The Development and External Testing of Deep Learning Model for Automated Classification of Intervertebral Disc...

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Healthcare
• Spine Surgery and Diagnostics

Background:

• Intervertebral disc degeneration is a common cause of spinal pain.
• Accurate classification of disc degeneration is crucial for effective treatment.
• Current diagnostic methods can be time-consuming and subjective.

Purpose of the Study:

• To develop and externally validate a deep learning model for automated classification of intervertebral disc degeneration using CT images.
• To assess the model's performance in identifying lesion type, zone, and grade.

Main Methods:

• A retrospective study included patients aged ≥18 years with diagnosed disc herniation or bulge.
• A 3D U-Net model was used for disc lesion segmentation.
• The EfficientNet-B4 architecture classified lesion type, zone, and grade based on key points.
• Model performance was evaluated using Dice score, accuracy, recall, and precision on internal and external datasets.

Main Results:

• The deep learning model achieved a mean Dice score of 0.735 for segmentation in external testing.
• External testing showed high performance metrics: 0.938 accuracy, 0.938 recall, and 0.944 precision for lesion type.
• External testing also demonstrated 0.919 accuracy, 0.986 recall, and 0.976 precision for lesion zone.
• The model achieved 0.979 for grade metrics in external testing.

Conclusions:

• The developed deep learning model demonstrates good performance in automatically classifying intervertebral disc degeneration on CT images.
• The model's ability to classify by type, zone, and grading shows potential for improving diagnostic efficiency and accuracy.
• This AI tool could assist radiologists and clinicians in managing spinal conditions related to disc degeneration.