A CNN Autoencoder for Learning Latent Disc Geometry from Segmented Lumbar Spine MRI.

Area of Science:

• Biomedical Imaging
• Machine Learning in Healthcare
• Spine Biomechanics

Background:

• Low back pain is a leading cause of disability globally.
• Lumbar intervertebral disc pathology is a frequent pain driver.
• Disc geometry provides insights into mechanical behavior and pathology.

Purpose of the Study:

• Develop a convolutional neural network (CNN) autoencoder for latent feature extraction from segmented lumbar disc MRI.
• Interpret these latent features to identify disc pathology.
• Complement standard geometric measures for enhanced diagnostic capabilities.

Main Methods:

• Utilized 195 sagittal T1-weighted lumbar spine MRIs from a public dataset.
• Implemented a pipeline involving MRI segmentation, CNN autoencoder training, and latent feature extraction.
• Measured standard geometric features and predicted disc narrowing using both latent and standard features.

Main Results:

• Achieved high segmentation accuracy (IoU 0.82, DSC 0.90).
• CNN autoencoder converged with a 4x1 bottleneck size, yielding high IoU (0.9984).
• Combined latent and geometric features improved disc narrowing prediction accuracy.

Conclusions:

• A CNN autoencoder effectively extracts interpretable latent features from lumbar disc MRI.
• These latent features enhance the prediction of disc narrowing.
• Future research will incorporate voxel intensity for compositional analysis.