Updates in surgery · 2026
Training the diagnostic artificial intelligence in thyroid sonography: how well is deep learning truly learning?
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.
Lozan Oana, Musholt Petra B, Lederer Ann-Kathrin, Bouzakri Nabila, Lang Hauke, Musholt Thomas J
Аннотация
INTRODUCTION: The use of artificial intelligence (AI) for TI-RADS classification in neck sonography for the assessment of thyroid nodules has been proven beneficial. We evaluated the learning curve of such AI in a real-world setting.
METHODS: Between 03/2023 and 06/2024, 110 patients with 176 thyroid nodules were examined and classified according to ACR TI-RADS classification using 3D-ultrasound PIUR tUS Infinity software. After software training and update, the study was repeated examining 133 patients with 228 nodules (03/2023 until 10/2024). Every AI-based TI-RADS evaluation was compared to that of an experienced endocrine surgeon (assessor), unaware of the software results.
RESULTS: First phase: AI-supported TI-RADS classification corresponded to the assessor's in 128/176 (73%) of examined nodules; second phase: correspondence in 210/227 (92.6%). Re-evaluating the initial 110 patients after software update, more nodules (194 vs. 176) were correctly assessed. In the first phase, AI "misinterpreted" in 36/176 (20%) cases nodules with microcalcifications or echogenic foci (leading to 1-3 points differences to accessor). After update, only 11/227 (4.8%) such nodules remained and relevant differences to the accessor were found in only 6/227 (2.6%) unusual cases (autoimmune thyroiditis, non-descended thymus, hemorrhaged cyst), compared to the initial 7%.
CONCLUSIONS: After several rounds of deep learning, a significant improvement in correct ACR TI-RADS classification was demonstrated, especially the assessment of nodules within conglomerates. The AI-supported ultrasound is already a solid tool in the diagnosis of true thyroid nodules in non-inflamed tissue, but cannot yet replace an experienced clinician in complex or unusual cases; however, the fast learning curve is encouraging.
METHODS: Between 03/2023 and 06/2024, 110 patients with 176 thyroid nodules were examined and classified according to ACR TI-RADS classification using 3D-ultrasound PIUR tUS Infinity software. After software training and update, the study was repeated examining 133 patients with 228 nodules (03/2023 until 10/2024). Every AI-based TI-RADS evaluation was compared to that of an experienced endocrine surgeon (assessor), unaware of the software results.
RESULTS: First phase: AI-supported TI-RADS classification corresponded to the assessor's in 128/176 (73%) of examined nodules; second phase: correspondence in 210/227 (92.6%). Re-evaluating the initial 110 patients after software update, more nodules (194 vs. 176) were correctly assessed. In the first phase, AI "misinterpreted" in 36/176 (20%) cases nodules with microcalcifications or echogenic foci (leading to 1-3 points differences to accessor). After update, only 11/227 (4.8%) such nodules remained and relevant differences to the accessor were found in only 6/227 (2.6%) unusual cases (autoimmune thyroiditis, non-descended thymus, hemorrhaged cyst), compared to the initial 7%.
CONCLUSIONS: After several rounds of deep learning, a significant improvement in correct ACR TI-RADS classification was demonstrated, especially the assessment of nodules within conglomerates. The AI-supported ultrasound is already a solid tool in the diagnosis of true thyroid nodules in non-inflamed tissue, but cannot yet replace an experienced clinician in complex or unusual cases; however, the fast learning curve is encouraging.