Model more accurate for predicting adolescent idiopathic scoliosis than physicians

Model more accurate for predicting adolescent idiopathic scoliosis than physicians | Image Credit: © jeffy1139 - © jeffy1139 - stock.adobe.com.

According to a recent study published in JAMA Network Open, a model for predicting adolescent idiopathic scoliosis (AIS) could help clinicians with planning and treatment.

AIS, presenting as 3-dimensional spine malformation, is seen in up to 2.2% of boys and 4.8% of girls. It has high rates of progression during puberty, making early detection, close follow-up, and proper interventions vital. However, school screening programs risk disruptions such as those experienced during the COVID-19 pandemic.

Along with the risk of disruptions, AIS detection and follow-up involves clinical expertise, with assessments of external appearance not always correctly detecting malformation severity and type. AIS progression is also monitored through radiographic examinations, increasing radioexposure.

Smartphone images of patients could be used for AIS detection but have limited variability and are difficult for widespread use because of different severity and curve types in patients. To overcome these issues, investigators developed a platform called AlignProCARE for out-of-hospital spine malformation evaluation.

A diagnostic study was conducted to determine the reliability of the AlignProCARE model. Participants had images of their backs taken by a parent or guardian using a smartphone. The app provided a built-in protocol for obtaining photos, allowing it to be used with minimal training.

Exclusion criteria included having systematic neural disorders which may influence mobility, diagnosis or signs of psychological disorders which would impact adherence, oncological diseases, severe skin disorders or lesions on the back, any other systematic disease, and a body mass index over 30. Patients unable to complete the consent process were also excluded.

Two spine specialists with over 20 years of AIS management experience provided ground truth labels of all data. AIS severity was defined by degrees, with no or mild AIS being a Cobb angle of 20° or less, moderate AIS a Cobb angle of 20° to 40°, and severe AIS a Cobb angle over 40°. Treatment planning recommendations were based on AIS severity

There were 1780 participants recruited into population cohort 1 between October 2018 and September 2020, while 378 were recruited into a prospective testing cohort from October 2020 to March 2022. Of participants 20.2% required no intervention, 57.9% required nonsurgical intervention with regular follow-ups, and 11.9% were under consideration for surgery.

No or mild AIS was found in 31.2% of cohort 1, moderate AIS in 59.3%, and severe AIS in 9.6%. In cohort 2, these rates were 25.7%, 51.6%, and 22.7% respectively. Receiver operating characteristic (ROC) curves were generated, with the model predicting an area under the ROC curve (AUC) of 0.839 for no or mild AIS and 0.902 for severe AIS.

AUC curves generated from the model indicated recognition of AIS severities as well or better than surgeons’ estimates. Investigators concluded this model could contribute to planning and treatment for patients at risk of AIS.

Reference

Zhang T, Zhu C, Zhao Y, et al. Deep learning model to classify and monitoridiopathic scoliosis in adolescents using a single smartphone photograph. JAMA Netw Open. 2023;6(8):e2330617. doi:10.1001/jamanetworkopen.2023.30617