P15.09.A Canny edge detection algorithm for quantitative differentiation between diffuse and circumscript glioma growth patterns on MRI. (5th September 2022)

• Record Type: Journal Article
• Authors: Thiele, F; Weller, J; Katzendobler, S; Trumm, C; Thon, N; Tonn, J
• Is Part Of: Neuro-oncology. Volume 24(2022)Supplement 2
• Page Start: ii85
• Page End: ii86
• Publication Date: 2022-09-05
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noac174.299
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: Background: Lower grade gliomas show heterogenous appearance on T2-weighted MRI. Some tumors grow diffusely along axonal structures whereas others distort adjacent brain tissue through local mass effect. The diagnostic, therapeutic and prognostic implication of differential growth patterns on MRI remain unknown and are difficult to assess quantitatively. Material and Methods: A web-based application allowing for image preprocessing and providing a comprehensive edge detection tool by means of quantifying tumor border delineation on T2-weighted images based on the canny edge detection algorithm was developed. A sigma value between 1 and 100 determined the threshold where tumor borders where not detected by the algorithm anymore, with 1 equating to the lowest threshold and thus detection of all edges contained in the image. Two experienced faculty members assigned sigma values to axial T2 images of a random sample of 20 WHO grade 2 astrocytomas, IDH-mutant and 1p/19q-non-codeleted. The sigma values were then compared with a binary, subjective rating by the same faculty staff according to the perceived predominant growth pattern (diffuse versus circumscript) of each glioma. Results: When subjectively categorizing tumors binarily (diffuse versus circumscript), there was moderate interrater variability between observers (cohen's kappa=0.6). Raters agreed in 16 of 20 cases, terming 7 gliomas unanimously diffuse and 9 gliomas circumscript. In 4 cases, the raters opinions diverged. The sigma values differed significantly between diffuse and circumscript tumors in both raters (rater 1, p=0.002; rater 2, p=0.018). For rater 1, the mean sigma difference between diffuse and circumscript tumors was 10.7 and 9.3 for rater 2. Conclusion: Edge detection algorithms can be efficiently applied on MRI scans and are highly accurate in differentiating diffuse from circumscript gliomas. Objectification demands defining imaging criteria for diffuse and circumscript appearance of lower grade gliomas on MRI.