A Novel Quantitative Automated Measurement of CSF Area for the Diagnosis of Lumbar Spinal Stenosis

Introduction A grading system for LSS is needed to allow precise communication between clinicians and for assessment of treatment responses. At present, there are two categories of grading systems- qualitative and quantitative. However, there are always individual subjectivity of radiologists in those qualitative classifications systems. The quantitative classifications systems are time-consuming and not relevant to clinical symptoms even with the same values for DSCSA or DSAPD. Herein, we propose a novel quantitative automated method to measure the cross-section area of the CSF. This study is to introduce this novel quantitative automated method, evaluate its reliability, and compare it to other two semiquantitative CSF-based classification systems. Methods All lumbar MRI studies were selected at random. Each study included T1/T2-weighted axial and sagittal images taken between April 2012 and July 2013. Two practicing neurosurgeons with spine specialty participated in this study and did the reviews of the grading for each disc level. Each of two neurosurgeons independently measured DSCSA (dural sac cross-sectional area, mm2) on T2-weighted axial images manually. The final DSCSA were the average of two results at each disc level of the lumbar spine. They also graded each disc level according to the qualitative grading system- the Lee's and Schizas' grading systems. The final grading of each disc level of the lumbar spine was determined by the most consistent grade out of the four results obtained from the two trials of two neurosurgeons. CSF area of each disc level (ACSFA) was calculated automatedly by the algorithm we developed. For each disc level, linear regressions were performed between DSCSA and ACSFA, and ACSFA was categorized based on the final grading of Lee's and Schizas' grades. Results Seventeen female and six male patients were studied (mean age, 65.21 ± 12.1years). The inter-reader reliability of Lee's and Schiza's grading systems were “moderate” and “good” (o.437 and 0.739), respectively. ACSFA was correlated well with DSCSA linearly (r=.8988,p,.001)(Fig. 1). Using one-way ANOVA, ACSFA was able to differentiate each grade of Lee's or Schizas' systems ( p < .0001 and p < .0001)(Fig. 2 & 3). Conclusion This novel automated quantitative measurement of CSF area can provide at least the some diagnostic value for the diagnosis of LSS, and, at the same time, it avoids the variability of the quantitative system between readers and overcomes the disadvantages of the present qualitative system (DSCSA). This method can serve as a useful diagnostic tool for the clinicians.