Predicting acute pancreatitis severity with enhanced computed tomography scans using convolutional neural networks

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Computational Pathology

Background:

• Acute pancreatitis (AP) severity assessment is crucial for patient management.
• Current methods for AP severity evaluation can be subjective and time-consuming.
• Advanced imaging techniques and machine learning offer potential for objective assessment.

Purpose of the Study:

• To evaluate the efficacy of convolutional neural network (CNN) models in predicting AP severity.
• To utilize enhanced computed tomography (CT) scans for AI-driven severity classification.
• To compare CNN performance using two distinct AP severity grading systems.

Main Methods:

• Development and training of 3D DenseNet CNN models using enhanced CT scans.
• Independent model training and validation using CT scans labeled with Computed Tomography Severity Index (CTSI) and Atlanta classification.
• Performance evaluation via confusion matrices, AUC-ROC, accuracy, precision, recall, and F1 scores.

Main Results:

• The DenseNet model achieved accuracy > 0.7 and AUC-ROC > 0.8 for both labeling methods.
• Models trained with CTSI-labeled scans showed a macro-average F1 score of 0.835 and AUC-ROC of 0.980.
• The study included 1,798 enhanced CT scans, randomly divided into training (n=1618) and testing (n=180) sets.

Conclusions:

• CNN models, specifically DenseNet, demonstrate significant feasibility in predicting AP severity from enhanced CT scans.
• AI-powered analysis of CT scans offers a reliable approach for objective AP severity assessment.
• These findings support the integration of AI tools in clinical radiology workflows for AP management.