Dimensionality Reduction Hybrid U-Net for Brain Extraction in Magnetic Resonance Imaging

Abstract

In various applications, such as disease diagnosis, surgical navigation, human brain atlas analysis, and other neuroimage processing scenarios, brain extraction is typically regarded as the initial stage in MRI image processing. Whole-brain semantic segmentation algorithms, such as U-Net, have demonstrated the ability to achieve relatively satisfactory results even with a limited number of training samples. In order to enhance the precision of brain semantic segmentation, various frameworks have been developed, including 3D U-Net, slice U-Net, and auto-context U-Net. However, the processing methods employed in these models are relatively complex when applied to 3D data models. In this article, we aim to reduce the complexity of the model while maintaining appropriate performance. As an initial step to enhance segmentation accuracy, the preprocessing extraction of full-scale information from magnetic resonance images is performed with a cluster tool. Subsequently, three multi-input hybrid U-Net model frameworks are tested and compared. Finally, we propose utilizing a fusion of two-dimensional segmentation outcomes from different planes to attain improved results. The performance of the proposed framework was tested using publicly accessible benchmark datasets, namely LPBA40, in which we obtained Dice overlap coefficients of 98.05%. Improvement was achieved via our algorithm against several previous studies.

Keywords: U-Net; brain extraction; magnetic resonance imaging (MRI); semantic segmentation; whole-brain segmentation.

Figure 1

Schematic diagram of the proposed method.

Figure 2

Traditional hybrid input-layer net.

Figure 3

Full connect hybrid U-Net.

Figure 4

Decoder connection with two separate encoder net.

Figure 5

Comparison of two-dimensional U-Net effects of marginal region voxels under different profiles. (a) The position of the upper edge region voxel (in red) of the transverse section in the two-dimensional coronal section (in blue). (b) Poorly segmented two-dimensional U-net results for edge transverse plane images; the left blue mask shows the poor predicted result with U-Net, and the right red mask shows the label. (c) Two-dimensional U-Net segmentation results in the coronal plane; the left blue mask shows the predicted result with U-Net, and the right red mask shows the label.

Figure 6

A comparative analysis of prediction accuracy and brain tissue proportion across various serial number sections.

Figure 7

LPBA40 dataset; ground truth mask overlaid on MRI data.

Figure 8

Comparison of evaluation scores of BET, U-Net, Auto-U-Net, and the proposed method: (a) Dice; (b) sensitivity; (c) specificity.

Figure 8

Comparison of evaluation scores of BET, U-Net, Auto-U-Net, and the proposed method: (a) Dice; (b) sensitivity; (c) specificity.

Figure 9

The predicted masks superimposed on MRI data in the coronal plane. These four images show the improvement of the predicted brain mask in different steps of the proposed algorithm. (a) The predicted brain mask with the original U-Net, and several deficiencies are marked with boxes in yellow. (b) The prediction performance is improved with K-means preprocessing, whereas one can still see the drop-off at the edges of the brain. (c) The image demonstrates the enhancement of the edge area through the utilization of a hybrid net, which can be seen within the red box. (d) The overlay of the referenced label mask onto the MRI data.

Figure 10

Predicted masks overlaid on MRI data (transverse section). (a) The brain mask predicted using the original U-Net model. (b) The improvement in prediction performance achieved through K-means preprocessing. (c) The enhancement in the edge area achieved by employing a hybrid net. (d) The good performance obtained using the PHC-U-Net model. (e) The superimposed reference label mask on the MRI data.