Copyright © 2026 Authors retain the copyright of this article. This article 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.

@article{196054,
        author = {Shaik Khwaja Mohammed Junaid and Petakamsetty Eswar Sai and Malla Srinivas and Ruttal Dheshik Sai and Md. Imam Khader Sharif},
        title = {An EfficientNet-Based Framework for Automated Skin Lesion Diagnosis with Explainable AI and Clinical Report Generation},
        journal = {International Journal of Innovative Research in Technology},
        year = {2026},
        volume = {12},
        number = {11},
        pages = {1765-1771},
        issn = {2349-6002},
        url = {https://ijirt.org/article?manuscript=196054},
        abstract = {Automated diagnosis of skin lesions from thermoscopic images plays a vital role in improving early detection of skin cancer and supporting clinical decision-making. However, accurate analysis is challenging due to low contrast, irregular lesion boundaries, visual similarity among classes, and the presence of artifacts such as hair and noise. This paper presents an end-to-end deep learning framework that integrates preprocessing, segmentation, classification, explainability, and report generation for automated skin lesion diagnosis.
Initially, thermoscopic images undergo preprocessing to enhance image quality by reducing noise and correcting illumination variations. A lightweight U-Net-based segmentation model is used to extract the lesion region of interest (ROI), enabling the model to focus on relevant features. The segmented ROI is then passed to an EfficientNetB1-based transfer learning classifier, pretrained on ImageNet and fine-tuned on the HAM10000 dataset, to classify lesions into eight diagnostic categories based on texture, color, and structural patterns.
To improve interpretability, Grad-CAM is employed to generate heatmaps highlighting important regions influencing predictions. Additionally, a Retrieval-Augmented Generation (RAG) pipeline using Lang Chain, Llama 3.2, and a FAISS-based knowledge base generates both clinician-oriented and patient-friendly reports.
The proposed system achieves 87.4% accuracy, a macro F1-score of 0.863, and an AUC of 0.962, demonstrating effectiveness for real-world dermatological applications.},
        keywords = {EfficientNet, Grad-CAM, HAM10000, RAG, Skin Lesion Diagnosis, Transfer Learning, U-Net Segmentation.},
        month = {April},
        }