TMS: Ensemble Deep Learning Model for Accurate Classification of Monkeypox Lesions Based on Transformer Models with...

Area of Science:

• Medical Informatics
• Computer Vision
• Epidemiology

Background:

• Monkeypox (MPX) outbreaks outside endemic regions necessitate rapid diagnostic tools.
• Differentiating MPX from similar dermatological conditions like chickenpox and measles is clinically challenging.
• The Monkeypox Skin Lesion Dataset (MSLD) was curated for this study, containing 770 images from 162 patients across four classes: MPX, measles, chickenpox, and normal.

Purpose of the Study:

• To develop and evaluate an ensemble machine learning model for accurate monkeypox skin lesion classification.
• To address the challenge of early clinical differentiation between monkeypox and other exanthematous diseases.
• To create a computational tool for enhanced disease monitoring and outbreak management.

Main Methods:

• An ensemble model was created by integrating transformer models and a Support Vector Machine (SVM) classifier.
• Seven Convolutional Neural Network (CNN) architectures were evaluated for feature extraction.
• The top four performing CNNs (EfficientNetB0, ResNet50, MobileNet, Xception) were selected for feature extraction, with subsequent concatenation and optimization before SVM classification.

Main Results:

• The proposed ensemble model achieved a high diagnostic accuracy of 95.45% for monkeypox detection.
• The model demonstrated excellent performance metrics, including precision (95.51%), recall (95.45%), and F1 score (95.46%).
• These results underscore the model's effectiveness in accurately identifying monkeypox lesions within the dataset.

Conclusions:

• The hybrid ensemble framework exhibits robust diagnostic performance for monkeypox detection.
• The model's high accuracy and computational efficiency suggest its potential as a valuable clinical decision support tool.
• This approach can contribute to improved disease surveillance and management strategies during outbreaks.