Melanoscope AI CDSS Validated for Skin Cancer Screening in Russia
A recent study released on arXiv (2605.27561) details the clinical validation of Melanoscope AI, a mobile dermoscopy clinical decision support system (CDSS) aimed at the early identification of malignant skin lesions. This research tackles the issue of dermatologist shortages in various regions of Russia by introducing a cascade deep learning model that incorporates quantitative interpretability assessment through attention map visualization (using attention rollout for ViT and Swin, and Grad-CAM for ConvNeXt and EfficientNetV2) alongside IoU-based agreement with expert annotations. Additionally, a three-zone patient routing algorithm was created, and initial single-centre prospective clinical validation was performed across four Melan... The goal is to enhance screening coverage and standardize patient routing.
Key facts
- Melanoscope AI is a mobile dermoscopy CDSS.
- Early detection of malignant skin lesions is critical for prognosis.
- Dermatologist shortages in Russian regions limit screening coverage.
- Model interpretability and standardised patient routing are key barriers.
- Two-stage cascade classification of dermoscopic images is used.
- Attention map visualisation includes attention rollout for ViT and Swin, and Grad-CAM for ConvNeXt and EfficientNetV2.
- Quantitative IoU-based agreement assessment between activation maps and expert annotations.
- Preliminary single-centre prospective clinical validation across four Melan...
Entities
Institutions
- arXiv
Locations
- Russia