Autonomous Classification and Prediction of Progression of Scoliosis

Spinal disorders present a significant cause of pain and distress for millions of people and a significant public health burden with a very high cost to health services. Scoliosis causes a curvature and rotation of the vertebrae, producing a lateral curvature of the spine. This can have severe health implications for patients including: life-long back pain, posture issues, mental health problems and in severe cases: heart problems and respiratory issues. Adolescent Idiopathic Scoliosis (AIS; Scoliosis in children and teenagers, without a known cause) has a prevalence of approximately 2-3% in the British population and globally. AIS often progresses rapidly during periods of adolescent growth, making monitoring Scoliosis progression essential. Monitoring is incredibly resource intensive. Automated monitoring of cases would decrease the workload of clinicians, allowing them to focus on complex cases, increasing the quality and cost- effectiveness of patient care. Automated classification could also be used for population level screening to identify cases. Epidemiological research into Scoliosis is also limited, partially because mild-moderate cases of the condition are often not captured in administrative health datasets. Robust automated classification of medical images would enable further quantitative epidemiological research into the causes of the disease. We have unique medical imaging data from the UK Biobank and the transgenerational Avon Longitudinal Study of Parents and Children (ALSPAC). The UK Biobank has approximately 48,000 full-body Dual Energy X-ray Absorptiometry (DXA) scans. ALSPAC has approximately: 33,000 full-body DXA scans of 9,000 children taken as they grow up with medical images at ages: 9, 13, 15, 17, 24. Allowing us to use a timeseries of images to predict Scoliosis progression in adolescents.
Our aims are to: 1) develop artificial intelligence tools to: classify the presence of Scoliosis in medical images for research use and clinical use; 2) predict the progression of Scoliosis cases. 3) We also aim to use these tools to assist with and conduct epidemiological research into the causes of Adolescent Idiopathic and Degenerative Scoliosis.
We will develop novel methods in machine learning (ML) and Artificial Intelligence (AI) to classify the presence of Scoliosis in medical images and predict the likelihood of Scoliosis progressing. To achieve this, we will use state of the art methods from computer vision and other disciplines within ML and AI. We will apply industry leading Neural Network architectures to the problem of medical image classification (to autonomously diagnose Scoliosis) and will also modify, and develop, novel architectures and configurations for this purpose. Further, we will develop novel methods in timeseries analysis for use in medical imaging (to predict the likelihood of Scoliosis progression). To summarize: the novel aspects of this project will include applying and adapting existing industry leading ML methods to medical image analysis as well as developing novel ML and AI methods for this purpose. Specifically, this will focus on the development of Transformer architectures and Convolutional Neural Networks for the analysis of medical images.

This project falls within the EPSRC information and communications technologies research area where Artificial Intelligence Technologies is one of the themes listed. Our focus specifically aligns to developing AI tools for computer vision in medicine. The project also falls within the Biological Informatics theme as we are developing artificial intelligence technologies for use in medicine and medical research for the automated analysis of medical data.

The project is completed in collaboration with specialist clinical researchers at Oxford University and The University of Bristol Musculoskeletal Research Unit, who provide a clinical foundation for our project.