Chronic liver disease detection using deep convolutional neural networks with MRI data: a deep learning approach

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Hepatology

Background:

• Chronic liver disease (CLD) poses a significant global health challenge.
• Early and accurate detection of CLD is critical for effective treatment and patient outcomes.
• Magnetic Resonance Imaging (MRI) is a key modality for liver assessment, but diagnostic accuracy can be improved.

Purpose of the Study:

• To develop and evaluate a deep learning-based convolutional neural network (DeepCNN) for automated CLD detection using MRI.
• To compare the diagnostic performance of the DeepCNN model against traditional machine learning algorithms.
• To assess the utility of non-segmented MRI images for CLD classification.

Main Methods:

• A retrospective study utilized 1112 MRI images from 184 patients (460 normal, 652 CLD) acquired between 2018-2024.
• Images underwent preprocessing including resizing, augmentation, and normalization.
• The DeepCNN model was trained and its performance compared with logistic regression, k-nearest neighbor, support vector machines, and random forest using accuracy, precision, recall, and F1-score.

Main Results:

• The DeepCNN model achieved 93% accuracy and an F1-score of 0.939, with 97% precision and 98% recall for CLD classification.
• Traditional machine learning algorithms showed lower accuracies, ranging from 72.31% to 83.16%.
• Excluding coronal views reduced DeepCNN accuracy to 86%, highlighting the importance of multi-planar imaging.

Conclusions:

• The developed DeepCNN model demonstrates superior diagnostic accuracy for CLD detection from MRI compared to traditional machine learning methods.
• This non-segmented MRI approach offers a practical advancement for CLD diagnosis, supporting earlier intervention.
• Future research could explore attention mechanisms and advanced deep learning architectures for further performance enhancement.