Lumbar Disc Herniation Automatic Detection in Magnetic Resonance Imaging Based on Deep Learning

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Radiology

Background:

• Lumbar disc herniation (LDH) is a primary cause of lower back pain and sciatica, with multifactorial origins.
• Magnetic resonance imaging (MRI) is crucial for visualizing spinal soft tissues and diagnosing LDH.
• Deep learning (DL) offers potential for rapid and accurate LDH detection from medical images.

Purpose of the Study:

• To develop an automated detection system for initial lumbar disc herniation (LDH) examinations.
• To investigate the efficacy of deep learning models with enhanced medical image features on small-scale datasets.
• To assess the performance of the YOLOv3 model for detecting LDH in MRI scans.

Main Methods:

• Utilized the YOLOv3 deep learning model for object detection of LDH on MRI.
• Processed and annotated a dataset of MRI images based on radiologist diagnoses.
• Employed data augmentation techniques to enhance a small-scale dataset and prevent overfitting.

Main Results:

• Achieved a highest mean average precision (mAP) of 92.4% with 550 augmented images (550-aug).
• The YOLOv3 model demonstrated 100% training accuracy with optimal average precision on the 550-aug dataset.
• Validated the feasibility of using DL with limited medical image data through data augmentation.

Conclusions:

• Data augmentation is vital for successful YOLOv3 training and detection performance.
• The proposed method shows promise for rapid initial testing and auto-detection using limited clinical datasets.
• This approach can aid in the early detection of lumbar disc herniation (LDH).