When you visit any website, it may store or retrieve information on your browser, mostly in the form of cookies. This information might be about you, your preferences or your device and is mainly used to make the site work as you expect it to. The information does not directly identify you but can give you a more personalized web experience. Because we respect your right to privacy, you can choose not to allow some types of cookies. Click on the different category headings to learn more and change our default settings. However, blocking some types of cookies may impact your experience of the site. You can find more information, including a detailed cookie explanation, on our Cookie Policy.
Independent Validation of a Deep Learning nnU-Net Tool for Neuroblastoma Detection and Segmentation in MR Images
Tumor segmentation is crucial in oncologic imaging processing. Using a large, diverse dataset, we validated a fully automatic deep learning nnU-Net-based model for detecting and segmenting neuroblastic tumors on MR images in a substantial children cohort.
We utilized an international multicenter, multivendor imaging repository of 300 children with neuroblastic tumors, encompassing 535 MR T2-weighted sequences. The automatic segmentation, part of the PRIMAGE project, was validated against manual editing by an expert radiologist, recording the time needed for manual adjustments and comparing overlaps and spatial metrics.
The median Dice similarity coefficient (DSC) was high at 0.997; 0.944–1.000 (median; Q1–Q3), with 6% of MR sequences unidentifiable by the net. No significant variations were observed in net performance post-chemotherapy or concerning MR magnetic field, sequence type, or tumor location. Visual inspection time averaged 7.9 ± 7.5 seconds, while manual editing, when needed (136 masks), took 124 ± 120 seconds.
The automatic CNN accurately located and segmented the primary tumor in 94% of cases with high concordance to manually edited masks, marking the first validation of such model for neuroblastic tumor identification and segmentation with body MR images. The semi-automatic approach, necessitating minor manual editing, bolstered radiologist confidence while reducing workload.
https://www.mdpi.com/2072-6694/15/5/1622
