Machine Learning Based Prediction Model for Closed-Loop Small Bowel Obstruction Using Computed Tomography and...

Area of Science:

• Radiology and Medical Imaging
• Surgical Gastroenterology
• Machine Learning in Healthcare

Background:

• Accurate diagnosis of closed-loop small bowel obstruction (CLSO) is crucial for timely surgical intervention.
• Traditional diagnostic methods often have limitations in sensitivity and specificity.
• Integrating diverse data sources may enhance diagnostic accuracy.

Purpose of the Study:

• To develop and validate a predictive model for CLSO using computed tomography (CT) and clinical findings.
• To identify key imaging and clinical features that are most predictive of CLSO.
• To improve the diagnostic performance beyond individual CT signs.

Main Methods:

• Retrospective review of radiology databases and surgical reports for patients with suspected CLSO.
• Independent review of CT scans by two observers for CLSO imaging features.
• Development and validation of a random forest prediction model combining CT and clinical data, with an 80/20 training/testing split.

Main Results:

• The study confirmed CLSO in 185 of 223 patients.
• Age over 52 was associated with a significantly higher risk of CLSO.
• The random forest model achieved an area under the receiver operating curve of 0.73, with a sensitivity of 0.72 and specificity of 0.8.
• Important predictors included prior surgery, age, lactate, whirl sign, U/C-shaped bowel, and fecalization.

Conclusions:

• A random forest model integrating clinical and imaging factors effectively predicts CLSO.
• Clinical factors (prior surgery, age, lactate) and imaging signs (whirl sign, fecalization, U/C-shaped bowel) are valuable predictors.
• Accurate CLSO diagnosis necessitates a systematic assessment of multiple CT signs rather than relying on individual findings.