Healthy Kidney Segmentation in the Dce-Mr Images Using a Convolutional Neural Network and Temporal Signal Characteristics

Abstract

Quantification of renal perfusion based on dynamic contrast-enhanced magnetic resonance imaging (DCE-MRI) requires determination of signal intensity time courses in the region of renal parenchyma. Thus, selection of voxels representing the kidney must be accomplished with special care and constitutes one of the major technical limitations which hampers wider usage of this technique as a standard clinical routine. Manual segmentation of renal compartments-even if performed by experts-is a common source of decreased repeatability and reproducibility. In this paper, we present a processing framework for the automatic kidney segmentation in DCE-MR images. The framework consists of two stages. Firstly, kidney masks are generated using a convolutional neural network. Then, mask voxels are classified to one of three regions-cortex, medulla, and pelvis-based on DCE-MRI signal intensity time courses. The proposed approach was evaluated on a cohort of 10 healthy volunteers who underwent the DCE-MRI examination. MRI scanning was repeated on two time events within a 10-day interval. For semantic segmentation task we employed a classic U-Net architecture, whereas experiments on voxel classification were performed using three alternative algorithms-support vector machines, logistic regression and extreme gradient boosting trees, among which SVM produced the most accurate results. Both segmentation and classification steps were accomplished by a series of models, each trained separately for a given subject using the data from other participants only. The mean achieved accuracy of the whole kidney segmentation was 94% in terms of IoU coefficient. Cortex, medulla and pelvis were segmented with IoU ranging from 90 to 93% depending on the tissue and body side. The results were also validated by comparing image-derived perfusion parameters with ground truth measurements of glomerular filtration rate (GFR). The repeatability of GFR calculation, as assessed by the coefficient of variation was determined at the level of 14.5 and 17.5% for the left and right kidney, respectively and it improved relative to manual segmentation. Reproduciblity, in turn, was evaluated by measuring agreement between image-derived and iohexol-based GFR values. The estimated absolute mean differences were equal to 9.4 and 12.9 mL/min/1.73 m² for scanning sessions 1 and 2 and the proposed automated segmentation method. The result for session 2 was comparable with manual segmentation, whereas for session 1 reproducibility in the automatic pipeline was weaker.

Keywords: convolutional neural networks; dynamic contrast-enhanced MRI; glomerular filtration rate; kidney segmentation; perfusion quantification; pharmocokinetic modeling.

Figure 1

Overview of the Designed Segmentation Pipeline.

Figure 2

U-Net architecture of the convolutional neural network implemented for semantic segmentation of kidneys in the DCE-MR images.

Figure 3

Examples of training image patches extracted from left and right kidneys from two time frames of Subject 1. Data augmentation was realized by image flipping in horizontal direction and vertical shifting of patch location relative to image center.

Figure 4

Preparation of training data for supervised learning of classifiers: (a) ROI placement in a DCE-MRI frame; (b) signal time courses assigned to corresponding ROI voxels; (c) three-dimensional visualization of PCA feature vectors representing cortex (blue), medulla (red) and pelvis (magenta) ROIs. The visualization was obtained by transforming 20 PCA features using t-SNE method.

Figure 5

Examples of output segmentation masks compared against manual annotations for Subjects 1 (a) and 5 (b).

Figure 6

Evolution of the loss function and validation metric over the training epochs for Subjects 1 and 5.

Figure 7

Comparison of segmentation results obtained by the proposed method with ground truth annotations and two alternative approaches postulated elsewhere (Subject 2, MR session 1).

Figure 8

Comparison of single kidney GFR estimates obtained based on mean signals calculated in manually or automatically annotated cortex regions.

Figure 9

Bland–Altman plots of agreement for automatically (left) and manually (right) determined kidney segments. Measurements were evaluated against normality using Shapiro–Wilk test.

Figure 10

Cross section of the left kidney (Subject 1, examination session 1) and its corresponding segmentation result (solid border lines) overlaid on the manual annotation (semi-transparent fill). White arrows indicate false classifications made by the proposed method potentially due to erroneous image registration. Red arrow shows an example of a segmentation error due to partial volume effect.