Automated detection of superior mesenteric artery occlusion on post-contrast CT Using a 3D deep learning model

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Superior mesenteric artery occlusion (SMAO) is a critical condition requiring prompt diagnosis.
• Current diagnostic methods for SMAO on CT scans can be time-consuming and prone to missed findings.
• There is a need for advanced tools to improve SMAO detection accuracy and efficiency.

Purpose of the Study:

• To develop and validate a 3D deep learning model for detecting SMAO on post-contrast abdominal CT.
• To assess the model's performance, including sensitivity, specificity, and AUC.
• To evaluate the clinical impact of the model in a prospective setting, focusing on detection delays and missed diagnoses.

Main Methods:

• A natural language processing (NLP) model identified SMAO reports for training data creation.
• A 3D convolutional neural network (CNN) was trained for SMAO localization.
• The model was prospectively deployed on 79,163 CT examinations over 6 weeks, measuring performance metrics and quality assurance (QA).

Main Results:

• The model achieved 67.6% sensitivity and 99.6% specificity (AUC=0.917) prospectively.
• It flagged 237 cases for QA, identifying 83 (35.0%) as missed SMAO, indicating 40.7% of cases were initially undiagnosed.
• Positive model results significantly reduced median delay time (5.1 min vs 27.9 min; p < .001).

Conclusions:

• A 3D deep learning model effectively detects SMAO on CT scans with high accuracy.
• The model significantly reduces reporting delays and identifies clinically important missed SMAO cases.
• This AI tool demonstrates substantial potential for improving the diagnosis of SMAO in clinical practice.