作者: Thomas Sauvigny , Patrick M. House , Brigitte Holst , Stefan Stodieck , Sirko Pelzl
DOI: 10.1016/J.EPLEPSYRES.2021.106594
关键词:
摘要: Abstract Purpose Focal cortical dysplasias (FCDs) represent one of the most frequent causes pharmaco-resistant focal epilepsies. Despite improved clinical imaging methods over past years, FCD detection remains challenging, as FCDs vary in location, size, and shape commonly blend into surrounding tissues without clear definable boundaries. We developed a novel convolutional neural network for segmentation validated it prospectively on daily-routine MRIs. Material The was trained 201 T1 FLAIR 3 T MRI volume sequences 158 patients with mainly FCDs, regardless type, 7 PMG. Non-FCD/PMG MRIs, drawn from 100 normal MRIs 50 non-FCD/PMG pathologies, were added to training. applied algorithm consecutive epilepsy daily practice. results compared corresponding neuroradiological reports morphometric analyses evaluated by an experienced epileptologist. Results Best training reached sensitivity (recall) 70.1 % precision 54.3 detecting FCDs. Applied out 9 detected segmented correctly 77.8 specificity 5.5 %. conventional visual 33.3 94.5 %, respectively (3/9 detected); overall epileptologic evaluation both (9/9 detected) thus served reference. Conclusion 3D autoencoder regularization segmentation. Our employs largest dataset date various types some It provided higher than analyses. its low specificity, number false positively predicted lesions per lower analysis. consider our already useful pre-screening everyday