Automated detection of lacunar infarcts from MRI images

Background When MRI images of patients with memory problems are assessed, the presence lacunar infarcts is evaluated. Lacunar infarcts, fluid-filled cavities 3-15 mm in diameter, one possible result cerebral small vessel disease and therefore relevant dementia diagnostics. In this study, we propose an automated tool for detecting on images. Method We adapted our previous approach [1] three ways: 1) post-processing steps were added to convolutional neural network segmentation pipeline, e.g., exclude enlarged perivascular spaces segmentations located inside sulci or ventricles, 2) clinic data from Amsterdam Dementia Cohort PredictND cohort included addition LADIS data, 3) ground truth was defined by experienced neuroradiologists (A, B, C) instead rater. T1 T2-FLAIR available all cases T2 a subset visual rating. The dataset composed 393 subjects each expert rated 1156 potential as “1 = no”, “2 probably “3 yes” “4 yes”. A consensus (CO) created calculating average using cutoff at 2.5. agreement measured between individual raters (A-B, B-C, A-C), automatic rating (AU-A, AU-B, AU-C), (AU-CO) Kappa. Correlation coefficient number per subject defined. Cross-validation used training algorithm. Result Out candidate regions evaluated, 540 considered based consensus. agreed other 74% whether shown lesion infarct (kappa 0.48). Correspondingly, 0.47). compared consensus, agreement/accuracy 78% 0.55). correlation detected automatically whole image visually (consensus) 0.85 (Figure 1). Conclusion This study suggests that comparable inter-rater variability neuroradiologists. Jokinen et al. Stroke 2020 Nov 9; 50(1)