Abstract
MRI is the primary diagnostic modality for spinal cord tumors. However, its validity has never been vigorously scrutinized in daily routine clinical practice, where MRI tissue diagnosis is usually not a single one but multiple ones with several differential diagnoses. Here, we aimed to assess the validity of MRI in terms of predicting the pathology and location of the tumor in routine clinical settings. We analyzed 820 patients with primary spinal cord tumors, who have a pathological diagnosis and location in the operation record which were confirmed. We modified traditional measures for validity based upon a set of diagnoses instead of a single diagnosis. Sensitivity and specificity and positive and negative predictabilities were evaluated for the tumor location and pathology. For tumor location, 456 were intradural extramedullary; 165 were intramedullary, and 156 were extradural. The overall sensitivity and specificity were over 90.0%. However, the sensitivity became lower when the tumor resided simultaneously in two spaces such as in the intradural-and-extradural or intramedullary-and-extramedullary space (54.6% and 30.0%, respectively). Most common pathology was schwannoma (n = 416), followed by meningioma (114) and ependymoma (87). Sensitivities were 93.3%, 90.4%, and 89.7%, respectively. Specificities were 70.8%, 82.9%, and 76.0%. In rare tumors such as neurofibromas, and diffuse midline gliomas, the sensitivity was much lower (less than 30%). For common locations and pathologies, the validity of MRI is generally acceptable. However, for rare locations and pathologies, MRI diagnosis still needs some improvement.