LumbarDiagnostics: A framework for diagnosing lumbar spine pathology
MetadataShow full item record
Low back pain is one of the major public health problems in industrialized countries. It not only imposes economic burden on a patient and his or her family members, but also decreases the productivity of the individual. It is reported that in regard to spine more than 30 million magnetic resonance imaging (MRI) exams are conducted annually in the United States and the number of scans is growing exponentially. However, compared to the number of scans, the number of radiologists is increasing linearly, resulting in a concern over lack of diagnostic radiologists. Therefore, there is an increasing need for the computer-aided effective management of pathology in lumbar spine using multi-protocol MR images. MRI is known to be the most effective primary diagnostic tool in the clinical evaluation of the lumbar spine since it provides more anatomic sources of pain including nerves, muscles, and ligaments than X-ray or computed tomography. Interestingly, some clinical studies show that computer-aided diagnosis enhances the number of breast cancer detection by about 10%. To this end, I propose a computer-aided diagnosis framework, LumbarDiagnostics, for computer-aided characterization and diagnosis of lumbar spine pathology using multi-protocol MRI in a reliable and rapid manner, allowing to enhance the effectiveness and efficiency of the examination procedures of the lumbar spine pathology. The proposed framework consists of the following components: target pathology selection, meta data analysis, inter- and intra-slice analysis, preprocessing, landmark localization and labeling, regions-of-interest determination, region-of-interest quantification, feature generation, classification, reference generation, and validation. Experiments on clinical MRI image data from 101 subjects based on the framework successfully detect disc herniation and spinal stenosis with more than 90% diagnostic accuracy, achieving a speedup factor of 30 by comparison with diagnosis by radiologists. The framework also works effectively in segmenting the spinal canal region in MRI. In addition, the proposed framework is successfully applied to the task of spinal canal boundary extraction. This study demonstrates that the framework can be applied to the diagnosis and quantification of various abnormalities in the lumbar spine.