Brain Tumor Segmentation and Radiomics Survival Prediction: Contribution to the BRATS 2017 Challenge

Abstract

Quantitative analysis of brain tumors is critical for clinical decision making. While manual segmentation is tedious, time consuming and subjective, this task is at the same time very challenging to solve for automatic segmentation methods. In this paper we present our most recent effort on developing a robust segmentation algorithm in the form of a convolutional neural network. Our network architecture was inspired by the popular U-Net and has been carefully modified to maximize brain tumor segmentation performance. We use a dice loss function to cope with class imbalances and use extensive data augmentation to successfully prevent overfitting. Our method beats the current state of the art on BraTS 2015, is one of the leading methods on the BraTS 2017 validation set (dice scores of 0.896, 0.797 and 0.732 for whole tumor, tumor core and enhancing tumor, respectively) and achieves very good Dice scores on the test set (0.858 for whole, 0.775 for core and 0.647 for enhancing tumor). We furthermore take part in the survival prediction subchallenge by training an ensemble of a random forest regressor and multilayer perceptrons on shape features describing the tumor subregions. Our approach achieves 52.6% accuracy, a Spearman correlation coefficient of 0.496 and a mean square error of 209607 on the test set.

Cardiac substructure segmentation with deep learning for improved cardiac sparing

“…, left) to aggregate high level information using context modules and an expansion pathway (Fig. , right) to combine feature and spatial information for localization . Context modules (Fig.…”

“…, right). Deep supervision allows for the injection of gradient signals deep into the network, as it speeds up convergence and enhances training efficiency when there is a small amount of available labeled training data . An elementwise summation with upsample was then applied across all added segmentation layers to generate the final segmentation.…”

Hippocampal segmentation for brains with extensive atrophy using three‐dimensional convolutional neural networks

“…(c) We employed deep supervision (Kayalibay et al, ; Lee, Xie, Gallagher, Zhang, & Tu, ) at the expanding pathway by adding earlier feature maps at different levels of the network and combining them via element‐wise summation to form the final network output. (d) Since the 3D network is memory expensive, we opted to use instance normalization (Isensee, Kickingereder, Wick, Bendszus, & Maier‐Hein, ; Ulyanov, Vedaldi, & Lempitsky, ) instead of the commonly used batch normalization as the stochasticity generated by a small batch size may destabilize batch normalization. (e) To generate segmentation maps from the entire input image, we relied on trainable deconvolution kernels as the upsampling operations.…”

Automated brain extraction of multisequence MRI using artificial neural networks

“…The topology of the ANN underlying the HD-BET algorithm was inspired by the U-Net image segmentation architecture (Ronneberger, Fischer, & Brox, 2015) and its 3D derivatives (Çiçek, Abdulkadir, Lienkamp, Brox, & Ronneberger, 2016;Kayalibay, Jensen, & van der Smagt, 2017;Milletari, Navab, & Ahmadi, 2016) and has recently been shown to have excellent performance in brain tumor segmentation both in an international competition (Isensee, Kickingereder, Wick, Bendszus, & Maier-Hein, 2018) as well as in the context of a largescale multi-institutional study (Kickingereder et al, 2019). Methods S2, Supporting Information, contain an extended description of the architecture, as well as the training and evaluation procedure.…”