Development and validation of the IVENUS artificial intelligence model for automated CEAP classification of early-stage chronic venous disease using lower limb photographs.

objectivesChronic venous disease (CVD) is one of the most common vascular disorders, affecting millions of people worldwide. Owing to the variability of clinical symptoms and the subjective nature of their interpretation, diagnosing CVD at an early stage is complicated, making it crucial for patients to consult a specialist. It was hypothesized that an artificial intelligence (AI) model could accurately classify CVD (C0-C2 Clinical, Etiology, Anatomy, and Pathophysiology (CEAP) clinical class) from lower limb photographs. Therefore, this study aimed to develop and validate such a model.

methodsA multicenter cross-sectional study (NCT17122021) was conducted from May 2020 to January 2024 in accordance with the Declaration of Helsinki. A dataset of 10,745 lower limb photographs was collected using smartphones and professional cameras across several Russian clinics, then standardized and anonymized before model training. CEAP clinical class was determined by consensus among three surgeons experienced in phlebology. The AI model IVENUS was developed to automatically assess lower limb photographs and classify early-stage CVD according to the CEAP clinical classification (C0-C2). The model was trained using a deep learning approach based on the Swin Transformer V2 architecture. To improve model robustness and reduce overfitting, Gaussian blurring and color jitter were applied as data augmentation methods during training. The standard performance metrics (sensitivity, recall, specificity, accuracy, and precision) were calculated.

resultsThe dataset consisted of 673 lower limbs of stage C0, 4445 lower limbs of stage C1, and 5627 lower limbs of stage C2. The overall diagnostic accuracy in the external validation subset of 1622 photographs was 84.8%, with a precision of 84.3%, sensitivity of 84.3%, and specificity of 92.3%.

conclusionsThe AI model IVENUS demonstrated high diagnostic value for early-stage CVD, sufficient for its application as a clinical decision support system. Therefore, this model may support patient self-screening and telemedicine triage and may be used by specialists for automated patient routing and tracking of treatment progress.