A deep-learning algorithm to classify skin lesions from mpox virus infection.

Area of Science:

• Dermatology
• Infectious Diseases
• Artificial Intelligence

Background:

• The mpox virus (MPXV) outbreak is exacerbated by undetected infections and delayed isolation of affected individuals.
• Early detection of MPXV is crucial for effective outbreak containment and public health management.

Purpose of the Study:

• To develop and validate an image-based deep convolutional neural network (MPXV-CNN) for the early identification of MPXV skin lesions.
• To assess the performance of the MPXV-CNN across diverse populations and skin tones.

Main Methods:

• A large dataset of 139,198 skin lesion images was curated, including non-MPXV and MPXV images.
• A deep convolutional neural network (MPXV-CNN) was trained and validated on this dataset.
• The model's performance was evaluated using sensitivity, specificity, and AUC metrics in validation, testing, and prospective cohorts.

Main Results:

• The MPXV-CNN demonstrated high diagnostic performance, with sensitivity and specificity reaching 0.91 and 0.898 in the testing cohort, respectively.
• The model exhibited robust classification performance across various skin tones and body regions.
• A web-based application was developed to provide accessible patient guidance using the MPXV-CNN.

Conclusions:

• The MPXV-CNN is a promising tool for the early detection of MPXV skin lesions, potentially aiding in outbreak mitigation.
• The developed AI tool can support clinical decision-making and improve the speed of diagnosis.
• Accessible AI-driven diagnostic tools can significantly contribute to managing infectious disease outbreaks.