Grant: Undergraduate: Summer Studentship Award - Institute Community Support

Scoliosis, which is an excessive curvature of the spine, affects approximately 1 in 1,000 individuals in early adolescence, progressing until the spine is fully developed. Bracing is a common treatment method, which decreases the progression of curves and significantly reduces the probability of needing surgery. For bracing to be effective, early referral is critical, but often difficult due to the lack of a safe and widely usable spinal curvature measurement method. One possible spinal measurement method is ultrasound; spinal curvatures can accurately be computed from skeletal landmarks, typically the transverse processes, localized in the ultrasound images. But obtaining ultrasound images of the spine that contain transverse processes with the necessary clarity is a technically difficult skill which requires considerable training and practice. In order to aid the sonographer and make spinal ultrasound scanning a more straightforward practice, we propose the use of deep learning to help with the detection and localization of transverse processes in the image. Deep learning using convolutional neural networks is very powerful for image classification and computer vision and it can be applied to the spinal ultrasound images. Use of deep learning may open the way for ultrasound to be deployed as a wide-scale safe, affordable and accurate scoliosis screening tool.