Funding outcome

Adolescent idiopathic scoliosis (AIS) is a three-dimensional spinal deformity that affects 1% to 3% of adolescents, making it the most common form of scoliosis. The deformity comprises of a lateral curvature and rotation of the spine. There is no known cause for it. The severity of the spinal curvature is typically assessed with an angle measured on an x-ray of the patient's spine. However, taking x-rays exposes patients to radiation, which can increase their chances of developing cancer. Because of this, clinicians have looked into alternative methods of imaging, such as ultrasound, that do not produce radiation. What makes treatment decision for this deformity difficult is that patients' curves can all progress at different rates. Therefore, a common parameter that clinicians desire is a predictive factor, describing how quickly a patient's scoliosis will progress. Knowing this can reduce an adolescent's exposure to radiation and improve the effectiveness of treatment. However, determining this predictive factor has been proven to be a difficult task. While we have an idea of what factors generally contribute to curve progression, the individual impact of each factor is unknown. Due to the complex nature of this prediction, many researchers have turned to using machine learning to predict the risk of progression in AIS patients. However, the algorithms in current research are insufficient in terms of both performance and validation. The purpose of my study is to build a machine learning algorithm that helps clinicians with their treatment decision for AIS patients by predicting the risk of progression in their spinal curves. This clinical decision support system will take patient data as inputs and output the optimal treatment option for the patient. Different machine learning techniques will be investigated and tested to determine which method provides the best performance for this application. Clinical data of over 2000 patients with AIS from the local hospital group will be used to train these algorithms to predict the best treatment option for an AIS patient. This data includes demographics, x-ray parameters, and unique ultrasound parameters which have been developed by our research team. With this project, I also aim to investigate how the presence of these novel ultrasound parameters affects the performance of the system. If the performance of the system is comparable to those that use only x-ray parameters, it will further validate the use of ultrasound imaging as a viable method of scoliosis imaging, and minimize the risk of cancer in AIS patients.