Related Experiment Video
Updated: Jun 30, 2025
Author Spotlight: Bridging Gaps in Anatomy and Establishing a Foundation for Algorithmic Studies
Published on: December 15, 2023
Fast and accurate 3-D spine MRI segmentation using FastCleverSeg
Jonathan S Ramos1, Mirela T Cazzolato2, Oscar C Linares2
1Computer Science Department, Federal University of Rondônia (DACC/UNIR), 364 BR, 76801-059, Rondônia, Brazil; Institute of Mathematics and Computer Sciences, University of Sao Paulo (ICMC/USP), 400 Trabalhador Saocarlense Avenue, 13566-590 São Carlos, São Paulo, Brazil.
FastCleverSeg offers efficient semi-automatic segmentation for spine imaging, reducing user effort and time. This method achieves high accuracy for vertebral bodies, muscles, and discs, improving spinal disease analysis.
Area of Science:
- Medical Imaging
- Computer Vision
- Biomedical Engineering
Background:
- Accurate segmentation of spinal structures (vertebrae, muscles, discs) is vital for diagnosing spinal diseases.
- Traditional segmentation methods are either manual (laborious) or fully automatic (require extensive data).
Purpose of the Study:
- To introduce FastCleverSeg, a semi-automatic segmentation approach that minimizes user interaction while maintaining high accuracy.
- To enable efficient and precise volumetric segmentation of spinal MRI data.
Main Methods:
- Reduced user interaction by requiring manual annotation of only 2-3 slices.
- Automatic estimation of annotations on intermediary slices (EANIS) using computer vision techniques.
- Improved voxel weight balancing for fast and precise volumetric segmentation.
Main Results:
- Demonstrated a processing time of 25 ms (30 ms SD) on a diverse MRI database (179 patients).
- Achieved a significant reduction in user interaction compared to existing methods.
- Maintained or surpassed competing methods' segmentation quality, reaching a 94% Dice score.
Conclusions:
- FastCleverSeg empowers physicians to efficiently generate reliable ground truths for spinal segmentation.
- The method expedites the segmentation process and shows potential for future integration with deep learning for fully automated analysis.
