Accuracy and processing time of kidney volume measurement methods in rodents polycystic kidney disease models: superiority of semiautomated kidney segmentation

Abstract

Measurement of total kidney volume (TKV) using magnetic resonance imaging (MRI) is a valuable approach for monitoring disease progression in autosomal dominant polycystic kidney disease (PKD) and is becoming more common in preclinical studies using animal models. Manual contouring of kidney MRI areas [i.e., manual method (MM)] is a conventional, but time-consuming, way to determine TKV. We developed a template-based semiautomatic image segmentation method (SAM) and validated it in three commonly used PKD models: Cys1^cpk/cpk mice, Pkd1^RC/RC mice, and Pkhd1^pck/pck rats (n = 10 per model). We compared SAM-based TKV with that obtained by clinical alternatives including the ellipsoid formula-based method (EM) using three kidney dimensions, the longest kidney length method (LM), and MM, which is considered the gold standard. Both SAM and EM presented high accuracy in TKV assessment in Cys1^cpk/cpk mice [interclass correlation coefficient (ICC) ≥ 0.94]. SAM was superior to EM and LM in Pkd1^RC/RC mice (ICC = 0.87, 0.74, and <0.10 for SAM, EM, and LM, respectively) and Pkhd1^pck/pck rats (ICC = 0.59, <0.10, and <0.10, respectively). Also, SAM outperformed EM in processing time in Cys1^cpk/cpk mice (3.6 ± 0.6 vs. 4.4 ± 0.7 min/kidney) and Pkd1^RC/RC mice (3.1 ± 0.4 vs. 7.1 ± 2.6 min/kidney, both P < 0.001) but not in Pkhd1^PCK/PCK rats (3.7 ± 0.8 vs. 3.2 ± 0.5 min/kidney). LM was the fastest (∼1 min) but correlated most poorly with MM-based TKV in all studied models. Processing times by MM were longer for Cys1^cpk/cpk mice, Pkd1^RC/RC mice, and Pkhd1^pck.pck rats (66.1 ± 7.3, 38.3 ± 7.5, and 29.2 ± 3.5 min). In summary, SAM is a fast and accurate method to determine TKV in mouse and rat PKD models.

NEW & NOTEWORTHY Total kidney volume (TKV) is a valuable readout in preclinical studies for autosomal dominant and autosomal recessive polycystic kidney diseases (ADPKD and ARPKD). Since conventional TKV assessment by manual contouring of kidney areas in all images is time-consuming, we developed a template-based semiautomatic image segmentation method (SAM) and validated it in three commonly used ADPKD and ARPKD models. SAM-based TKV measurements were fast, highly reproducible, and accurate across mouse and rat ARPKD and ADPKD models.

INTRODUCTION

In polycystic kidney diseases (PKDs), total kidney volume (TKV), determined by magnetic resonance imaging (MRI) or computed tomography (CT), reflects the severity of renal cystic burden. However, the utility of TKV for assessing disease progression and predicting its outcomes differs among the two major PKD forms.

In autosomal dominant PKD (ADPKD), TKV typically increases progressively over life (1). Since TKV changes in ADPKD manifest even before measurable changes in renal function (2), TKV-based indexes become invaluable for identifying patients at risk for severe renal disease progression and candidacy for therapeutic interventions (3). The recent approval of TKV as an enrichment biomarker for ADPKD clinical trials by the United States Food and Drug Administration (FDA) (4) has accelerated the development of ADPKD therapeutic strategies, including the first FDA-approved ADPKD therapy (3).

In autosomal recessive PKD (ARPKD), after an initial increase, TKV growth may cease and even decline over time (1). In contrast to ADPKD where isolated cysts originate from a relatively small number of nephrons and progressively enlarge, ARPKD is characterized by ectatic dilation of the collecting ducts affecting most, if not all, nephrons. These ectatic dilations typically do not continue to increase over time, whereas fibrotic changes become more prominent. These ARPKD-specific characteristics complicate the use of TKV in monitoring or predicting its renal outcomes. Currently, there are no FDA-approved biomarkers or therapies for the treatment of ARPKD.

MRI has emerged as the preferred method for evaluating renal cystic burden in PKDs. This mostly operator-independent method is relatively safe and can accurately determine TKV, even for very large kidneys. An additional advantage is that three-dimensional (3-D) visualization provides high contrast between the region of interest (ROI) and the surroundings (5, 6). A high ROI is required for kidney segmentation and subsequent TKV measurement.

The most accurate TKV measurement method is based on manual contouring of the kidney boundary in a 3-D image, a tedious and time-consuming process. To address this limitation, several automatic and semiautomatic methods were developed for human kidney segmentation from MRI images (7, 8). Since their use is still limited mostly to highly specialized centers, TKV assessment in clinical practice often relies on alternative methods such as estimating TKV based on the ellipsoid formula or using the longest kidney length (8, 9). More recently, the use of machine learning has opened the opportunity for fully automated TKV calculation (10).

