JISE


  [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15]


Journal of Information Science and Engineering, Vol. 40 No. 3, pp. 495-505


Measurement of Automatic Segmented Specific Lumbar Intervertebral Discs

SUNG-TAI WEI1,2,†, CHIEN-YU LI2,†, CHIH-LUNG LIN2,4,†, YI-CHI HUANG3,†,
JUI-CHI CHEN5, ZHEN-YOU LIAN5,6 AND CHENG-HUNG CHUANG6,+
1Department of Neurosurgery
China Medical University Hospital
Taichung, 404 Taiwan
2Department of Neurosurgery
3Department of Radiology
Asia University Hospital
Taichung, 413 Taiwan
4Department of Occupational Therapy
5Department of Computer Science and Information Engineering
Asia University
Taichung, 413 Taiwan
6Department of Artificial Intelligence and Computer Engineering
National Chin-Yi University of Technology
Taichung, 411 Taiwan
E-mail: chchuang@ncut.edu.tw+

The segmentation of intervertebral discs from medical images is an important task in clinical medicine. Deep learning methods perform well for automatic segmentation of all intervertebral discs, but fail to identify them. When some certain discs are required for diagnosis, only these specific discs are needed to be segmented. However, segmentation of those specific discs often suffers from segmentation errors, misalignment, and noise. To address this issue, a two-stage segmentation method based on MultiResUNet and distance transform has been proposed to segment specific intervertebral discs. But is this method valid for height measurement of segmented discs? In this study, the disc height measure-ment in the two-stage segmentation method is compared to different segmentation models, e.g. U-Net and MultiResUNet. When evaluating specific disc height measurements, the two-stage segmentation method has lower error rates compared to other models.

Keywords: deep learning, U-Net, MultiResUNet, spine image, intervertebral disc segmentation

  Retrieve PDF document (JISE_202403_04.pdf)