A Hybrid Framework for Mpox Diagnosis and Severity Assessment via Symptom Classification and Skin Lesion Segmentation

Md Jewel; Karim Mohammed Rezaul; Kazy Noor e Alam Siddiquee; Md Shabiul Islam; Tajnuva Jannat

Centre for Applied Research in Software & IT (CARSIT), UK, UK · University of Westminster, UK, UK · Military Institute of Science & Technology (MIST), Bangladesh, Bangladesh · Multimedia University, Malaysia, Malaysia · Centre for Applied Research in Software & IT (CARSIT), UK, UK

Pages: 1-38   Licence: CC BY 4.0

Abstract

Monkeypox has re-emerged as a global health concern, demanding diagnostic solutions that reflect the complex clinical presentation of the disease. Traditional machine learning and deep learning approaches often isolate either symptom-based or image-based analysis, limiting diagnostic comprehensiveness. To address this gap, we propose a novel multimodal AI-driven framework that integrates structured symptom data with skin lesion image analysis for reliable detection and severity assessment. Separate symptom and image datasets were utilized to develop the respective models. For symptom-based classification, an ensemble of six machine learning models including LightGBM, XGBoost, CatBoost, Logistic Regression, Random Forest, and Support Vector Machine is employed. Concurrently, a majority voting ensemble of seven pre-trained deep learning models, including ResNet50, InceptionV3, MobileNetV2, EfficientNet, VGG16, Xception, and DenseNet, is used for image-based diagnosis. Positive image classifications trigger lesion segmentation using a customized U-Net model to quantify lesion burden. A severity scoring algorithm, designed according to World Health Organization (WHO) clinical guidelines, combines symptom severity and lesion metrics to classify cases as mild, moderate, severe, or serious. The proposed framework offers explainable, scalable, and clinically interpretable outputs, achieving higher diagnostic accuracy and vigorous severity estimation. Tailored for real-world applicability, particularly in outbreak-prone and resource-limited settings, our system provides a valuable AI-based tool to support early intervention, personalized care, and improved public health response.

Keywords: Monkeypox, Mpox, MPXV, Skin Lesion, Artificial Intelligence, Deep Learning, CNN.