Related Experiment Videos
Fuzzy cognitive explainable AI framework integrating ResNet-50 and causal clinical concept reasoning for skin lesion
Ram Kumar Yadav1, Avishek Nandi2, Vinesh Kumar3
1Department of Data Science and Engineering, Manipal University Jaipur, Dehmi Kalan, Off Jaipur Ajmer Expressway, Jaipur, 303007, India. ram.yadav@jaipur.manipal.edu.
Scientific Reports
|July 7, 2026
Summary
This study introduces Fuzzy-Cognitive Explainable AI (FC-XAI) for melanoma detection, linking deep learning predictions to clinical concepts like asymmetry and border irregularity for improved trust and explainability in dermatology AI.
Area of Science:
- Dermatology
- Artificial Intelligence
- Medical Imaging
Background:
- Melanoma detection from dermoscopic images is challenging.
- Deep learning (DL) models offer high accuracy but lack clinical explainability.
- Existing methods provide pixel-level explanations, not clinically relevant concepts.
Purpose of the Study:
- To develop a hybrid AI framework (FC-XAI) for melanoma detection that provides clinically meaningful explanations.
- To bridge the gap between DL predictions and dermatologists' diagnostic criteria (ABCD).
- To quantify uncertainty when DL predictions and explanations disagree.
Main Methods:
- A two-stage framework: ResNet-50 for melanoma detection, followed by SHAP for region identification.
- Mapping salient regions to clinical concepts (Asymmetry, Border, Colour, Diameter) forming a Fuzzy Cognitive Map (FCM).
- Evaluating FC-XAI on ISIC and PH2 datasets for accuracy, recall, AUC, and explanation coherence.
Main Results:
- The DL model achieved an AUC of 0.9832 and 84% accuracy on the ISIC dataset.
- FC-XAI generated structured, concept-level explanations (e.g., high asymmetry predicts malignancy).
- Medium-uncertainty scenarios showed higher accuracy (82.4%) than low-uncertainty (57.1%), indicating triage potential.
Conclusions:
- FC-XAI provides clinically relevant, concept-level explanations for DL-based melanoma detection.
- The framework enhances trust and interpretability in AI for dermatology.
- FC-XAI represents a significant step towards reliable and trustworthy AI in clinical practice.