Deep learning for automatic bowel-obstruction identification on abdominal CT.

Area of Science:

• Artificial Intelligence in Medical Imaging
• Radiology and Diagnostic Imaging
• Machine Learning for Disease Detection

Background:

• The increasing incidence of bowel obstruction (BO) presents a significant workload challenge for radiologists.
• Automated analysis of abdominal computed tomography (CT) scans can enhance diagnostic efficiency and patient care.
• Developing reliable machine learning models for BO identification is crucial for timely intervention.

Purpose of the Study:

• To develop and evaluate a machine learning model for the automated detection of suspected bowel obstruction (BO) in abdominal CT scans.
• To assess the performance of a 3D mixed convolutional neural network (CNN) for binary classification of BO.

Main Methods:

• A dataset of 1345 abdominal CT scans from patients with suspected BO was utilized, with an external dataset of 88 scans for validation.
• A preprocessing pipeline was developed, including abdominal-pelvic region localization and 3D scan cropping.
• Several neural network architectures were trained and tested, with a 3D mixed CNN achieving the best performance.

Main Results:

• The 3D mixed CNN achieved an F1 score of 0.92 and balanced accuracy of 0.86 on the internal dataset.
• On the external dataset, the model demonstrated an F1 score of 0.89 and balanced accuracy of 0.89.
• When calibrated for sensitivity, the model achieved 1.00 sensitivity and 0.84 specificity on the internal dataset.

Conclusions:

• The developed 3D mixed CNN shows significant potential for automated binary classification of BO in abdominal CT scans.
• This AI tool can automate patient selection and CT prioritization, thereby enhancing radiologist workflow.
• The model's high accuracy and sensitivity suggest its utility in speeding up the diagnosis of bowel obstruction.