Journal of vascular surgery. Venous and lymphatic disorders · 2026
Artificial intelligence-assisted identification of skin lesions associated with chronic venous insufficiency.
СтатьяНазвание и аннотация: Перевод в очереди
Обработка статьи · этапы, ошибки и управление
Загрузка состояния…
Ошибки и ограничения
История · было → стало
Состояние обновляется каждые 5 секунд, пока блок открыт. Процент показывается только там, где известен объём работы. Пауза останавливает продолжение этапа; текущий запрос к источнику может завершиться.
Kim Ko Eun, Na Jin Cheol, Song SangJun, Jeon Jiehyun, Baek Seungjun, Kim Beom Suk
Аннотация
OBJECTIVE: This study aimed to develop and validate a deep learning framework to classify chronic venous insufficiency (CVI)-related skin lesions and differentiate them from other lower-extremity dermatological conditions using dermoscopic images.
METHODS: We retrospectively analyzed 677 high-resolution dermoscopic images from 248 patients, all histopathologically confirmed by skin biopsy. The dataset was categorized into three clinically distinct groups: (1) skin conditions related to CVI (eg, stasis dermatitis), (2) common inflammatory dermatological diseases, and (3) vasculitis. A Swin Transformer-based architecture (Microsoft Research)-a deep learning model that processes images at multiple spatial scales to capture both fine local detail and broader contextual patterns-was implemented to analyze dermoscopic images. A strict patient-level split was used to ensure model robustness, so that images from the same patient were never shared between training and test sets, which could otherwise artificially inflate performance estimates.
RESULTS: The proposed Swin Transformer model demonstrated superior diagnostic performance, achieving an overall area under the curve of 0.935 and classification accuracy of 0.848, significantly outperforming conventional convolutional neural networks and Vision Transformer baseline models. Group-specific area under the curve values were 0.942 for CVI-related conditions, 0.929 for inflammatory dermatoses, and 0.934 for vasculitis. Misclassifications were predominantly associated with overlapping dermoscopic features, such as purpuric patterns shared between stasis dermatitis and early-stage vasculitis. Notably, misclassification errors were directed more often toward the lower-risk pathway, with three CVI-related lesions misread as vasculitis, compared with only one vasculitis case misread as a CVI-related lesion.
CONCLUSIONS: This study demonstrates that a Swin Transformer-based deep learning model can effectively differentiate CVI-related skin lesions from inflammatory and vascular conditions, providing objective, noninvasive diagnostic support with the potential to streamline clinical workflows, facilitate early referrals to specialized care, and reduce reliance on invasive diagnostic procedures in the management of lower-extremity skin diseases.
METHODS: We retrospectively analyzed 677 high-resolution dermoscopic images from 248 patients, all histopathologically confirmed by skin biopsy. The dataset was categorized into three clinically distinct groups: (1) skin conditions related to CVI (eg, stasis dermatitis), (2) common inflammatory dermatological diseases, and (3) vasculitis. A Swin Transformer-based architecture (Microsoft Research)-a deep learning model that processes images at multiple spatial scales to capture both fine local detail and broader contextual patterns-was implemented to analyze dermoscopic images. A strict patient-level split was used to ensure model robustness, so that images from the same patient were never shared between training and test sets, which could otherwise artificially inflate performance estimates.
RESULTS: The proposed Swin Transformer model demonstrated superior diagnostic performance, achieving an overall area under the curve of 0.935 and classification accuracy of 0.848, significantly outperforming conventional convolutional neural networks and Vision Transformer baseline models. Group-specific area under the curve values were 0.942 for CVI-related conditions, 0.929 for inflammatory dermatoses, and 0.934 for vasculitis. Misclassifications were predominantly associated with overlapping dermoscopic features, such as purpuric patterns shared between stasis dermatitis and early-stage vasculitis. Notably, misclassification errors were directed more often toward the lower-risk pathway, with three CVI-related lesions misread as vasculitis, compared with only one vasculitis case misread as a CVI-related lesion.
CONCLUSIONS: This study demonstrates that a Swin Transformer-based deep learning model can effectively differentiate CVI-related skin lesions from inflammatory and vascular conditions, providing objective, noninvasive diagnostic support with the potential to streamline clinical workflows, facilitate early referrals to specialized care, and reduce reliance on invasive diagnostic procedures in the management of lower-extremity skin diseases.