Mpox-XDE: an ensemble model utilizing deep CNN and explainable AI for monkeypox detection and classification.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Dermatology

Background:

• Human monkeypox (Mpox) is a growing global health concern requiring early identification.
• Current deep learning (DL) models for Mpox detection need further reliability improvements for early-stage diagnosis.
• Accurate and timely diagnosis is essential to prevent the spread of Mpox.

Purpose of the Study:

• To develop a robust and accurate ensemble deep learning model for early Mpox detection.
• To enhance the classification performance of existing DL models for Mpox identification.
• To provide a reliable tool for healthcare professionals in diagnosing Mpox.

Main Methods:

• An ensemble model, Mpox-XDE, was created by combining three modified DL models: Xception, DenseNet201, and EfficientNetB7.
• The Mpox Skin Images Dataset (MSID) with 770 images was utilized for training and testing.
• The ensemble model incorporated Softmax, dense, and flattened layers with dropout, and a global average pooling layer for classification.

Main Results:

• The Mpox-XDE model achieved high performance metrics: 98.70% testing accuracy, 98.90% precision, 98.80% recall, and 98.80% F1-score.
• The model successfully classified images into four categories: chickenpox, measles, normal, and Mpox.
• Explainable AI (XAI) using Grad-CAM visualized the model's decision-making process, highlighting relevant image regions.

Conclusions:

• The proposed Mpox-XDE ensemble model demonstrates exceptional accuracy in early Mpox detection from skin images.
• This methodology offers a significant advancement in diagnostic tools for Mpox.
• The explainability feature aids in understanding and trusting the model's predictions for clinical application.