An Explainable AI-Driven Clinical Decision Support System for Skin Cancer Detection Using Deep and Texture Feature Fusion

Abstract

Skin cancer is among the most common forms of cancer worldwide, and its early detection plays a crucial role in improving patient survival rates. However, existing automated systems based purely on deep learning often lack interpretability and do not incorporate clinical reasoning. In this work, we propose an explainable Clinical Decision Support System (CDSS) for binary skin cancer classification using a hybrid feature fusion approach. The model combines deep features extracted from a pretrained ResNet18 network with handcrafted texture features obtained from Gray-Level Co-occurrence Matrix (GLCM) and wavelet transforms. This combination enables better differentiation between benign and malignant lesions by capturing both high-level and fine-grained image characteristics. Additionally, the system integrates ABCDE-based clinical rules for risk assessment and utilizes Grad-CAM to provide visual explanations of model predictions. Experimental evaluation on the HAM10000 dataset shows that the proposed approach achieves an accuracy of 84.23% and a malignant recall of 80.89%, outperforming baseline CNN models. The results demonstrate that the system not only improves prediction performance but also enhances transparency, making it more suitable for real-world clinical decision support.

Introduction

The text presents a deep learning–based Clinical Decision Support System (CDSS) for skin cancer detection, designed to improve both accuracy and interpretability in medical diagnosis.

Skin cancer, especially melanoma, requires early detection, but traditional diagnosis is subjective and depends heavily on dermatologist expertise. While deep learning models like CNNs (ResNet, DenseNet, EfficientNet) have improved classification performance, they often act as black-box systems and ignore clinically meaningful reasoning such as the ABCDE rule (Asymmetry, Border, Color, Diameter, Evolution), limiting trust and real-world adoption.

To address this, the proposed system introduces a hybrid explainable framework that combines:

• Deep features from ResNet18
• Handcrafted texture features using GLCM and wavelet transforms
• Clinical ABCDE-based rule analysis for structured assessment
• Grad-CAM for visual explanation of predictions

These features are fused and used for binary classification of skin lesions (benign vs malignant). The system is evaluated on the HAM10000 dataset, achieving 84.23% accuracy and 80.89% malignant recall, while also improving interpretability.

The literature review shows that existing methods either achieve good accuracy or good feature extraction, but rarely combine deep learning, handcrafted features, clinical rules, and explainability together. Most prior works also lack transparency, making clinical adoption difficult.

Conclusion

This study presents an explainable Clinical Decision Support System for binary skin cancer detection using a hybrid feature fusion approach. By combining deep features from ResNet18 with handcrafted texture descriptors such as GLCM and wavelet transforms, the system effectively captures both high-level and fine-grained characteristics of skin lesions. The experimental results on the HAM10000 dataset demonstrate an accuracy of 84.23% and a malignant recall of 80.89%, indicating strong performance in identifying cancerous cases. The inclusion of ABCDE-based clinical analysis and Grad-CAM visualization further enhances interpretability, making the system more aligned with real-world clinical practices. Overall, the integration of feature fusion, clinical reasoning, and explainable AI provides a reliable and practical framework for AI-assisted dermatological screening, with potential to support early diagnosis and improve decision-making in healthcare settings.

References

[1] A. Esteva et al., “Dermatologist-level classification of skin cancer with deep neural networks,” Nature, vol. 542, no. 7639, pp. 115–118, Feb. 2017. [2] P. Tschandl et al., “The HAM10000 dataset: A large collection of multi-source dermatoscopic images of common pigmented skin lesions,” Scientific Data, vol. 5, no. 1, 2018. [3] R. M. Haralick, K. Shanmugam, and I. Dinstein, “Textural features for image classification,” IEEE Transactions on Systems, Man, and Cybernetics, vol. 3, no. 6, pp. 610–621, Nov. 1973. [4] B. Harangi, “Skin lesion classification with ensembles of deep convolutional neural networks,” Journal of Biomedical Informatics, vol. 86, pp. 25–32, Oct. 2018. [5] M. A. Kassem et al., “Skin lesions classification into eight classes for HAM10000 dataset using deep CNN and transfer learning,” IEEE Access, vol. 8, pp. 102851–102863, 2020. [6] A. A. Adegun and S. Viriri, “Deep learning-based system for automatic melanoma detection,” IEEE Access, vol. 8, pp. 7160–7172, 2020. [7] R. R. Selvaraju et al., “Grad-CAM: Visual explanations from deep networks via gradient-based localization,” in Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2017, pp. 618–626. [8] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2016, pp. 770–778.

Copyright

Copyright © 2026 Mrs. G. Archana, Pasham Manikanta Reddy, Pamulaiah Gari Uday Kumar, Pichika Venkatesh. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.