2019/09/11 by Dennis Bontempi, Bontempi, Dennis, Sergio Benini +7 · 1 citation
Biochemistry, Genetics and Molecular Biology · Computer Science · #Advanced Neural Network Applications #Cell Image Analysis Techniques #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #FOS: Electrical engineering #Image and Video Processing (eess.IV) #Medical Image Segmentation Techniques #electronic engineering #information engineering
paper · pdf · doi:10.48550/arxiv.1909.05085
openalex publication_date 2019/09/11 · openalex created_date 2022/07/19 · openalex updated_date 2026/07/28
Many functional and structural neuroimaging studies call for accurate\nmorphometric segmentation of different brain structures starting from image\nintensity values of MRI scans. Current automatic (multi-) atlas-based\nsegmentation strategies often lack accuracy on difficult-to-segment brain\nstructures and, since these methods rely on atlas-to-scan alignment, they may\ntake long processing times. Recently, methods deploying solutions based on\nConvolutional Neural Networks (CNNs) are making the direct analysis of\nout-of-the-scanner data feasible. However, current CNN-based solutions\npartition the test volume into 2D or 3D patches, which are processed\nindependently. This entails a loss of global contextual information thereby\nnegatively impacting the segmentation accuracy. In this work, we design and\ntest an optimised end-to-end CNN architecture that makes the exploitation of\nglobal spatial information computationally tractable, allowing to process a\nwhole MRI volume at once. We adopt a weakly supervised learning strategy by\nexploiting a large dataset composed by 947 out-of-the-scanner (3 Tesla\nT1-weighted 1mm isotropic MP-RAGE 3D sequences) MR Images. The resulting model\nis able to produce accurate multi-structure segmentation results in only few\nseconds. Different quantitative measures demonstrate an improved accuracy of\nour solution when compared to state-of-the-art techniques. Moreover, through a\nrandomised survey involving expert neuroscientists, we show that subjective\njudgements clearly prefer our solution with respect to the widely adopted\natlas-based FreeSurfer software.\n