Deep transfer learning approaches for Monkeypox disease diagnosis.

Area of Science:

• Medical Imaging
• Computational Biology
• Artificial Intelligence

Background:

• Monkeypox presents significant global health challenges with increasing daily cases.
• Infected individuals exhibit diverse skin symptoms and pose a transmission risk through contamination.
• Machine Learning (ML) has demonstrated success in image-based medical diagnoses, including cancer and COVID-19.

Purpose of the Study:

• To develop and evaluate a Monkeypox diagnosis model utilizing Generalization and Regularization-based Transfer Learning approaches (GRA-TLA).
• To assess the model's performance in both binary and multiclass classification tasks.
• To investigate the computational efficiency and interpretability of the proposed GRA-TLA method.

Main Methods:

• Developed a Monkeypox diagnosis model using GRA-TLA for binary and multiclass classification.
• Tested the approach on ten different Convolutional Neural Network (CNN) models across three studies.
• Employed Local Interpretable Model-Agnostic Explanations (LIME) for model prediction interpretation and feature extraction.

Main Results:

• The GRA-TLA approach combined with Extreme Inception (Xception) achieved 77%-88% accuracy in distinguishing Monkeypox cases in Studies One and Two.
• Residual Network (ResNet)-101 demonstrated superior performance for multiclass classification, with accuracies ranging from 84% to 99% in Study Three.
• The proposed GRA-TLA method proved computationally efficient regarding the number of parameters (NP) and Floating-Point Operations per Second (FLOPs) compared to existing Transfer Learning (TL) approaches.

Conclusions:

• The developed GRA-TLA model shows significant potential for accurate and efficient Monkeypox diagnosis from skin images.
• The model's interpretability using LIME provides valuable insights into disease indicators.
• This ML-based approach offers a promising tool for addressing the global Monkeypox challenge.