Explainable AI-Driven Detection of Human Monkeypox Using Deep Learning and Vision Transformers: A Comprehensive Analysis

Limited training data remains a critical barrier to AI performance: The study found that building deep learning models from scratch was significantly hampered by dataset limitations, highlighting that publicly available mpox skin lesion datasets are insufficient for training robust classifiers without additional techniques.

Transfer learning substantially improves diagnostic accuracy: By using pre-trained models rather than training from scratch, the researchers achieved much better results, with MobileNet-v2 reaching 93.15% accuracy—demonstrating that leveraging existing knowledge can overcome data scarcity challenges in medical AI applications.

Multiple AI approaches can achieve clinically relevant performance: Three different models (MobileNet-v2, Vision Transformer B16, and ResNet-50) all achieved accuracy above 86%, suggesting that responsible AI deployment doesn't require a single "perfect" solution but rather multiple validated approaches that can be cross-checked.

Explainability was essential for validation: The researchers applied explainable AI techniques to further validate model performance, indicating that for medical applications, understanding why an AI makes a diagnosis is as important as the accuracy metric itself—a key requirement for clinical trust and responsible deployment.

AI shows promise for addressing diagnostic challenges in infectious disease: Since mpox symptoms closely resemble measles and chickenpox, AI-assisted analysis of skin lesions could help clinicians make faster, more accurate early diagnoses, potentially improving public health response times.