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

Abstract

The recent re-emergence of Monkeypox emphasizes the critical importance of ensuring precise and prompt diagnostic capabilities. Traditional diagnosis typically depends on visually examining skin lesions and listening to patients describe their symptoms. However, this approach can be subjective and sometimes leads to misjudgment. To complicate matters further, Monkeypox shares many symptoms with other skin conditions and viral infections, making it difficult for doctors to identify the disease accurately in its early stages. In response, we present Tri-Net, a novel deep learning approach that integrates skin lesion analysis with symptom-based prediction for reliable detection of Monkeypox infection. The skin lesion module leverages an ensemble of three powerful architectures EfficientNetB4, Inception-ResNet V2, and DenseNet201 within the Tri-Net framework to classify images across different stages of the disease. Complementing this, a dedicated Convolutional Neural Network (CNN) processes patient-reported symptoms to further improve diagnostic precision. Our proposed Tri-Net model was trained on publicly available Monkeypox Skin Lesion Dataset (MSLD) comprises diverse dermatological lesion imagery and clinical symptom profiles. The combined approach demonstrated high accuracy, robustness, and significantly outperformed existing methods. Notably, the symptom-based module alone achieved a prediction accuracy of 97.86%. By uniting visual and clinical features, Tri-Net offers a non-invasive, rapid, and highly accurate diagnostic tool that supports early intervention and effective disease control. This study marks a meaningful step toward AI-driven diagnostics for emerging infectious diseases like Monkeypox.