Automated Characterization of Abdominal MRI Exams Using Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Magnetic resonance imaging (MRI) data is rapidly increasing in volume and complexity.
• Heterogeneity in imaging protocols, scanner technology, and data labeling across institutions poses challenges for research.
• Standardized MRI data identification, characterization, and labeling are essential for developing robust machine learning models.

Purpose of the Study:

• To develop a standardized methodology for classifying abdominal MRI sequences, orientation, and contrast.
• To leverage convolutional neural networks (CNNs) for automated MRI data characterization.
• To enhance the usability of diverse MRI datasets for machine learning applications.

Main Methods:

• Three distinct CNN models with similar backbone architectures were developed.
• Models were trained to classify single abdominal MRI slices.
• Classification tasks included 12 MRI sequences, 4 orientations, and 2 contrast types.

Main Results:

• The CNN models demonstrated high performance in classifying MRI data.
• Accuracies achieved were 96.9% for sequence, 97.4% for orientation, and 97.3% for contrast.
• The automated classification method proved effective for abdominal MRI data.

Conclusions:

• Automated classification of abdominal MRI sequences, orientation, and contrast using CNNs is feasible and accurate.
• This methodology can facilitate the standardization and utilization of diverse MRI datasets.
• The developed models support the advancement of machine learning in medical imaging research.