Tri-Net: unified deep learning for skin lesion and symptom-based monkeypox detection

Area of Science:

• Medical Informatics
• Artificial Intelligence
• Epidemiology

Background:

• The re-emergence of Monkeypox necessitates improved diagnostic tools due to challenges in traditional visual inspection and symptom overlap with other diseases.
• Current diagnostic methods for Monkeypox can be subjective and prone to misjudgment, particularly in early stages.
• Accurate and timely diagnosis is crucial for effective Monkeypox management and control.

Purpose of the Study:

• To develop and evaluate Tri-Net, a novel deep learning approach for reliable Monkeypox detection.
• To integrate dermatological lesion analysis with symptom-based prediction for enhanced diagnostic accuracy.
• To provide a rapid, non-invasive, and highly accurate AI-driven tool for early Monkeypox intervention.

Main Methods:

• Tri-Net utilizes an ensemble of EfficientNetB4, Inception-ResNet V2, and DenseNet201 for skin lesion image classification.
• A dedicated Convolutional Neural Network (CNN) processes patient-reported symptoms for diagnostic precision.
• The model was trained and validated on the Monkeypox Skin Lesion Dataset (MSLD), including diverse lesion imagery and clinical profiles.

Main Results:

• The integrated Tri-Net model demonstrated high accuracy and robustness in detecting Monkeypox.
• The symptom-based module alone achieved a prediction accuracy of 97.86%.
• Tri-Net significantly outperformed existing diagnostic methods for Monkeypox.

Conclusions:

• Tri-Net offers a promising AI-driven solution for non-invasive and rapid Monkeypox diagnosis.
• The integration of visual and clinical data enhances diagnostic capabilities for emerging infectious diseases.
• This approach supports early intervention strategies and effective public health responses to Monkeypox outbreaks.