fold cross validation methods were used for experiments, and performance variability was analyzed across repeated runs.
Results: The proposed SA-ResNet50 consistently outperformed the baseline models on both datasets while exhibiting lower performance variability. On the PH2 dataset, the proposed model achieved an accuracy of 98.9 ± 0.4%, sensitivity of 99.2 ± 0.3%, and specificity of 99.6 ± 0.2%. On the HAM10000 dataset with the first data partitioning strategy, SA-ResNet50 achieved an accuracy of 90.2 ± 1.6%, sensitivity of 85.7 ± 2.0%, and specificity of 92.8 ± 1.3%. In second strategy, the proposed model achieved an accuracy of 84.31±3.48%, Macro-F1 score of 71.86±7.23%, Balanced Accuracy of 70.29±9.36%, Matthews Correlation Coefficient (MCC) of 68.15±9.36, and an Area Under the ROC Curve (AUC) of 92.47±1.91, demonstrating robust performance on the highly imbalanced HAM10000 dataset with second data partitioning strategy.
Conclusion: The experimental results demonstrate that integrating the Shuffle Attention module into the ResNet50 backbone improves discriminative feature representation and enhances model robustness and convergence stability for dermoscopic skin lesion classification across benchmark datasets. These findings suggest that the proposed SA-ResNet50 constitutes an effective and reliable framework for automated skin lesion classification.
Keywords
Skin Lesion Classification
Shuffle Attention
Deep Learning
Dermoscopy
Subjects
1.           Bray F, Laversanne M, Sung H, Ferlay J, Siegel RL, Soerjomataram I, et al. Global cancer statistics 2022: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin. 2024;74(3):229–63.
2.           Goyal M, Knackstedt T, Yan S, Hassanpour S. Artificial intelligence-based image classification methods for diagnosis of skin cancer: Challenges and opportunities. Comput Biol Med. 2020;127:104065.
3.           Adebiyi A, Abdalnabi N, Hoffman E, Md S, Hirner J, Simoes EJ, et al. Accurate Skin Lesion Classification Using Multimodal Learning on the HAM10000 Dataset. medRxiv. 2024;1–17.
4.           Minarno AE, Fadhlan M, Munarko Y, Chandranegara DR. Classification of Dermoscopic Images Using CNN-SVM. JOIV Int J Informatics Vis. 2024;8(2):606–12.
5.           Gurung S, Gao YR. Classification of melanoma (skin cancer) using convolutional neural network. In: 2020 5th International Conference on Innovative Technologies in Intelligent Systems and Industrial Applications (CITISIA). IEEE; 2020. p. 1–8.
6.           Nasir M, Attique Khan M, Sharif M, Lali IU, Saba T, Iqbal T. An improved strategy for skin lesion detection and classification using uniform segmentation and feature selection based approach. Microsc Res Tech. 2018;81(6):528–43.
7.           Zghal NS, Derbel N. Melanoma skin cancer detection based on image processing. Curr Med Imaging Rev. 2020;16(1):50–8.
8.           Sreedhar B, BE MS, Kumar MS. A comparative study of melanoma skin cancer detection in traditional and current image processing techniques. In: 2020 Fourth International Conference on I-SMAC (IoT in Social, Mobile, Analytics and Cloud)(I-SMAC). IEEE; 2020. p. 654–8.
9.           Brinker TJ, Hekler A, Enk AH, Klode J, Hauschild A, Berking C, et al. A convolutional neural network trained with dermoscopic images performed on par with 145 dermatologists in a clinical melanoma image classification task. Eur J Cancer. 2019;111:148–54.
10.         Vivek N, Ramesh K. Deep Learning Multi-Modal Melanoma Detection: Algorithm Development and Validation. JMIR AI. 2025;4:e66561.
11.         Hu J, Shen L, Sun G. Squeeze-and-excitation networks. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2018. p. 7132–41.
