Application of machine learning algorithms for multiparametric MRI-based evaluation of murine colitis

Area of Science:

• Medical Imaging
• Gastroenterology
• Machine Learning in Medicine

Background:

• Inflammatory bowel disease (IBD) diagnosis relies on assessing gut pathology.
• Magnetic resonance imaging (MRI) offers non-invasive evaluation of IBD, providing morphological and functional data.
• Existing MRI techniques capture relaxation times and diffusion capacity of bowel segments.

Purpose of the Study:

• To evaluate multiparametric MRI for diagnosing experimentally induced colitis in mice.
• To assess the diagnostic benefit of combining MRI parameters using machine learning.
• To compare the accuracy of combined parameters against individual MRI metrics.

Main Methods:

• Colitis was induced in mice using Dextran Sodium Sulfate (DSS).
• In vivo and ex vivo MRI of mouse colons were performed.
• Machine learning (Model Averaged Neural Network - avNNet) was applied to multiparametric MRI data.
• Diagnostic accuracy was assessed using Receiver Operating Characteristics (ROC) analysis, calculating sensitivity, specificity, PPV, and NPV.

Main Results:

• The avNNet model demonstrated high diagnostic accuracy in both in vivo and ex vivo settings.
• In vivo: Sensitivity 91.3%, Specificity 92.3%, PPV 96.9%, NPV 80.0%.
• Ex vivo: Sensitivity 87.4%, Specificity 82.9%, PPV 88.9%, NPV 80.8%.
• The combined multiparametric approach significantly outperformed single MRI parameters in diagnostic accuracy.

Conclusions:

• Multiparametric MRI combined with machine learning significantly enhances diagnostic accuracy for experimental colitis.
• This pilot study suggests potential for developing an MR-derived score for IBD diagnosis and surveillance.
• Further research can optimize MRI-based tools for improved inflammatory bowel disease management.