Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Comprehensive Study on Lumbar Disc Segmentation Techniques Using MRI Data

Serkan Salturk, Irem Sayin, Ibrahim Cem Balci, Taha Emre Pamukcu, Zafer Soydan, Huseyin Uvet

TL;DR

This study conducts a comprehensive comparison of multiple deep learning architectures for lumbar intervertebral disc segmentation on MRI data, including UNet variants, ResUNet/ResUNext, TransUNet, EF3 Net, and PSP Net. It employs 10-fold cross-validation and standard segmentation metrics (Pixel Accuracy, Mean IoU, Dice) to assess performance, with a post-processing filter that retains the five largest connected components to reduce noise. The results identify ResUnext as the strongest performer (Pixel Accuracy ≈ 0.949, Dice ≈ 0.843), with TransUNet and other networks also delivering solid segmentation quality; filtering generally provides modest gains, notably for Dense UNet. The findings support the suitability of transformer-augmented and EfficientNet-based encoders for clinical lumbar disc segmentation and point to future work integrating automated disease classification from segmented regions.

Abstract

Lumbar disk segmentation is essential for diagnosing and curing spinal disorders by enabling precise detection of disk boundaries in medical imaging. The advent of deep learning has resulted in the development of many segmentation methods, offering differing levels of accuracy and effectiveness. This study assesses the effectiveness of several sophisticated deep learning architectures, including ResUnext, Ef3 Net, UNet, and TransUNet, for lumbar disk segmentation, highlighting key metrics like as Pixel Accuracy, Mean Intersection over Union (Mean IoU), and Dice Coefficient. The findings indicate that ResUnext achieved the highest segmentation accuracy, with a Pixel Accuracy of 0.9492 and a Dice Coefficient of 0.8425, with TransUNet following closely after. Filtering techniques somewhat enhanced the performance of most models, particularly Dense UNet, improving stability and segmentation quality. The findings underscore the efficacy of these models in lumbar disk segmentation and highlight potential areas for improvement.

Comprehensive Study on Lumbar Disc Segmentation Techniques Using MRI Data

TL;DR

This study conducts a comprehensive comparison of multiple deep learning architectures for lumbar intervertebral disc segmentation on MRI data, including UNet variants, ResUNet/ResUNext, TransUNet, EF3 Net, and PSP Net. It employs 10-fold cross-validation and standard segmentation metrics (Pixel Accuracy, Mean IoU, Dice) to assess performance, with a post-processing filter that retains the five largest connected components to reduce noise. The results identify ResUnext as the strongest performer (Pixel Accuracy ≈ 0.949, Dice ≈ 0.843), with TransUNet and other networks also delivering solid segmentation quality; filtering generally provides modest gains, notably for Dense UNet. The findings support the suitability of transformer-augmented and EfficientNet-based encoders for clinical lumbar disc segmentation and point to future work integrating automated disease classification from segmented regions.

Abstract

Lumbar disk segmentation is essential for diagnosing and curing spinal disorders by enabling precise detection of disk boundaries in medical imaging. The advent of deep learning has resulted in the development of many segmentation methods, offering differing levels of accuracy and effectiveness. This study assesses the effectiveness of several sophisticated deep learning architectures, including ResUnext, Ef3 Net, UNet, and TransUNet, for lumbar disk segmentation, highlighting key metrics like as Pixel Accuracy, Mean Intersection over Union (Mean IoU), and Dice Coefficient. The findings indicate that ResUnext achieved the highest segmentation accuracy, with a Pixel Accuracy of 0.9492 and a Dice Coefficient of 0.8425, with TransUNet following closely after. Filtering techniques somewhat enhanced the performance of most models, particularly Dense UNet, improving stability and segmentation quality. The findings underscore the efficacy of these models in lumbar disk segmentation and highlight potential areas for improvement.

Paper Structure

This paper contains 20 sections, 2 figures, 1 table.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Method
  4. Data Collection and Labelling
  5. Methodological Framework
  6. UNet
  7. UNet++
  8. UNet BN
  9. Dense UNet
  10. Attention UNet
  11. ResUNet
  12. ResUNext
  13. Multires UNet
  14. TransUNet
  15. EF3 Net
  16. ...and 5 more sections

Figures (2)

  • Figure 1: Graphical Abstract
  • Figure 2: (a) Original image (b) Images masked by specialists with ImageJ application (c) Masked images converted to black&white