Multi-Modal Explainable AI Predictions for Dermatological Disease and Allergy Diagnosis

View at publisher Summarise with VIRAT
Authors: Britney Spierce G, Rajeswari M, Stewart Kirubakaran S
Cited by: 0
References: 16

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Millions of people around the world have suffered from skin diseases, and a lot of morbidity has resulted from delays or incorrect diagnoses. Dermatological diagnosis traditionally is based on subjective visual judgment by experts, which usually faces accessibility issues. This work proposes an AI-assisted framework for multi-class skin disease diagnosis that is enhanced with Explainable Artificial Intelligence for enhancing the trust, interpretability, and clinical usability of the developed model. The proposed system uses three deep learning architectures: ResNet50, DenseNet121, and a hybrid ResNet50 integrated with the Convolutional Block Attention Module (CBAM), to extract highly discriminative visual cues by focusing on relevant lesion areas both spatially and channel-wise. A curated dataset of dermoscopic skin lesion images has been used; several preprocessing and augmentation techniques like flipping, rotation, and color jittering were utilized to enhance robustness against diverse clinical imaging conditions. Weighted Random Sampling and class balanced loss are applied to mitigate challenges offered by class imbalance. Grad-CAM-based heatmaps provide clear visual explanations through highlighting of the disease specific lesion regions, strengthening clinical transparency. Also, it integrates the complete diagnostic workflow into one web interface with FastAPI, providing image upload, symptom-aware evaluation, differential diagnosis, and personalized food recommendations according to the predicted condition. The results affirm that the combination of attention enhancement, feature level interpretability, and calibrated prediction will improve the diagnostic accuracy and clinical trustworthiness of AI algorithms by leaps and bounds. Experimental evaluation has established that the proposed Hybrid ResNet50 + CBAM model achieves the topmost performance with 89 % accuracy, 87 % precision, 86 % recall, and$\mathbf{8 6. 5 \%}$F1-score, outperforming the baseline models ResNet50 and DenseNet121.

Fields of Study & Subjects

Cutaneous Melanoma Detection and Management Explainable Artificial Intelligence (XAI) Artificial Intelligence in Healthcare

Publication details

Journal / source
International Conference on Intelligent Data Communication Technologies and Internet of Things (IDCIoT)
Pages
1726–1733
Publication date
4 February 2026
Language
EN