Spinal Cord Image Denoising Using Dncnn Algorithm

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computational Neuroscience

Background:

• Accurate diagnosis of disc herniation (DH) is crucial and relies heavily on spinal image quality.
• Traditional denoising methods struggle with directional information, hindering early DH detection and classification of its stages.
• Detecting small DH (below 2mm) in MR images remains a significant challenge.

Purpose of the Study:

• To develop advanced denoising techniques for spinal MR images to improve DH diagnosis.
• To enhance the visualization of the nucleus pulposus region for better analysis.
• To achieve earlier and more accurate detection and classification of DH stages.

Main Methods:

• Spinal cord MR images from the SPIDER dataset were processed using a Parrot optimization tuned Denoising Convolutional Neural Network (Po-DnCNN).
• The enhanced images were then analyzed for DH detection and stage classification using a Hippopotamus optimization-Fast Hybrid Vision Transformer (Ho-FastViT).
• The proposed methods were quantitatively and qualitatively evaluated against manual Pfirrman Grade values.

Main Results:

• The Po-DnCNN and Ho-FastViT algorithms demonstrated significant improvements in image denoising and enhancement of the nucleus pulposus region.
• Accurate classification of DH stages (Degeneration, Prolapse, Extrusion, Sequestration) was achieved.
• The proposed methods showed superior performance compared to traditional denoising techniques.

Conclusions:

• The novel Po-DnCNN and Ho-FastViT algorithms offer superior performance for spinal image denoising compared to traditional methods.
• These AI-driven approaches enable earlier and more accurate detection of disc herniation (DH).
• The methods achieved high diagnostic accuracy, with Po-DnCNN reaching 98% and Ho-FastViT reaching 97%.