[공지사항]을 빙자한 안부와 근황 
Show more

Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

Multi-Centre Validation of a Deep Learning Model for Scoliosis Assessment

Created by
  • Haebom

Author

\V{S}imon Kubov, Simon Kl i\v{c}n ik, Jakub Dand ar, Zden\v{e}k Straka, Karol ina Kvakov a, Daniel Kvak

Outline

This study evaluated the accuracy of automated deep learning software (Carebot AI Bones) for Cobb angle measurement, which is essential for diagnosing scoliosis. The results of measurement by two musculoskeletal radiologists and the measurement results of AI software were compared and analyzed for 103 anteroposterior spine radiographs collected from 10 hospitals. The performance of AI was evaluated using Bland-Altman analysis, mean absolute error (MAE), root mean square error (RMSE), Pearson correlation coefficient, and Cohen kappa coefficient, and it was reported that AI showed specialist-level performance in Cobb angle measurement and grading.

Takeaways, Limitations

Takeaways:
We demonstrate that the Cobb angle of scoliosis can be accurately measured using deep learning-based automated software.
It can contribute to solving the problems of time consumption and differences between measurers in the existing manual measurement method.
It may help increase the efficiency of the scoliosis diagnosis and classification process and improve patient management.
Limitations:
The size of the dataset used in the study was relatively small (103 cases).
Further research is needed to determine generalizability across different types of scoliosis and radiographic quality.
Larger prospective studies are needed to fully validate the clinical utility of AI software.
Although the judgment of two radiologists was used as the standard, the agreement between the judgments of the experts may not be perfect (inter-reader kappa 0.59). The MAE between the AI and the experts was 3.89–3.90, which may be within the clinically meaningful error range.
👍