12.         Wang Q, Wu B, Zhu P, Li P, Zuo W, Hu Q. ECA-Net: Efficient channel attention for deep convolutional neural networks. In: Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2020. p. 11534–42.
13.         Woo S, Park J, Lee JY, Kweon IS. CBAM: Convolutional block attention module. In: Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics). 2018. p. 3–19.
14.         Park J, Woo S, Lee JY, Kweon IS. Bam: Bottleneck attention module. arXiv Prepr arXiv180706514. 2018;
15.         Ding X, Qian K, Zhang Q, Jiang X, Dong L. Dual-channel compression mapping network with fused attention mechanism for medical image segmentation. Sci Rep. 2025;15(1):8906.
16.         Zhang QL, Yang YB. Sa-net: Shuffle attention for deep convolutional neural networks. In: ICASSP 2021-2021 IEEE international conference on acoustics, speech and signal processing (ICASSP). IEEE; 2021. p. 2235–9.
17.         Nasir IM, Alshaya H, Tehsin S, Bouchelligua W. Transformer-Based Foundation Learning for Robust and Data-Efficient Skin Disease Imaging. Diagnostics 2026, Vol 16, Page 440. 2026 Feb 1;16(3):440.
18.         Hark BD. Beyond Accuracy: Methodological Evaluation of Performance Stability in Deep Learning Models on the HAM10000 Multi-Class Skin Lesion Benchmark. Türkiye Klin Biyoistatistik. 2025;17(3):140–50.
19.         He K, Zhang X, Ren S, Sun J. Deep residual learning for image recognition. In: Proceedings of the IEEE conference on computer vision and pattern recognition. 2016. p. 770–8.
20.         Mendonça T, Ferreira PM, Marques JS, Marcal ARS, Rozeira J. PH 2-A dermoscopic image database for research and benchmarking. In: 2013 35th annual international conference of the IEEE engineering in medicine and biology society (EMBC). IEEE; 2013. p. 5437–40.
21.         Tschandl P, Rosendahl C, Kittler H. Data descriptor: The HAM10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Sci Data. 2018;5.
22.         Khan MA, Mazhar T, Ali MD, Khattak UF, Shahzad T, Saeed MM, et al. Automatic melanoma and non-melanoma skin cancer diagnosis using advanced adaptive fine-tuned convolution neural networks. Discov Oncol. 2025;16(1):645.
23.         Balaha HM, Hassan AES. Skin cancer diagnosis based on deep transfer learning and sparrow search algorithm. Neural Comput Appl. 2023;35(1):815–53.
24.         Çevik E, Zengin K. Classification of skin lesions in dermatoscopic images with deep convolution network. Avrupa Bilim ve Teknol Derg. 2019;309–18.
25.         Kondaveeti HK, Edupuganti P. Skin cancer classification using transfer learning. In: 2020 IEEE International Conference on Advent Trends in Multidisciplinary Research and Innovation (ICATMRI). IEEE; 2020. p. 1–4.
26.         Reddy ND. Classification of dermoscopy images using deep learning. arXiv Prepr arXiv180801607. 2018;
27.         Maniraj SP, Maran PS. A hybrid deep learning approach for skin cancer diagnosis using subband fusion of 3D wavelets. J Supercomput. 2022;78(10):12394–409.
28.         Vidya Lakshmi V, Leena Jasmine JS. A Hybrid Artificial Intelligence Model for Skin Cancer Diagnosis. Comput Syst Sci Eng. 2021;37(2):233–45.
29.         Jacob S, Rosita JD. Fractal Model for Skin Cancer Diagnosis Using Probabilistic Classifiers. Int J Adv Signal Image Sci. 2021;7(1):21–9.
Iranian Journal of Medical Physics
Volume 23, Issue 2 - Serial Number 2
March and April 2026
Pages 132-141

  • Receive Date 23 June 2026
  • Revise Date 07 August 2026
  • Accept Date 05 September 2026