Radiologist-AI Collaboration for Ischemia Diagnosis in Small-Bowel Obstruction: Multicentric Development and External...

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Gastroenterology

Background:

• Small-bowel obstruction (SBO) is a common surgical emergency.
• Detecting ischemia, a critical complication of SBO, is challenging and requires timely intervention.
• Current diagnostic methods have limitations in accurately and rapidly identifying SBO-related ischemia.

Purpose of the Study:

• To develop and validate a multimodal artificial intelligence (AI) model for detecting ischemia in patients with SBO.
• To assess the performance of AI models integrating 3D CT data, laboratory markers, and clinical text.
• To evaluate the impact of AI assistance on radiologist performance in diagnosing SBO ischemia.

Main Methods:

• A multimodal AI model was developed using 3D CT scans, C-reactive protein, neutrophil count, and radiology report text from 771 SBO cases.
• Models were trained as unimodal and multimodal variants (image-plus-laboratory, image-plus-text, full multimodal).
• External validation was performed on 66 independent cases, and AI assistance was tested with four radiologists of varying experience levels.

Main Results:

• The image-plus-laboratory multimodal model achieved the best external validation performance (AUC 0.69, sensitivity 0.89, specificity 0.44).
• Adding report text did not improve external validation performance.
• AI assistance showed consistent, though not statistically significant, improvements in radiologist diagnostic accuracy (AUC increase).

Conclusions:

• A multimodal AI model integrating CT imaging and laboratory data is effective for detecting 24-hour ischemia in SBO.
• This AI model shows potential as a triage-support tool to assist radiologists.
• Further research may refine multimodal AI for improved SBO ischemia detection and patient outcomes.