Clinical imaging · 2026
Automated detection of superior mesenteric artery occlusion on post-contrast CT Using a 3D deep learning model.
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.
Harris Robert J, Baginski Scott G, Bronstein Yulia, Schultze Dietrich, Lozano Kay D, Mehta Parthiv N, Rethy Michael C, Sparrman Kohl L, Kim Shwan
Аннотация
PURPOSE: To develop and evaluate a 3D deep learning model for detecting superior mesenteric artery occlusion (SMAO) on post-contrast abdominal CT examinations and to assess its performance and clinical impact in a prospective setting.
METHODS: A natural language processing (NLP) model was used to identify reports positive for SMAO, which were manually annotated to create training data. A 3D convolutional neural network was trained to localize SMAO and was tested on cases with previously missed findings. The model was prospectively deployed over a 6-week period to analyze 79,163 post-contrast CT examinations. Sensitivity, specificity, area under the curve (AUC), delay times, and quality assurance (QA) metrics were measured.
RESULTS: On the test dataset, a classification threshold of 0.90 yielded 50.0% sensitivity and 100.0% specificity. Prospectively, sensitivity was 67.6% and specificity was 99.6% (AUC = 0.917). The model flagged 237 cases for QA review, of which 83 (35.0%) were confirmed as missed SMAO, indicating that 40.7% of SMAO cases were undiagnosed on the initial read. SMAO incidence was 0.26%. Median delay time was shorter for cases with positive image model results (5.1 min [IQR 2.9-9.6] vs 27.9 min [IQR 8.7-52.7]; p < .001). QA-detected SMAO occurred more often on non-CTA exams (2.4% vs 46.3%; p < .001) and never mentioned SMAO or mesenteric ischemia in their clinical indication (0.0% vs 17.4%; p < .001).
CONCLUSION: A 3D deep learning model accurately detected SMAO on post-contrast abdominal CT with high specificity, reduced reporting delays, and identified clinically important missed findings.
METHODS: A natural language processing (NLP) model was used to identify reports positive for SMAO, which were manually annotated to create training data. A 3D convolutional neural network was trained to localize SMAO and was tested on cases with previously missed findings. The model was prospectively deployed over a 6-week period to analyze 79,163 post-contrast CT examinations. Sensitivity, specificity, area under the curve (AUC), delay times, and quality assurance (QA) metrics were measured.
RESULTS: On the test dataset, a classification threshold of 0.90 yielded 50.0% sensitivity and 100.0% specificity. Prospectively, sensitivity was 67.6% and specificity was 99.6% (AUC = 0.917). The model flagged 237 cases for QA review, of which 83 (35.0%) were confirmed as missed SMAO, indicating that 40.7% of SMAO cases were undiagnosed on the initial read. SMAO incidence was 0.26%. Median delay time was shorter for cases with positive image model results (5.1 min [IQR 2.9-9.6] vs 27.9 min [IQR 8.7-52.7]; p < .001). QA-detected SMAO occurred more often on non-CTA exams (2.4% vs 46.3%; p < .001) and never mentioned SMAO or mesenteric ischemia in their clinical indication (0.0% vs 17.4%; p < .001).
CONCLUSION: A 3D deep learning model accurately detected SMAO on post-contrast abdominal CT with high specificity, reduced reporting delays, and identified clinically important missed findings.