Automatic Detection and Classification of Modic Changes in MRI Images Using Deep Learning: Intelligent Assisted...

Area of Science:

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

Background:

• Modic changes (MCs) are common intravertebral MRI findings requiring expert interpretation.
• Current interpretation of MCs on MRI is complex and time-consuming.
• Deep learning offers potential for automating MCs detection and classification.

Purpose of the Study:

• To evaluate the performance of a deep learning model (SSD and ResNet18) for automatic detection and classification of MCs.
• To compare the model's diagnostic accuracy with human observers.
• To assess inter-observer and observer-classifier agreement for MCs diagnosis.

Main Methods:

• Retrospective analysis of 140 patients for internal dataset and 28 for external dataset.
• Developed a deep learning model combining Single Shot Multibox Detector (SSD) for localization and ResNet18 for classification.
• Evaluated model performance using accuracy, recall, precision, F1 score, and Kappa values, comparing with two physicians.

Main Results:

• Internal dataset: Accuracy 86.25%, Recall 87.77%, Precision 84.92%, F1 85.60%. Inter-observer Kappa: 0.768. Observer-classifier Kappa: 0.717.
• External dataset: Accuracy 75%, Recall 77.08%, Precision 77.80%, F1 74.97%. Inter-observer Kappa: 0.681. Observer-classifier Kappa: 0.519.
• The model demonstrated strong performance and high agreement with physician diagnoses.

Conclusions:

• Deep learning models, specifically SSD and ResNet18, show significant potential for accurate MCs detection and classification.
• The developed model achieved high agreement with experienced physicians, indicating its clinical utility.
• AI-assisted diagnosis can enhance efficiency and consistency in spine research and clinical practice.