Developing AI techniques to quantify vertebral damage and predict progressive vertebral deformity using plain radiographs
Vertebral compression fracture is one of the most common fractures in the elderly with osteoporosis. Although most patients with vertebral compression fracture have favourable clinical outcomes after conservative treatment, there is a subgroup of patients who develop progressive vertebral collapse over time, resulting in disabling back pain and spinal deformity with associated significant comorbidity. These long-term complications substantially impair quality of life in the elderly. For optimal practice, these patients should be identified when they present with a vertebral fracture so that timely preventive clinical interventions can be applied to stop the pathogenesis of progressive vertebral collapse. However, currently there is no screening tool available to help clinicians identify these patients.
Our previous research shows that a key risk factor for the progression of vertebral fracture is the extent of initial vertebral damage. Subsequent research developed a mathematical model to predict progressive deformity using vertebral damage intensity. Based on this model our latest research has developed a novel method to quantify vertebral damage using morphometric measurements that can be made in routine clinical imaging. Specifically, the method estimates vertebral damage by measuring the change of vertebral shape on plain lateral radiograph taken concurrently in lying and then standing postures. It has the potential to be developed into a clinical tool that can help clinicians predict whether a patient is at risk of progressive vertebral collapse. However, the ability of this method to assess damage is dependent on the resolution of current imaging analysis techniques. Although plain radiographs have one of the best spatial resolutions (0.1 mm) among clinical imaging modalities, morphometric measurements have to be made through a manual digitization process that involves the subjective selection of specific landmark points on radiographs, which may decrease the precision to 0.5 mm. The precision may be further decreased when measuring changes of vertebral shape as this involves the manual digitization on two different radiographs. It is possible to use artificial intelligence (AI) to address these issues as deep learning systems may be able to identify specific landmarks on two related radiographs (e.g. radiographs taken for the same patient at standing and lying postures) more effectively and more consistently. AI may also help develop the method into a tool that can automatically measure vertebral damage to facilitate its application in clinical settings.
Therefore, the purpose of the proposed project is to develop AI techniques that enables automatic and high-precision measurements of vertebral shape changes on related plain radiographs.
For more information about the project, please contact Dr Jin Lou
