Combining Deep Learning Techniques and Image Analysis Methods for Vertebrae, Disc, and Spinal Stenosis...

Area of Science:

• * Medical Imaging Analysis
• * Artificial Intelligence in Healthcare
• * Computational Anatomy

Background:

• * Artificial neural networks (ANN) offer advanced decision support in medicine.
• * Optimizing time and resource management is crucial for medical diagnostics.
• * Accurate segmentation of spinal structures is key for diagnosing low back diseases.

Purpose of the Study:

• * To develop an ANN-aided decision support method for classifying and quantifying low back disease from MRI.
• * To accurately segment vertebrae and intervertebral discs for improved spinal stenosis assessment.
• * To evaluate the performance of U-Net models in segmenting lumbar spine structures.

Main Methods:

• * Utilized a dataset of 1960 T2-weighted lumbar spine MRI slices from 200 patients.
• * Employed U-Net architectures for the segmentation of vertebrae and intervertebral discs.
• * Trained and evaluated neural networks on the prepared MRI dataset.

Main Results:

• * U-Net models achieved a mean Dice similarity coefficient of 0.79 for vertebrae segmentation.
• * Achieved a mean Dice similarity coefficient of 0.76 for intervertebral disc segmentation.
• * Demonstrated the efficacy of the ANN approach in analyzing lumbar spine MRI data.

Conclusions:

• * A computer-aided diagnosis (CAD) methodology using U-Net for lumbar spine MRI segmentation is proposed.
• * The developed approach requires minimal user input, enhancing efficiency.
• * This method shows significant potential as a clinical support system for radiologists in diagnosing low back conditions.