Automated Rigo Classification: An Innovative Artificial Intelligence System.

Authors

Abstract

Although there is variation in clinicians' application of the Rigo classification system (RCS) to adolescent idiopathic scoliosis, no artificial intelligence (AI) models have been developed to standardize the RCS specifically. This study aimed to develop an AI model that applies the RCS to X-rays, and compare its parameter measurements, processing time, and classification accuracy to human observers. Using an open-source AI model from Scoliosis Tools, an AI model was developed to apply the RCS. Twenty non-surgical, non-in-brace X-rays of AIS patients were chosen from a SpineWeb public dataset. The RCS AI model measured parameters and classified each of these PA-view X-rays, while being timed. The X-rays were also independently classified by three human observers, while being timed. Afterwards, all observers agreed upon the most accurate RCS classification for each X-ray. Then, RCS parameters were measured by two observers for each X-ray. A descriptive analysis examined timing, parameters, and accuracy. P<0.05 was considered significant. The average time to output each RCS classification was 14.6 ± 5.4 seconds for the AI model, and 37.1 ± 10.7 seconds for human observers (P< 0.0001). When RCS measurements done by the AI model were compared to human observers, there were no significant differences (P>0.05). The AI model had 60.0% accuracy, compared to the average human observer accuracy of 75.0%. There were no significant differences between the parameter measurements of the AI model and human observers. The AI model can serve as a high-speed tool to help clinicians standardize RCS classification using its parameters. Future iterations will play an important role in enhancing efficiency of treatment plans, including physiotherapeutic scoliosis specific exercise and bracing design.

Tags

Topics