2019/12/26 by Endre Grøvik, Darvin Yi, Grøvik, Endre +22
Medicine · #Advanced MRI Techniques and Applications #Brain Metastases and Treatment #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #MRI in cancer diagnosis #Medical Imaging Techniques and Applications #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1912.11966
openalex publication_date 2019/12/26 · openalex created_date 2022/10/02 · openalex updated_date 2026/07/28
The purpose was to assess the clinical value of a novel DropOut model for\ndetecting and segmenting brain metastases, in which a neural network is trained\non four distinct MRI sequences using an input dropout layer, thus simulating\nthe scenario of missing MRI data by training on the full set and all possible\nsubsets of the input data. This retrospective, multi-center study, evaluated\n165 patients with brain metastases. A deep learning based segmentation model\nfor automatic segmentation of brain metastases, named DropOut, was trained on\nmulti-sequence MRI from 100 patients, and validated/tested on 10/55 patients.\nThe segmentation results were compared with the performance of a\nstate-of-the-art DeepLabV3 model. The MR sequences in the training set included\npre- and post-gadolinium (Gd) T1-weighted 3D fast spin echo, post-Gd\nT1-weighted inversion recovery (IR) prepped fast spoiled gradient echo, and 3D\nfluid attenuated inversion recovery (FLAIR), whereas the test set did not\ninclude the IR prepped image-series. The ground truth were established by\nexperienced neuroradiologists. The results were evaluated using precision,\nrecall, Dice score, and receiver operating characteristics (ROC) curve\nstatistics, while the Wilcoxon rank sum test was used to compare the\nperformance of the two neural networks. The area under the ROC curve (AUC),\naveraged across all test cases, was 0.989+-0.029 for the DropOut model and\n0.989+-0.023 for the DeepLabV3 model (p=0.62). The DropOut model showed a\nsignificantly higher Dice score compared to the DeepLabV3 model (0.795+-0.105\nvs. 0.774+-0.104, p=0.017), and a significantly lower average false positive\nrate of 3.6/patient vs. 7.0/patient (p<0.001) using a 10mm3 lesion-size limit.\nThe DropOut model may facilitate accurate detection and segmentation of brain\nmetastases on a multi-center basis, even when the test cohort is missing MRI\ninput data.\n