- 44 views
- 37 downloads
Automated Rod Length Measurements on Radiographs and Sonograms in Children with Early Onset Scoliosis
-
- Author / Creator
- Kabir, Mohammad Humayun
-
Early Onset Scoliosis (EOS) is a medical condition that is defined as a lateral curvature of the spine with vertebral rotation in children under age 10. Approximately 2-3% of children worldwide have scoliosis. Surgical intervention is the most effective management to treat these children who have severe scoliosis. Currently, the magnetically controlled growing rods (MCGR) surgery which can gradually extend externally to straighten the curve is the most cost-effective method to treat EOS. Radiographs are taken at every visit twice, once before the rod length adjustment and once after, to measure manually the rod length change.
The existing manual measurement techniques face various obstacles, including inconsistencies between different raters, poor quality of radiographs, patients’ motion, and their different postures during radiography taken, and disparities in image resolution. In addition, taking radiography exposes children to harmful ionizing radiation. Therefore, the ultrasound (US) imaging method has also been reported in the literature to image the MCGR for measuring the change of the growing rods.
The objective of this work was to develop machine learning (ML) algorithms to automatically measure the rod length adjustment on both radiographs and US images at each visit. Clinical data were collected, and studies were conducted to validate both algorithms for radiographs and sonograms.
Driven by the imperative for accuracy, this study utilized ML algorithms to construct an autonomous system. Both radiography and US automated systems were developed using Mask Regional Convolution Network (Mask RCNN) techniques, which were widely recognized and commonly used in the field of computer vision for object detection and instance segmentation. The application of Mask RCNN was implemented utilizing the Detectron2 framework.
Instead of using absolute measurements, a calibration technique was applied. Three ML models: rod model, 58mm fixed length model, and head-piece model were developed to extract the rod length from radiographs. Three-hundred and eighty-seven radiographs were used for model development, and 60 radiographs with 118 rods were used for testing. The radiography automated system eventually demonstrated an acceptable inter-method correlation coefficient of 0.90 and a mean absolute difference (MAD) of 0.98 ± 0.88 mm when compared to manual adjustment measurements.
In the US imaging system, similar ML algorithms were developed. Two ML models: the Boundary model and the Rod model, were developed. A study that included 90 US images acquired from 23 EOS patients was conducted. Among the 90 images, 70 images were used for model development, and 20 images were used for testing. The MAD between the Artificial Intelligence (AI) measurements versus the manual measurements was 1.2 ± 1.46 mm and the reliability of the inter-method correlation coefficient was 0.96.
In summary, this thesis reported a new and automated approach to measuring rod adjustments in children who have MCGR installation. The automated approaches were accurate to measure the adjustments in a faster way, saving time of clinicians, while reducing the chance of error made by the traditional manual approach. -
- Graduation date
- Fall 2024
-
- Type of Item
- Thesis
-
- Degree
- Master of Science
-
- License
- This thesis is made available by the University of Alberta Library with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.