Improving Automatic Renal Segmentation in Clinically Normal and Abnormal Paediatric DCE-MRI via Contrast Maximisation and Convolutional Networks for Computing Markers of Kidney Function

Abstract

There is a growing demand for fast, accurate computation of clinical markers to improve renal function and anatomy assessment with a single study. However, conventional techniques have limitations leading to overestimations of kidney function or failure to provide sufficient spatial resolution to target the disease location. In contrast, the computer-aided analysis of dynamic contrast-enhanced (DCE) magnetic resonance imaging (MRI) could generate significant markers, including the glomerular filtration rate (GFR) and time-intensity curves of the cortex and medulla for determining obstruction in the urinary tract. This paper presents a dual-stage fully modular framework for automatic renal compartment segmentation in 4D DCE-MRI volumes. (1) Memory-efficient 3D deep learning is integrated to localise each kidney by harnessing residual convolutional neural networks for improved convergence; segmentation is performed by efficiently learning spatial-temporal information coupled with boundary-preserving fully convolutional dense nets. (2) Renal contextual information is enhanced via non-linear transformation to segment the cortex and medulla. The proposed framework is evaluated on a paediatric dataset containing 60 4D DCE-MRI volumes exhibiting varying conditions affecting kidney function. Our technique outperforms a state-of-the-art approach based on a GrabCut and support vector machine classifier in mean dice similarity (DSC) by 3.8% and demonstrates higher statistical stability with lower standard deviation by 12.4% and 15.7% for cortex and medulla segmentation, respectively.

Keywords: DCE-MRI; GFR; MR urography; cortex; kidney; medulla; renal compartment; segmentation; time–intensity curve.

Figure A1

Graphs (a–f) represent the time–intensity curves of three clinically “normal” cases (scans or 4D volumes). The first column shows the relative contrast intensity enhancement of the automatic renal compartment segmentation in both the left and right kidneys over time (minutes). The second column shows the corresponding ground-truth (GT) time–intensity curves.

Figure A1

Graphs (a–f) represent the time–intensity curves of three clinically “normal” cases (scans or 4D volumes). The first column shows the relative contrast intensity enhancement of the automatic renal compartment segmentation in both the left and right kidneys over time (minutes). The second column shows the corresponding ground-truth (GT) time–intensity curves.

Figure A2

Graphs (a–f) represent the time–intensity curves of three clinically “abnormal” cases (scans or 4D volumes). The first column shows the relative contrast intensity enhancement of the automatic renal compartment segmentation in both the left and right kidneys over time (minutes). The second column shows the corresponding ground-truth (GT) time–intensity curves.

Figure A2

Graphs (a–f) represent the time–intensity curves of three clinically “abnormal” cases (scans or 4D volumes). The first column shows the relative contrast intensity enhancement of the automatic renal compartment segmentation in both the left and right kidneys over time (minutes). The second column shows the corresponding ground-truth (GT) time–intensity curves.

Figure 1

Overview of the proposed automatic kidney segmentation approach. The training stage simultaneously develops a network (3D Rb-UNet) for localising the organ and a segmentation network (3D FC-DenseNet) to predict the labels that correspond to kidney and non-kidney tissue. The testing stage processes an original scan (a 4D volume), performs a coarse segmentation to generate a bounding box capturing the main kidney region and then processes the cropped image volume to predict the labels of that organ.

Figure 2

Overview of the proposed automatic renal segmentation approach. Using the input 4D DCE-MRI series, Process 1 detects the individual left and right kidneys (if present) via the automatic kidney segmentation approach. For each identified kidney, Process 2 performs medulla and cortex segmentation for all 3D volumes in the 4D DCE-MRI series. Process 3 generates the resulting single “optimum” volumetric medulla and cortex segmentation.

Figure 3

Whole-kidney segmentation results in eight different DCE-MRI scans (4D volumes). Every column corresponds to one MRI volume. The first row displays a sample DCE-MRI coronal slice with the segmentation outcome (green) overlapping the ground truth (red) and dice similarity coefficient (DSC). The second row displays a 3D reconstruction of the kidney and DSC. (a) Segmentations in four clinically “normal” cases; (b) Segmentations in four clinically “abnormal” cases.

Figure 4

Box plots for two datasets depicting the medulla and cortex dice score coefficients (DSCs) for clinically “normal” and “abnormal” kidneys.

Figure 5

Medulla and cortex segmentation results of three different clinically “normal” kidneys. The first column (a,d,g) shows the results from the proposed approach; the second column (b,e,h) shows the respective results using the baseline approach from Yoruk et al. [16]; the third column (c,f,i) shows the respective ground truth.

Figure 6

Medulla and cortex segmentation results of three different clinically “abnormal” kidneys. The first column (a,d,g) shows the results from the proposed approach; the second column (b,e,h) shows the respective results using the baseline approach from Yoruk et al. [16]; the third column (c,f,i) shows the respective ground truth.

Figure 7

Graphs (a–f) represent clinically “normal” and “abnormal” cases (scans or 4D volumes), respectively. The relative contrast intensity enhancement of the (automatically segmented) medulla and cortex in both the left and right kidneys is shown over time (minutes).

Figure 8

Graphs (a–f) represent clinically “normal” and “abnormal” cases (scans or 4D volumes), respectively.The tracer concentration is shown over time for both the right (blue) and left (red) kidneys; the corresponding kidney percentage, volume (mL) and GFR (mL/min) were computed to discern and evaluate separate kidney functions.