Academic Journal of Computing & Information Science, 2026, 9(1); doi: 10.25236/AJCIS.2026.090103.
Tongzhu Zhao1, Yu Xiang1, Wei Wang1, Yonghao Wu1, Tiancai Zhu1
1School of Information Science and Technology, Yunnan Normal University, Kunming, China
Early diagnosis of skin cancer is crucial for improving patient survival rates. Existing deep learning-based classification models predominantly employ fixed, static attention mechanisms, which struggle to adaptively capture subtle features of lesions that vary in scale and morphology. This study proposes a dynamic attention network based on deep reinforcement learning (ResNetRL), aiming to enhance the model’s classification performance for melanoma by dynamically regulating multi-scale attention weights in real-time. We embed Dual-Scale Attention Modules (DAS) into the four stages of ResNet50, where the scaling factors for channel attention and spatial attention are dynamically adjusted by the DDPG (Deep Deterministic Policy Gradient) algorithm. A hybrid reward function is innovatively designed, incorporating three metrics: classification accuracy, loss trend, and lesion region stability. Evaluated on the ISIC skin cancer datasets (ISIC 2017 and ISIC 2019), the proposed method achieves classification accuracies of 89.81%, 88.83%, and 91.38%, respectively. This study validates the effectiveness of dynamic attention mechanisms in medical image analysis and provides a novel research approach and methodology for the field.
Melanoma Classification; Residual Network; Dynamic Attention; Deep Reinforcement Learning; DDPG
Tongzhu Zhao, Yu Xiang, Wei Wang, Yonghao Wu, Tiancai Zhu. Melanoma Image Classification Based on Reinforcement Learning and Dynamic Attention Mechanisms. Academic Journal of Computing & Information Science (2026), Vol. 9, Issue 1: 24-32. https://doi.org/10.25236/AJCIS.2026.090103.
[1] Kong, X.: Research on Melanoma Classification Based on Deep Learning [D]. Zhejiang University of Technology, 2019. DOI: 10.27463/d.cnki.gzgyu.2019.000407.
[2] Ruffano, L.F., Takwoingi, Y., Dines, J., et al.: Computer-assisted diagnosis techniques (dermoscopy and spectroscopy-based) for diagnosing skin cancer in adults [J]. Cochrane Database of Systematic Reviews, 12: CD012526 (2018).
[3] Shen, D., Wu, G., Suk, H.I.: Deep learning in medical image analysis [J]. Annual Review of Biomedical Engineering, 19: 221–248 (2017).
[4] Yu, Z., Ni, D., Chen, S., et al.: Hybrid dermoscopy image classification framework based on deep convolutional neural network and Fisher vector [C]. In Proceedings of the IEEE 14th International Symposium on Biomedical Imaging, Melbourne: IEEE, pp. 301–304 (2017).
[5] Tian, D., Li, F., Zhang, X., Zhang, J.: Construction and Application of a Classification Model Combining Feature Selection and Deep Reinforcement Learning [J]. Electronic Design Engineering, 30(12): 93–97 (2022).
[6] Zhang, J., Xie, Y., Xia, Y., et al.: Attention residual learning for skin lesion classification [J]. IEEE Transactions on Medical Imaging, 38(9): 2092–2103 (2019).
[7] Lin, G., Gong, T., Lin, T.: A Survey on Attention Mechanisms in Image Classification Tasks [J]. Telecommunication Engineering, 65(08): 1349–1362 (2025). DOI: 10.20079/j.issn.1001-893x.241105001.
[8] Hu, J., Shen, L., Sun, G.: Squeeze-and-excitation networks [C]. In Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, Piscataway: IEEE, pp. 7132–7141 (2018). DOI: 10.1109/CVPR.2018.00745.
[9] Thrun, S., Littman, M.L.: Reinforcement Learning: An Introduction [J]. IEEE Transactions on Neural Networks, 16(1): 285–286 (2005).
[10] Busoniu, L., Babuska, R., De Schutter, B., Bart, P.: A comprehensive survey of multiagent reinforcement learning [J]. IEEE Transactions on Systems, Man, and Cybernetics Part C (Applications and Reviews), 38: 156–172 (2008).
[11] Chen, Z., Liu, Z., Wan, L., et al.: A Survey on Multi-Agent Reinforcement Learning Theory and Appli-cations [J]. Pattern Recognition and Artificial Intelligence, 37(10): 851–872 (2024). DOI: 10.16451/j.cnki.issn1003- 6059.202410001.
[12] Meng, X., Li, J., Yu, C., et al.: A Review of Deep Learning in Skin Lesion Image Segmentation [J/OL]. Journal of Frontiers of Computer Science and Technology, 1–27 (2025-12-05).
[13] Rao, A., M.V.M.K., Moonesar, I.A., Atalla, S., Prashanth, B.S., Joshi, G., Soni, T.K., Le, T., Verma, A., Marashdeh, H.: Cross Country Determinants of Investors’ Sentiments Prediction in Emerging Markets Using ANN [J]. Frontiers in Artificial Intelligence, 5: 912403 (2022). DOI: 10.3389/frai.2022.912403.
[14] Zhou, S.K., Le, H.N., Luu, K., Nguyen, H.V., Ayache, N.: Deep reinforcement learning in medical imaging: A literature review [J]. Medical Image Analysis, 73: 102193 (2021). DOI: 10.1016/j.media.2021.102193.
[15] Wu, M., Gao, Y., Zheng, J., Zhang, Q., Du, S.: Actor-Dueling-Critic Methods in Reinforcement Learn- ing [J]. Sensors (Basel), 19(7): 1547 (2019). DOI: 10.3390/s19071547.
[16] Liu, Y., Wu, P., Liu, W.: Research on Intelligent Diagnosis of Knee Joint Fracture Medical Images Based on ResNet18 and MSC Attention Mechanism [J]. Software Engineering, 28(11): 16–21 (2025). DOI: 10.19644/j.cnki.issn2096-1472.2025.011.004.
[17] Li, Y., Zhao, P., Zhang, X., et al.: MRI Radiomics Combined with ResNet101 Deep Learning for Dis- criminating Lumbar Brucellar Spondylitis and Spinal Metastatic Cancer [J]. Chinese Journal of Medical Imaging Technology, 41(06): 958–962 (2025). DOI: 10.13929/j.issn.1003-3289.2025.06.023.
[18] Cheng, C., Wang, Y., Song, Q., et al.: A Lightweight ECA-ResNeXt for Smart Agriculture and Its Application in Rice Disease Recognition [J]. Agricultural Engineering, 15(07): 28–35 (2025). DOI: 10.19998/j.cnki.2095-1795.202507304.