Frontiers in medicine · 2026
Development and clinical validation of an artificial intelligence based model for thyroid nodule malignancy risk assessment using C-TIRADS guidelines.
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.
Ye Rongzhou, Liu Yao, Wu Xiuming, Wang Kangjian
Аннотация
BACKGROUND: Thyroid nodules refer to discrete lesions within the thyroid gland, caused by abnormal proliferation of thyroid cells and local growth. Thyroid ultrasound imaging is a non-invasive, widely used technique primarily employed to assess the benign or malignant nature of nodules, the extent of invasion into adjacent tissues, and lymph node metastasis. Early ultrasound diagnosis can help reduce the incidence of thyroid cancer. However, the diagnostic results from ultrasound examinations are often subjective, labor-intensive, and highly dependent on the clinical experience of the ultrasound physician.
METHODS: We developed a C-TIRADS-guided computer-aided diagnostic framework consisting of a nodule detection module, a C-TIRADS feature classification module, and a rule-based risk scoring module. Thyroid nodules were first localized using the detection network and then classified according to key C-TIRADS ultrasound features, including composition, echogenicity, margin, shape, and echogenic foci. The predicted feature scores were aggregated to determine the benign or malignant risk category. The model was trained using task-specific datasets and clinically validated on an independent cohort of 303 thyroid nodules/images from 284 patients.
RESULTS: In the independent clinical validation set, the AI model achieved an overall accuracy of 0.862 [95% CI, 0.822-0.901]. Physician accuracy was 0.705 [95% CI, 0.657-0.756] without AI assistance and increased to 0.845 [95% CI, 0.802-0.884] with AI assistance. These findings suggest the potential value of the proposed system as a C-TIRADS-guided decision-support tool.
CONCLUSION: The proposed C-TIRADS-guided framework can localize thyroid nodules, classify guideline-defined ultrasound features, and provide standardized malignancy risk stratification. The model may assist physicians in thyroid ultrasound interpretation; however, larger blinded multicenter studies are required before routine clinical application.
METHODS: We developed a C-TIRADS-guided computer-aided diagnostic framework consisting of a nodule detection module, a C-TIRADS feature classification module, and a rule-based risk scoring module. Thyroid nodules were first localized using the detection network and then classified according to key C-TIRADS ultrasound features, including composition, echogenicity, margin, shape, and echogenic foci. The predicted feature scores were aggregated to determine the benign or malignant risk category. The model was trained using task-specific datasets and clinically validated on an independent cohort of 303 thyroid nodules/images from 284 patients.
RESULTS: In the independent clinical validation set, the AI model achieved an overall accuracy of 0.862 [95% CI, 0.822-0.901]. Physician accuracy was 0.705 [95% CI, 0.657-0.756] without AI assistance and increased to 0.845 [95% CI, 0.802-0.884] with AI assistance. These findings suggest the potential value of the proposed system as a C-TIRADS-guided decision-support tool.
CONCLUSION: The proposed C-TIRADS-guided framework can localize thyroid nodules, classify guideline-defined ultrasound features, and provide standardized malignancy risk stratification. The model may assist physicians in thyroid ultrasound interpretation; however, larger blinded multicenter studies are required before routine clinical application.