Development of an automated image processing tool for spinal canal measurements

Dr Cheung, Jason Pui Yin   (Principal Investigator (PI))

Dr Hu Yong   (Co-Investigator)

18

2018-06-15

2019-12-14

138980

Development of an automated image processing tool for spinal canal measurements

Automated, Developmental spinal stenosis, Machine learning, Measurements, MRI, Spinal Canal

Orthopaedics/Traumatology,Artificial Intelligence and Machine learning

201711160001

Seed Fund for PI Research – Translational and Applied Research

2017

Completed

Patients with lumbar spinal stenosis requiring surgery generally have good clinical response after operation.[1-3] However, reoperation after decompression for lumbar spinal stenosis is not an uncommon event. In the Spine Patient Outcomes Research Trial (SPORT) for spinal stenosis, 8% of patients who underwent an operation required another operation within two years of the index operation and 13% had another operation within four years.[2, 3] In a similar study, 23% of patients were reported to require a repeat surgery by ten years.[4] In Korea, Kim et al. showed that reoperation rates were up to 74% at 1 year, 9% at 2 years and 13.4% at 5 years after surgery.[5, 6] Reasons for reoperation vary from case to case including pseudoarthrosis, persistent pain or recurrent symptoms, instrumentation failure or progressive degeneration at another spine level.[7] This raises an important concern in clinical practice. Although adjacent level symptoms are known to be associated with fusion surgery, new evidence has shown that it also occurs with only decompression surgery.[8] This leads us to believe that these patients are prone to developing symptoms at multiple levels. Reoperation has been reported to occur in 13% of patients with approximately 50% of reoperations performed at adjacent levels.[8] This leads to an average of 3.3% of patients per year. Other studies showed that reoperation after decompressive laminectomy varied from 5-23%.[9-11] Patients may develop symptoms at adjacent levels due to progressive degenerative changes in the presence pre-existing narrowed canals. Characteristically, these patients have generalized short pedicles suggestive of genetic disturbances during the fetal and postnatal period.[12] A narrowed spinal canal can be due to developmental problems such as the articular processes in an early embryonic stage, disproportional growth of the lamina and pedicles, and also of the spinal canal and spinal roots.[12] Pathological changes in developmental spinal stenosis include narrowing of dorsal aspect of spinal canal due to bulging of the inferior articular facets. The lamina is also enlarged with narrowed interlaminar spaces. Pedicles are shortened leading to decreased interpedicular distances.[13] Vertebral bodies may also be wedged and has posterior lipping, contributing to the narrow spinal canal.[14] Occasionally, the lower lumbar vertebral levels may produce a trefoil or three-leaf clover shaped canal.[15] This configuration may also predispose compression of lumbar and sacral nerve roots.[15] Thus, with this distorted canal morphology, patients may develop claudication symptoms more readily than patients with normally developed canals. This condition known as developmental spinal stenosis (DSS) results in reduction of space available for accommodating neural contents, increasing not only the risk of developing stenotic symptoms, but also symptom recurrence requiring repeated surgery at other involved levels.[16] The investigators published a series of novel magnetic resonance imaging (MRI) based studies. This includes assessment of canal dimensions on T1- and T2- weighted scans[17], and comparative study between symptomatic patients who underwent surgery and asymptomatic individuals finding the anteroposterior (AP) bony spinal canal diameter as the most relevant MRI measurement of DSS.[18] Level-specific cut-off values with the best sensitivity and specificity for each vertebral level has been generated for the diagnosis of DSS.[18] Although the investigators have found a paradoxical relationship between the degree of LF fibrosis and the canal size[19] and several single nucleotide polymorphisms and candidate genes that are associated with canal narrowing[20], its pathophysiology and etiology still requires further research. In particular, with regards to the genetic origins of DSS, large population-based cohorts are useful tools for identifying genetic polymorphisms. However, precise phenotyping of subjects is needed for these genetic association studies to be useful. Manually phenotyping every subject in a large cohort is time-consuming. Thus, this project aims to develop an automated method to measure different parameters of the lumbar spine thereby providing a quick and efficient way of phenotyping a large population of individuals. Objectives: 1) Algorithm development for automatic image processing and region-of-interest segment of original lumbar spine MR images 2) Software development with user-friendly interface 3) Clinical validation