Endocrine practice : official journal of the American College of Endocrinology and the American Association of Clinical Endocrinologists · 2025
Deep Learning-based Multiview Facial Identification as a Screening Tool for Cushing Syndrome.
СтатьяНазвание и аннотация: Перевод пока не подготовлен
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.
Qiang Jiaqi, Liu Hongjun, Guo Xiaoyuan, Song Changwei, Lu Lin, Long Xiao, Pan Hui, Zhao Qing, Huang Jiuzuo, Li Jianqiang, Chen Shi
Аннотация
OBJECTIVES: Cushing syndrome (CS) can significantly impair quality of life. It is often diagnosed late due to misdiagnosis as simple obesity under current strategies. Given that CS is characterized by distinctive facial features, for example, the "moon face," employing facial identification algorithms to analyze multiview facial images may aid in CS diagnosis.
METHODS: A total of 42 pairs of patients with CS and age-, sex-, and body mass index-matched controls were enrolled in the training cohort, with 13 matched pairs separately included in the external validation cohort. Multiview facial images were collected and integrated. Four deep learning models-DenseNet, ResNet, Swin, and ViT-were trained using fivefold cross-validation. These models were evaluated on the external validation cohort. Diagnostic performance metrics were calculated and compared with the diagnostic assessments of 18 endocrinologists.
RESULTS: Four facial recognition models were established. Within the training cohort, all 4 models demonstrated stable diagnostic performance across the 5 cross-validation folds. When applied to the external validation cohort, the models achieved an accuracy of 92.3% to 100%, sensitivity of 84.6% to 100%, specificity of 100%, and area under the curve of 94.4% to 100%. The deep learning models outperformed endocrinologists in accuracy and demonstrated sensitivity and specificity that equaled or exceeded those of the endocrinologists.
CONCLUSIONS: These findings indicate that multiview facial identification using deep learning could provide an accurate and accessible diagnostic method for CS. Given that the Artificial intelligence models outperformed endocrinologists, this approach holds potential as a novel diagnostic tool to assist clinicians in identifying various types of CS.
METHODS: A total of 42 pairs of patients with CS and age-, sex-, and body mass index-matched controls were enrolled in the training cohort, with 13 matched pairs separately included in the external validation cohort. Multiview facial images were collected and integrated. Four deep learning models-DenseNet, ResNet, Swin, and ViT-were trained using fivefold cross-validation. These models were evaluated on the external validation cohort. Diagnostic performance metrics were calculated and compared with the diagnostic assessments of 18 endocrinologists.
RESULTS: Four facial recognition models were established. Within the training cohort, all 4 models demonstrated stable diagnostic performance across the 5 cross-validation folds. When applied to the external validation cohort, the models achieved an accuracy of 92.3% to 100%, sensitivity of 84.6% to 100%, specificity of 100%, and area under the curve of 94.4% to 100%. The deep learning models outperformed endocrinologists in accuracy and demonstrated sensitivity and specificity that equaled or exceeded those of the endocrinologists.
CONCLUSIONS: These findings indicate that multiview facial identification using deep learning could provide an accurate and accessible diagnostic method for CS. Given that the Artificial intelligence models outperformed endocrinologists, this approach holds potential as a novel diagnostic tool to assist clinicians in identifying various types of CS.