Multi-parameter-based radiological diagnosis of Chiari Malformation using Machine Learning Technology.

BACKGROUND: The known primary radiological diagnosis of Chiari Malformation-I (CM-I) is based on the degree of tonsillar herniation (TH) below the Foramen Magnum (FM). However, recent data also shows the association of such malformation with smaller posterior cranial fossa (PCF) volume and the anatomical issues regarding the Odontoid. This study presents the achieved result regarding some detected potential radiological findings that may aid CM-I diagnosis using several machine learning (ML) algorithms. MATERIALS AND METHODS: Midsagittal T1-weighted MR images were collected in 241 adult patients diagnosed with CM, eleven morphometric measures of the posterior cerebral fossa were performed. Patients whose imaging was performed in the same centre and on the same device were included in the study. By matching age and gender, radiological exams of 100 clinically/radiologically proven symptomatic CM-I cases and 100 healthy controls were assessed. Eleven morphometric measures of the posterior cerebral fossa were examined using 5 designed ML algorithms. RESULTS: The mean age of patients was 29.92 ± 15.03 years. The primary presenting symptoms were headaches (62%). Syringomyelia and retrocurved-odontoid were detected in 34% and 8% of patients, respectively. All of the morphometric measures were significantly different between the groups, except for the distance from the dens axis to the posterior margin of FM. The Radom Forest model is found to have the best 1.0 (14 of 14) ratio of accuracy in regard to 14 different combinations of morphometric features. CONCLUSION: Our study indicates the potential usefulness of ML-guided PCF measurements, other than TH, that may be used to predict and diagnose CM-I accurately. Combining two or three preferable osseous structure-based measurements may increase the accuracy of radiological diagnosis of CM-I.