DOI 10.1007/s12020-025-04278-9
Human-machine collaborative risk assessment model for thyroid nodules based on local attention and multi-scale feature extraction: a multi-center clinical study.
Аннотация
PURPOSE: We aimed to evaluate a human-machine collaborative risk assessment model for thyroid nodules using local attention mechanisms and multi-scale feature extraction and compare its performance with those of radiologists of varying experience levels.
METHODS
A multi-center diagnostic study was conducted using ultrasound image datasets from six hospitals in China. The model was trained on 8397 images from 8063 patients (training set) and validated on 253 images from 245 patients across multiple centers. The diagnostic performance of the model was compared with those of radiologists with varying levels of experience. An assistive strategy was developed where radiologists adjusted their diagnoses based on model results.
RESULTS
The model achieved recognition accuracies of 0.966, 0.809, 0.826, 0.837, and 0.861 for composition, echogenicity, margin, echogenic foci, and orientation, respectively. The area under the receiver operating characteristic curve (AUROC) for the model in diagnosing benign and malignant nodules was 0.882, significantly higher than that of the junior radiologist (0.789; P < 0.0001). The AUROC of the model was between that of the intermediate (0.837) and senior (0.892) radiologists, with no significant difference compared to either group (both P > 0.05). The assistive strategy improved the AUROC for the junior radiologist from 0.789 to 0.859 (P < 0.0001) and increased sensitivity from 66.11% to 80.00% (P < 0.05), with specificity unchanged.
CONCLUSION
The model accurately identified thyroid nodule risk features and enhanced diagnostic performance, particularly for the junior radiologist, improving sensitivity in diagnosing thyroid nodules.
Предметные рубрики MeSH
Подготовка и перевод публикации
Обработка статьи · этапы, ошибки и управление
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.
Тип публикации указан по данным источника. Уровень доказательности автоматически не присваивается.