Spinal Cord Image Denoising Using Dncnn Algorithm

Spinal image denoising plays a vital role in the accurate diagnosis of disc herniation (DH).

Traditional denoising algorithms perform less due Limited Directional Selectivity problem and do not adequately capture directional information in pixels. Traditional algorithms' edge representation and texture details are insufficient for the earlier detection of DH. Limited Directional Selectivity leads to inaccurate diagnosis and classification of Disc Herniation (DH) stages. The DH stages are (i) Degeneration (ii) Prolapse (iii) Extrusion and (iv) Sequestration. Moreover, detection of DH size below 2mm using MR image is the major problem.

To solve the above problem, spinal cord MR images fed to the proposed Parrot optimization tuned Denoising Convolutional Neural Network (Po- DnCNN) algorithm for perspective enhancement of nucleus pulposus region in the spinal cord, vertebrae. The perspective enhancement of Spinal cord image led to the accurate classification of stages and earlier detection of DH by using the proposed Hippopotamus optimization- Fast Hybrid Vision Transformer (Ho-FastViT) algorithm. For this study, spinal cord MR images are obtained from the Grand Challenge website - SPIDER dataset.

The proposed Po-DnCNN method and Ho-FastViT results are analysed quantitatively and qualitatively based on the edge, contrast, classification of the stage, and enhancement of the projected nucleus pulposus region in the spinal cord and vertebrae. The predicted DH results using the proposed method are compared with the manual Pfirrman Grade value of the spinal card method.

Proposed method is better than traditional methods for earlier detection of DH. Po-DnCNN and Ho-FastViat methods give high accuracy of about 98% and 97% compared to traditional methods.