Machine Learning-based Tools for Predicting Neurological Deterioration in Non-operative Degenerative Cervical Myelopathy Patients
Date
2024-06-27
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
Background
Degenerative cervical myelopathy (DCM) is the most common form of atraumatic spinal cord injury globally. DCM is characterized by the progressive compression of the cervical spinal cord as a result of vertebral column spondylotic degeneration. While surgery is currently the only effective treatment of DCM, clinical guidelines remain unclear on the benefit of surgery for patients with mild forms of DCM. This thesis utilizes machine learning based tools to elucidate clinical and imaging indicators of neurological deterioration in non-operative DCM patients. This thesis is comprised of two independent studies, one focussing on clustering patient groups at risk of deteriorating and the second focusing on developing a supervised machine learning (ML) model capable of predicting neurological deterioration. Methods DCM patients recruited from 2016-2023 underwent MRI scans, including T2w, diffusion tensor imaging (DTI), and magnetization transfer (MT) scans, along with a series of clinical metrics. These were collected every six months, resulting in 124 overall entries. T2w imaging scans were assessed for spinal cord compression, and cervical spinal canal diameter (SCD) was measured. Clustering was achieved through PaCMAP dimensionality reduction and K-Means clustering for the first study. Logistic regressions, support vector classifiers, and random forest classifiers were trained and tested for the second study. Findings We elucidated five patient groups with their associated risks of deterioration, according to both SCD range and cord compression pattern. Furthermore, we found that the compression pattern is unimportant at SCD extremes (≤14.5 mm or >15.75mm). Our best-performing supervised ML model had a testing balanced accuracy of 0.830 and ROC-AUC of 0.87. The three most important metrics for predicting neurological deterioration based on the model were MT ratio above the maximally compressed cervical level in the dorsal and ventral funiculi, and moderate tingling in the arm, shoulder, or hand (quickDASH item 10). Significance and Conclusion SCD and focal cord compression alone do not reliably predict an increased risk of neurological deterioration, their combination does. Furthermore, MT and DTI scans improve the prediction of neurological deterioration in non-operative mild DCM patients.
Description
Keywords
Machine Learning, Spinal Cord Injury, Degenerative Cervical Myelopathy
Citation
Al-Shawwa, A. -J. (2024). Machine learning-based tools for predicting neurological deterioration in non-operative degenerative cervical myelopathy patients (Master's thesis, University of Calgary, Calgary, Canada). Retrieved from https://prism.ucalgary.ca.