Related Experiment Videos
Explainable AI-driven hybrid deep learning framework for accurate skin cancer diagnosis
Abdullah Al Sakib1, Sm Masfequier Rahman Swapno2, Fahim Ahamed3
1Department of Information Technology, Westcliff University, Irvine, CA, USA.
Digital Health
|April 3, 2026
Summary
This study introduces a hybrid deep learning model for accurate skin cancer classification, enhancing diagnostic robustness and providing interpretable AI explanations for clinical decision support.
Area of Science:
- Dermatology
- Artificial Intelligence
- Medical Imaging
Background:
- Skin cancer classification accuracy is crucial for patient outcomes.
- Deep learning models offer potential but often lack interpretability.
- Robustness under varying image acquisition conditions remains a challenge.
Purpose of the Study:
- To develop and evaluate a hybrid, partially interpretable deep learning (DL) approach for multi-class skin cancer classification.
- To improve model robustness under diverse acquisition conditions.
- To deliver clinically meaningful explanations for DL predictions.
Main Methods:
- A hybrid DL pipeline combining EfficientNetB0 CNN with a Random Forest classifier.
- Preprocessing steps include hair artifact removal and noise reduction.
- Probability-level fusion of model outputs and Grad-CAM for explainability.
Main Results:
- Achieved 98.61% accuracy and 98.60% F1-score on a combined dataset.
- Demonstrated 95.02% accuracy and 95.06% F1-score on the HAM10000 dataset.
- High sensitivity and AUC for melanoma detection, with Grad-CAM highlighting diagnostic areas.
Conclusions:
- Partially interpretable DL architectures show promise for robust skin cancer classification.
- The framework, with Grad-CAM and a web interface, can serve as a clinical decision-support tool.
- This approach enhances diagnostic accuracy and provides valuable insights into model reasoning.
