Fully Automated MR Detection and Segmentation of Brain Metastases in Non‐small Cell Lung Cancer Using Deep Learning

Stephanie T. Jünger, Ulrike Cornelia Isabel Hoyer, Diana Schaufler, Kai Roman Laukamp, Lukas Goertz, Frank Thiele, Jan‐Peter Grunz, Marc Schlamann, Michael Perkuhn, Christoph Kabbasch, Thorsten Persigehl, Stefan Grau (+4 others)
2021 Journal of Magnetic Resonance Imaging  
Non-small cell lung cancer (NSCLC) is the most common tumor entity spreading to the brain and up to 50% of patients develop brain metastases (BMs). Detection of BMs on MRI is challenging with an inherent risk of missed diagnosis. To train and evaluate a deep learning model (DLM) for fully automated detection and 3D segmentation of BMs in NSCLC on clinical routine MRI. Retrospective. Ninety-eight NSCLC patients with 315 BMs on pretreatment MRI, divided into training (66 patients, 248 BMs) and
more » ... ependent test (17 patients, 67 BMs) and control (15 patients, 0 BMs) cohorts. T1 -/T2 -weighted, T1 -weighted contrast-enhanced (T1 CE; gradient-echo and spin-echo sequences), and FLAIR at 1.0, 1.5, and 3.0 T from various vendors and study centers. A 3D convolutional neural network (DeepMedic) was trained on the training cohort using 5-fold cross-validation and evaluated on the independent test and control sets. Three-dimensional voxel-wise manual segmentations of BMs by a neurosurgeon and a radiologist on T1 CE served as the reference standard. Sensitivity (recall) and false positive (FP) findings per scan, dice similarity coefficient (DSC) to compare the spatial overlap between manual and automated segmentations, Pearson's correlation coefficient (r) to evaluate the relationship between quantitative volumetric measurements of segmentations, and Wilcoxon rank-sum test to compare the volumes of BMs. A P value <0.05 was considered statistically significant. In the test set, the DLM detected 57 of the 67 BMs (mean volume: 0.99 ± 4.24 cm3 ), resulting in a sensitivity of 85.1%, while FP findings of 1.5 per scan were observed. Missed BMs had a significantly smaller volume (0.05 ± 0.04 cm3 ) than detected BMs (0.96 ± 2.4 cm3 ). Compared with the reference standard, automated segmentations achieved a median DSC of 0.72 and a good volumetric correlation (r = 0.95). In the control set, 1.8 FPs/scan were observed. Deep learning provided a high detection sensitivity and good segmentation performance for BMs in NSCLC on heterogeneous scanner data while yielding a low number of FP findings. Level of Evidence 3 Technical Efficacy Stage 2.
doi:10.1002/jmri.27741 pmid:34032344 fatcat:cfbkv5acwfcjnlj2ix2hkfrz6e