Comprehensive Summary
The following study introduces CHASHNIt (Combined Hybrid Architecture for Scalable High-performance in Neural Iteration), a deep learning model for the classification of skin diseases. The framework integrates three state-of-the-art networks-EfficientNetB7, DenseNet201, and InceptionResNetV2 addressing class imbalance through GAN-based data augmentation. The following model is trained on 19,5500 images across 23 dermatological conditions from the DermNet database, after pre-processing steps including normalization, feature selection, and augmentation. Results demonstrate that the model has superior performance with 97.8% accuracy, 98.1% precision, 97.55% recall, 97.6% F1 score, and 92.3% IoU, which significantly outperforms other models like ResNet101, Swin Transformer, and MobileNetV3. The use of explainable AI tools provides heatmaps that align with dermatologist annotations, enhancing interpretability. Ablation studies confirmed that the hybrid model achieved substantial performance gain beyond its individual components.
Outcomes and Implications
The project addresses the significant obstacles in dermatological AI: a lack of transparency in prediction and data imbalance. Through a combination of GAn augmentation and XAI visualizations, CHASHNIt ensures both diagnostic reliability and interpretability, a critical step for clinical trust. Its exceptionally high accuracy suggests strong potential for integration into telehealth platforms and resource-limited settings where dermatologists are scarce. However, the reliance on a single dataset imposes high computational demands and remaining limitations, with the authors proposing future optimization for multi-center validation and edge deployment. In further validation, CHASHNIt could serve as a decision-support tool for dermatology, aiding in the timely recognition, triage, and referral of complex skin conditions.