Article
Evaluation of an Artificial Intelligence (AI) based scoliosis measurement program
Search Medline for
Authors
-
Miao Wang
- Aarhus University Hospital, Aarhus N, Denmark
-
Ming Sun
- Aarhus University Hospital, Aarhus N, Denmark
-
Haisheng Li
- Aarhus University Hospital, Aarhus N, Denmark
-
Peter Helmig
- Aarhus University Hospital, Aarhus N, Denmark
-
Kestutis Valancius
- Aarhus University Hospital, Aarhus N, Denmark
-
Yukang Yang
- Tsinghua University, Beijing, China
-
Tianyu Liu
- Tsinghua University, Beijing, China
-
Yu Wang
- Beijing University Hospital, Beijing, China
-
Kristian Høy
- Aarhus University Hospital, Aarhus N, Denmark
-
Cody Bünger
- Aarhus University Hospital, Aarhus N, Denmark
-
Ebbe Hansen
- Aarhus University Hospital, Aarhus N, Denmark
| Published: | October 21, 2024 |
|---|
Outline
Text
Introduction: Scoliosis refers to the three-dimensional deformity of the spine with one or several segments of the spine bent laterally with vertebral rotation. Reliable measurement of spinal curve is crucial for determining therapeutic decision for scoliosis patients. Cobb Angle is the gold standard, but it is an objective measurement with variant from surgeon to surgeon. A solid and reliable measurement tool is needed. Artificial Intelligence has showed great potential in image measurement.
Aim: To compare the performance of an AI based scoliosis measurement tool with senior scoliosis surgeons in Denmark.
Methods: Trained the AI algorithm with 650 scoliosis X-ray images by using Convolutional Neural Network (CNN).
Another 100 scoliosis X-ray have been assigned into two groups randomly. Each group has been measured by AI and two surgeons. All four surgeons measured Cobb angles twice with minimal 3 weeks interval. Intraclass correlation coefficients (ICC) were used to determine the interobserver and intraobserver reliabilities. (ICC can range from 0 to 1, where 0 means no reliability and 1 means perfect reliability, ICC between 0.9 to 1 means excellent reliability). The correlation of scoliosis curve angle measurements between AI program and senior surgeons have been tested with Pearson correlation coefficient and the mean absolute error.
Results: Coefficient was 0.956 in group 1 and 0.930 in group 2. Spearman rank-order correlation was 0.960 (p<0.001) in group 1 and 0.900 (p<0.001) in group 2. The absolute error between AI and surgeons are 3.5°±3.1° in group 1 and 5.0°±3.8° in group 2. In total the absolute error is 4.2°±3.3°(Figure 1). In 67% of all cases, there were only 0°–5° different between AI program and spine surgeons.
Conclusion: There is statistic correlation of Cobb angle measurement between our new developed AI program and senior spine surgeons. The reliability is statistic excellent in both patients’ groups. Our new AI program can provide reliable Cobb angle measurement as good as senior spine surgeons.
Acknowledgements: Central Region Denmark Research Foundation, and A.P. Moeller Foundation.
