Automating Post-Processing of Lumbar Disc MRI Segmentation

• Hanaa Abdallah, Mechanical Engineering, University of Delaware

• Dawn Elliott, Biomedical Engineering, University of Delaware

Low back pain is a prevalent problem, to better understand its cause and develop interventions, we need to assess the intervertebral disc mechanics. This study included 86 subjects spread evenly across sex and age groups (18-85 years old), with no history of back pain to establish a baseline for healthy disc mechanics. Using magnetic resonance imaging (MRI) we can analyze the lumbar spines’ health and function. The discs must be segmented from the scans to enable mechanical analysis: including 2D and 3D disc geometry. Many programs allow for manual segmentation but it is time-consuming (about 2 hours per spine) and there is potential for human error. To improve analysis efficiency, an auto-segmenting code was developed using Python and is run through Google Collab. After running the images through the auto segmenting code, they are not immediately in a usable format. The files require post-processing before they can be aligned with the original images and fully analyzed. I developed a post-processing code in Matlab that restores original image size, converts the file to binary, and complements an existing code. This post-process makes the segmentation compatible with the original image for further analysis. Automating this process reduces potential human error and severely decreases the required time (about 5 minutes for one spine). Lastly, the output images undergo minor manual checking and corrections before entering the analysis pipeline. The total segmentation process has drastically decreased time, allowing for efficient completion of segmentations for the remaining subjects. Future analysis will investigate trends in the disc mechanical behavior with sex.