Bridging the Diagnostic Divide: Classical Computer Vision and Advanced AI methods for distinguishing ITB and CD through CTE Scans

Automation improves diagnostic consistency: The study developed a computer vision algorithm to automatically calculate the visceral-to-subcutaneous fat ratio from CT scans, replacing manual measurements by radiologists. This automation aims to reduce variability and enhance objectivity in distinguishing between Intestinal Tuberculosis and Crohn's Disease, two conditions that are clinically difficult to tell apart.

Multiple approaches were compared for reliability: The researchers benchmarked their custom 2D algorithm against TotalSegmentator (a deep learning tool) and manual radiologist calculations. This comparative approach helps establish which method is most trustworthy for clinical use, an important consideration for responsible AI deployment in healthcare.

AI model achieved 75% accuracy with interpretability focus: A ResNet10 deep learning model trained on CT scans achieved 75% diagnostic accuracy. Notably, the authors emphasized integrating explainability features into their model—a responsible AI practice that helps clinicians understand why the AI made a particular diagnosis rather than treating it as a "black box."

Combines classical and modern AI methods: Rather than relying solely on deep learning, the study integrated traditional computer vision techniques with advanced AI and radiological expertise. This hybrid approach suggests that responsible AI in healthcare may benefit from combining multiple methodologies rather than replacing human expertise entirely.

Clinical validation through expert consensus: The research built upon radiologist consensus about the fat ratio biomarker, grounding the AI development in established clinical knowledge rather than purely algorithmic optimization.