Monkeypox detection using deep neural networks.

Area of Science:

• Medical Imaging Analysis
• Artificial Intelligence in Healthcare
• Dermatology

Background:

• The 2022 Monkeypox epidemic highlighted the need for rapid diagnostic tools.
• Lessons from the COVID-19 pandemic underscore the importance of pandemic preparedness.
• Global health concerns necessitate advanced disease detection methods.

Purpose of the Study:

• To develop and evaluate deep neural network (DNN) models for Monkeypox detection using skin images.
• To compare the performance of seven different DNN architectures in identifying Monkeypox.
• To enhance diagnostic transparency and clinical trust through explainable AI.

Main Methods:

• A dataset of skin images including Monkeypox, Chickenpox, Measles, and normal cases was utilized.
• Seven DNN models were developed and tested in two-class (Monkeypox vs. others) and four-class (all categories) scenarios.
• Local Interpretable Model-Agnostic Explanations (LIME) and Gradient-weighted Class Activation Mapping (Grad-Cam) were employed for model interpretability.

Main Results:

• The DenseNet201 model achieved the highest performance, with 97.63% accuracy and 90.51% F1-Score in the two-class scenario.
• In the four-class scenario, DenseNet201 reached 95.18% accuracy and 89.61% F1-Score.
• LIME and Grad-Cam successfully identified key image regions contributing to Monkeypox diagnosis, increasing model explainability.

Conclusions:

• The DenseNet201 model demonstrates superior performance for Monkeypox detection from skin images.
• Explainable AI techniques (LIME, Grad-Cam) are crucial for understanding diagnostic decisions and building clinical trust.
• The proposed model serves as a valuable auxiliary tool for diagnosing Monkeypox and differentiating it from similar skin conditions.