RESUMO
In this article, we compare the performance of a state-of-the-art segmentation network (UNet) on two different glioblastoma (GB) segmentation datasets. Our experiments show that the same training procedure yields almost twice as bad results on the retrospective clinical data compared to the BraTS challenge data (in terms of Dice score). We discuss possible reasons for such an outcome, including inter-rater variability and high variability in magnetic resonance imaging (MRI) scanners and scanner settings. The high performance of segmentation models, demonstrated on preselected imaging data, does not bring the community closer to using these algorithms in clinical settings. We believe that a clinically applicable deep learning architecture requires a shift from unified datasets to heterogeneous data.
Assuntos
Aprendizado Profundo , Glioblastoma , Algoritmos , Glioblastoma/diagnóstico por imagem , Humanos , Imageamento por Ressonância Magnética , Estudos RetrospectivosRESUMO
AIM OF THE STUDY: It was found that the mutations in the SDHD gene, encoding one of subunits of the succinate dehydrogenase complex, lead to the development of head and neck paraganglioma (HNPGL). We analyzed this gene in 91 patients with HNPGL from Russia. MATERIALS AND METHODS: DNA was isolated from the whole blood. A screening for mutations was performed by Sanger sequencing. RESULTS: We revealed three missense mutations that have been described previously: p.Pro81Leu, p.His102Arg, p.Tyr114Cys. Moreover, we identified a novel potentially pathogenic variant (p.Trp105*). CONCLUSIONS: We found that mutations in the SDHD gene were less common in Russian patients compared with the majority of European populations. It was shown that the p.His102Arg mutation is a major mutation in Russia. We confirmed the previous suggestion that a bilateral localization of the tumor and the carotid type represent a marker of the genetically determined form of HNPGL.