Deep Learning Framework for Liver Segmentation from T1-Weighted MRI Images

Abstract

The human liver exhibits variable characteristics and anatomical information, which is often ambiguous in radiological images. Machine learning can be of great assistance in automatically segmenting the liver in radiological images, which can be further processed for computer-aided diagnosis. Magnetic resonance imaging (MRI) is preferred by clinicians for liver pathology diagnosis over volumetric abdominal computerized tomography (CT) scans, due to their superior representation of soft tissues. The convenience of Hounsfield unit (HoU) based preprocessing in CT scans is not available in MRI, making automatic segmentation challenging for MR images. This study investigates multiple state-of-the-art segmentation networks for liver segmentation from volumetric MRI images. Here, T1-weighted (in-phase) scans are investigated using expert-labeled liver masks from a public dataset of 20 patients (647 MR slices) from the Combined Healthy Abdominal Organ Segmentation grant challenge (CHAOS). The reason for using T1-weighted images is that it demonstrates brighter fat content, thus providing enhanced images for the segmentation task. Twenty-four different state-of-the-art segmentation networks with varying depths of dense, residual, and inception encoder and decoder backbones were investigated for the task. A novel cascaded network is proposed to segment axial liver slices. The proposed framework outperforms existing approaches reported in the literature for the liver segmentation task (on the same test set) with a dice similarity coefficient (DSC) score and intersect over union (IoU) of 95.15% and 92.10%, respectively.

Keywords: MRI; T1-weighted contrast; automated liver segmentation; deep learning; diagnostic radiology.

Figure 1

Flow diagram explaining methodology for automated liver segmentation from $T_1$-weighted MRI scans.

Figure 2

Superficial visualization of relevant abdominal anatomy for describing underlying ambiguity in the segmentation task.

Figure 3

Visualization of MRI slices from (a) upper pole of the kidney, (b) inferior part of the liver, (c) middle part of the liver, and (d) superior part of the liver for different types of data available in the dataset.

Figure 4

Flow diagram explaining the methodology for fold creation and augmentation in the training set.

Figure 5

Visualization of image enhancement techniques.

Figure 6

Network architectures of different segmentation networks and investigation frameworks. The varying depth of pretrained dense, residual, and inception encoder backbones were investigated for UNet++, UNet, and FPN segmentation network architectures.

Figure 7

Cascaded network for handling anatomical ambiguity: the generalized network predicts an initial mask; if the pixel count for the predicted mask refers to a constrained or null liver content, then the input slice is fed into the specialized network.

Figure 8

Visualization of the predicted masks from the networks with DenseNet backbones for a sample axial slice showing the middle part of the liver (a) and the inferior part of the liver (b).

Figure 9

Visualization of the predicted masks from selected networks for a sample slice in (a) middle part of the liver (large liver content), (b) inferior part of the liver (small liver content), and (c) upper pole of the kidney (no liver content).

Figure 10

Comparison of the predicted masks from the generalized and specialized networks for sample MR slices with small liver content.