CEREBRUM: a fast and fully-volumetric Convolutional Encoder-decodeR for weakly-supervised sEgmentation of BRain strUctures from out-of-the-scanner MRI

Abstract

Many functional and structural neuroimaging studies call for accurate morphometric segmentation of different brain structures starting from image intensity values of MRI scans. Current automatic (multi-) atlas-based segmentation strategies often lack accuracy on difficult-to-segment brain structures and, since these methods rely on atlas-to-scan alignment, they may take long processing times. Alternatively, recent methods deploying solutions based on Convolutional Neural Networks (CNNs) are enabling the direct analysis of out-of-the-scanner data. However, current CNN-based solutions partition the test volume into 2D or 3D patches, which are processed independently. This process entails a loss of global contextual information, thereby negatively impacting the segmentation accuracy. In this work, we design and test an optimised end-to-end CNN architecture that makes the exploitation of global spatial information computationally tractable, allowing to process a whole MRI volume at once. We adopt a weakly supervised learning strategy by exploiting a large dataset composed of 947 out-of-the-scanner (3 Tesla T1-weighted 1mm isotropic MP-RAGE 3D sequences) MR Images. The resulting model is able to produce accurate multi-structure segmentation results in only a few seconds. Different quantitative measures demonstrate an improved accuracy of our solution when compared to state-of-the-art techniques. Moreover, through a randomised survey involving expert neuroscientists, we show that subjective judgements favour our solution with respect to widely adopted atlas-based software.

CEREBRUM-7T: Fast and Fully-volumetric Brain Segmentation of 7 Tesla MR Volumes

“…From the quantitative assessment presented in Section 3.1 emerges that there is a relative difference in performance between CEREBRUM-7T architectures and the automatic GT used for training. Nevertheless, since performance are measured with respect to the automatic GT, CEREBRUM-7T segmentations might be even superior to those provided in the automatic GT, as in Bontempi et al (2020) and Roy et al (2019), as we also suggest in Figure 2 where CEREBRUM-7T masks appear more accurate than GT masks.…”

“…In this work we reshape the framework presented in Bontempi et al (2020) to handle UHF high-resolution data, delivering CEREBRUM-7T, the first DL fully automatic solution for brain MRI segmentation on out-of-the-scanner 7T data 2 .…”

“…Similarly to Bontempi et al (2020), CEREBRUM-7T acts in a fully 3D fashion on brain volumes, as shown in Figure 1. This is possible simplifying the model architecture with respect to other DL-based segmentation methods, and also to CEREBRUM, since the increase in the data size and the GPU memory constraints.…”

“…This is possible simplifying the model architecture with respect to other DL-based segmentation methods, and also to CEREBRUM, since the increase in the data size and the GPU memory constraints. Such design choice enables to achieve a full brain segmentation in few seconds, as in Bontempi et al (2020), compared to several hours of other currently used methods for 7T data. Furthermore, exploiting the ability of DL methods to efficiently learn internal representations of data, segmentation happens in native space, with neither the support of ad-hoc preprocessing, nor the alignment to reference atlases.…”

“…Similarly to Bontempi et al (2020), the model architecture is designed so as to deal with the dimensionality of the training data (i.e., 256 × 352 × 224 voxels) at once. As shown in Figure 1, the model is a deep encoder/decoder network with three layers, with one, two, and three 3D convolutional blocks in the first, second, and third level respectively (n_f ilters = 24, 48, 96).…”