Background Advances in novel MRI techniques are promising for assessing progression of Ankylosing Spondylitis and effectiveness of treatment. Nevertheless, MRI is not widely utilized in clinical setting due to the lack of a fast, reproducible scoring system. A scoring method has been developed for the quantification of spinal lesions that is fast to process and reduces inter and intra observer variability and requires rough regions of interest (ROI). This method is built as a module of Dynamika, an image analysis diagnostic platform.
Objectives To provide a reproducible scoring method for the quantification of spinal lesions from static MRI scans.
Methods Baseline and follow-up cervical, thoracic and lumbar MRI scan series from 36 patients (216 scans in total) were processed by 3 different readers. The followed workflow for each reading was: 1) Draw spine outlining ROIs in sagittal views; 3) Compute a Norml parametric map  by using the middle slice ROI as a baseline; 4) Adjust MRI histogram centre and width to 1 and 2 respectively; 5) Group the 2D ROIs into a volume of interest (VOI); 6) Record the sum of all positive pixel values in the VOI (SumNormI).
The windowing of the MRI histogram is defined with the aim of quantifying the spine while excluding discs.
Pearson's Coefficient was calculated to assess the inter-observer variance. The coefficient was calculated for the SumNormI obtained for each VOI and for the normalized SumNormI (SumNormI divided by the number of slices comprising the VOI). After a week, readers were asked to repeat the analysis for 48 scans in order to measure the intra-observer variance with the Lin's Concordance Correlation Coefficient .
Results By adjusting the window as described, NormI focuses on intensities between healthy spine and discs, assuming lesion have intensities between both (see Figure).
Reproducibility was assessed in terms of inter and intra-observer variance (see Table). Pearson's Coefficient was significantly high for each pair of readers (p<0.005), which suggests that the proposed workflow is robust for different readers. Intra-observer reliability also achieved high values (p<0.005) when comparing the results recorded after a second reading.
Conclusions The presented workflow provides a method for scoring of spinal lesions from static MRI scans. This workflow is reproducible and only requires the drawing of rough ROIs.
O. Kubassova et al. A novel diagnostic platform for quantifying Ankylosing Spondylitis and general Spondylitis from MRI, in Proceedings of 99th Annual Meeting RSNA, December 1 – 6, 2013
Biometrics Vol. 45, No. 1, Mar., 1989
Acknowledgements We acknowledge Jener Basilio for his guidance and cooperation in the assessment of the reproducibility of this workflow.
Disclosure of Interest None declared