5-Year progression prediction of endplate defects: Utilizing the EDPP-Flow convolutional neural network based on...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Spine Degenerative Diseases

Background:

• Lumbar disc degeneration (LDD) is a primary cause of low back pain, often diagnosed using MRI.
• MRI biomarkers such as Schmorl's node, high intensity zone (HIZ), and Modic changes are associated with LDD.
• The natural history and predictive capacity of these MRI findings for LDD progression remain unclear.

Purpose of the Study:

• To develop and validate a deep learning pipeline, EDPP-Flow, for predicting the 5-year progression of Schmorl's node, HIZ, and Modic changes.
• To leverage clinical MRI data for accurate LDD progression prediction.
• To address the challenge of predicting LDD progression with limited historical data.

Main Methods:

• Utilized a dataset of 1152 volunteers with baseline and 5-year follow-up MRI scans.
• Employed a deep learning pipeline integrating MRI-SegFlow and a convolutional neural network for endplate defect progression prediction.
• Implemented resampling and data augmentation strategies to handle an unbalanced dataset with limited positive samples.

Main Results:

• Achieved high weighted accuracy, sensitivity, and specificity for predicting the progression of Schmorl's node, HIZ, and Modic changes.
• Demonstrated robust performance on an unbalanced dataset, with prediction accuracies ranging from 87.51% to 91.75%.
• Successfully predicted LDD progression despite low prevalence of positive cases (4.3%–11.7%).

Conclusions:

• A validated deep learning pipeline (EDPP-Flow) can accurately predict the 5-year progression of key endplate defects in LDD.
• The pipeline shows significant potential for clinical application in managing low back pain.
• Deep learning models can effectively predict disease progression even with imbalanced medical imaging datasets.