The Risser grade is a widely used indicator of bone maturity in the management panel of Adolescent Idiopathic Scoliosis (AIS). The best predictors of curve progression are the growth potential and kinetics, which both depend on the bone’s maturity. However, bone maturity assessment from radiographs is challenging and is subject to intra and inter-observer variability. This study aims at developing an automatic, reliable and reproducible method for the assessment of the Risser’s grade using deep learning.
A convolutional neural network was trained to automatically grade conventional radiographs according to the Risser method. A total of 1830 posteroanterior radiographs of AIS patients were retrospectively collected and graded using the American Risser’s definition. Each radiograph was pre-processed and cropped to include the entire pelvis region. The network was then validated by comparing its accuracy against the inter and intra-observer variability of six trained graders from our institution.
Overall agreement between observers was fair, with a kappa coefficient of 0.60 for the experienced graders and an agreement of 74.50%. The automatic grading method obtained a kappa coefficient of 0.72, which is a substantial agreement with the ground truth, and an overall accuracy of 78.00%.
This is the first study using deep learning for automatic assessment of the Risser grade. This work may provide a new method for standardization of Risser grading, and additional insights in the assessment of bone maturity from radiographs.
| Date | 4 Jan 2019 |
|---|
| Original language | French |
|---|
| Awarding Institution | - École de technologie supérieure
|
|---|
| Supervisor | Luc Duong (Supervisor) & Sylvie Ratté (Co-supervisor) |
|---|
Kaddioui, H. (Author),
Duong (Supervisor) &
Ratté (Co-supervisor),
4 Jan 2019Student thesis: Master's thesis › Master in Engineering: Engineering