Revolutionizing Crohn's disease detection: integrating AI with intestinal ultrasound for superior diagnosis

Intestinal ultrasound (IUS) is increasingly utilized by healthcare providers for the diagnosis of Crohn's disease (CD), which has highlighted new challenges related to standardized image interpretation and its limitations as a research instrument. The application of artificial intelligence methodologies may offer solutions to these challenges. This study aimed to assess the feasibility of integrating IUS with clinical features to enhance the diagnostic accuracy of CD. We compared the performance of five machine learning (ML) models to diagnose the most effective model for this purpose.

To evaluate and compare the diagnostic performance of five ML models in identifying CD using IUS features alone and in combination with clinical features; to determine the optimal ML approach for improving diagnostic accuracy in patients presenting with chronic diarrhea and to establish a framework for automated diagnosis of CD through the integration of imaging and clinical data.

Retrospective diagnostic accuracy study using ML methodology.

We conducted a retrospective analysis of clinical features and ultrasound data from a cohort of 1119 patients presenting with chronic diarrhea, categorized according to the diagnostic criteria recommended by the World Health Organization (WHO). Approximately 80% of the dataset was allocated for training purposes, while the remaining 20% was reserved for testing.

The logistic regression model that utilized ultrasound features exclusively demonstrated superior diagnostic performance compared to the other four models, achieving an accuracy of 78%, sensitivity of 85%, and specificity of 84.9%, with an Area Under the Curve (AUC) of 0.89 in the test set. This model accurately identified CD in patients presenting with chronic diarrhea. The integration of clinical features further enhanced the diagnostic performance of the XGBoost model, achieving an accuracy of 92%, sensitivity of 89%, specificity of 88.7%, and an AUC of 0.98 in the test set.

This study not only illustrates the feasibility of employing ultrasound analysis in patients with diarrhea but also highlights the potential for developing a methodology that integrates ultrasound data with clinical indicators. Our findings provide a framework for automated diagnosis of CD.