AUTOMATICALLY DIAGNOSING HIP CONDITIONS FROM X-RAYS USING LANDMARK DETECTION

When patients present with symptoms of hip pain a clinician might diagnose a condition called femoroacetabular impingement (FAI), where the ball and socket of the hip joint rub together during movement. To diagnose FAI a doctor inspects an x-ray, and records the angles between certain key points in the image. If the angles are 'too big' then FAI is diagnosed. We anticipate that these key points can be located in an x-ray using deep learning and thus the angles measured and FAI diagnosed automatically. In this paper we deploy a stacked hourglass network to automatically locate key-points in hip x-rays, which we then use to automatically diagnose FAI in a patient. On a test set of 112 hips our algorithm diagnoses cam impingement, one of two types of FAI, correctly 90% of the time. To our knowledge this is the first time any kind of FAI has been automatically diagnosed.