In contrast to patients with PKD, the renal cystic disease progresses relatively rapidly in most animal models of PKD, often resulting in death within weeks or months. In these models, cyst burden and TKV increase even before measurable changes in renal function and TKV usually continues to increase progressively over the animals’ lifespan (e.g., as reported in Ref. 11). Therefore, TKV has become an important readout to monitor drug efficacy on disease progression in preclinical studies.

Similar to clinical ADPKD trials, TKV in animal models can be determined before and after experimental interventions using noninvasive imaging modalities, such as MRI. This approach may substantially improve the power of such preclinical studies [e.g., reduce the number of studied animals, cost, and length of the study (12) compared with a single posttreatment timepoint assessment of total kidney weight or kidney length that is usually obtained at the end of the study when experimental animals are euthanized (13, 14)]. The pre- and posttreatment TKV comparisons seem especially relevant to models with high phenotypic variability, such as the Pkd1^RC/RC mouse (15). Despite its advantages, the adoption of TKV measurement in animal models by the broader scientific community is limited by the lack of access to relevant imaging and image analysis expertise.

To simplify the process of TKV measurement in animal models, we developed a new template-based semiautomatic method (SAM) for kidney image segmentation in mice and rats. We evaluated its performance compared with the conventional manual method (MM), TKV estimate based on the ellipsoid formula method (EM), and longest kidney length method (LM). We tested the accuracy and time requirements of these methods across three commonly used animal PKD models to guide the use of TKV assessment methods in small animal studies.

METHODS

Animal Models and Small Animal MRI

We studied three commonly used animal models of PKD: Pkd1^RC/RC mice and Pkhd1^pck/pck rats (to model renal ADPKD-like phenotype) and Cys1^cpk/cpk mice (an ARPKD phenocopy) (15–17). We studied 10 animals per model; all animals were male. Pkd1^RC/RC mice were used at 14 wk of age, Pkhd1^pck/pck rats at 10 wk of age, and Cys1^cpk/cpk mice at 21 days of age. Animals were anesthetized with isoflurane, positioned on a body temperature-regulating bed in the prone position, and imaged in the axial view using a 9.4-T small animal MRI scanner (Bruker) with a surface coil (Bruker BioSpin, Billerica, MA) placed on top of their body above the kidneys. A T2-weighted fast spin echo sequence (rapid acquisition with relaxation enhancement) was used to image the entire kidney region with a repetition time/echo time = 5,000/50 ms, field of view = 32.5 × 15 mm, number of excitations = 8, frequency/phase encoding = 278/128, flip angle = 180°, and slice thickness and spacing between slices were 0.5 mm for mice and 1 mm for rats. All protocols followed the National Institutes of Health Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Animal Care and Use Committee of the University of Alabama at Birmingham. The University of Alabama at Birmingham is fully accredited by the Association for Assessment and Accreditation of Laboratory Animal Care International.

Image Analyses

For each animal, TKV was measured (comprising the volumes of both the left and right kidneys, 10 animals per model). MM, which involves the manual outlining of the kidneys at each MRI image, is broadly recognized as a gold standard for 3-D kidney volumetrics. Therefore, in our analyses, we used this method as a reference to which we compared the other methods of analysis (Fig. 1). Specifically, we drew the line along the edge of the kidney in each image slice using ImageJ (National Institutes of Health, Bethesda, MD), a freely available software. The binary image of the kidney ROI was then created, and TKV was calculated by summing all voxels within the ROI.

Figure 1.

The three analyzed methods for kidney volume assessment in small animal polycystic kidney disease models. A: schema of the manual method (MM), ellipsoid formula method (EM), and semiautomated method (SAM). In each image, the dark gray oval represents the kidney that is being measured. In MM, the individual layers of the kidney, which are manually outlined (represented by the lighter gray overlays shown filling the entire kidney). The area of each of these individual layers is calculated using manual outlines, and total kidney volume (TKV) is calculated as a sum of all these layers (see provided equation). EM estimates TKV based on the measurement of two perpendicular lines drawn through the center image slice within the kidney. These measurements [D₁ and D₂ (corresponding to kidney width and depth) as well as D₃ (corresponding to kidney length)] are used to calculate TKV (see provided equation). SAM uses the imaging software shown. The lighter gray outline represents the three-dimensional reference template, which is adjusted in each plane to accurately fit the dimensions of the kidney being measured. The adjustments are illustrated by the blue arrows and the change in size of the light gray overlay. B: representative central kidney regions of the three studied animal models with the kidney boundaries determined either by MM, EM, or SAM. In this study, the region determined by MM was considered a reference to which we compared the performance of the other tested methods. The three studied models represent distinct patterns of renal cyst distribution; most renal tissue parenchyma was preserved in the Pkhd1^pck/pck rat and Pkd1^RC/RC mouse models, and most of the renal tissue was replaced by cysts in the Cys1^cpk/cpk mouse model. Specifically, in Pkhd1^pck/pck rats, by 10 wk of age, most cysts localized to the corticomedullary junction or medulla. In Pkd1^RC/RC mice (C57BL6/J genetic background), the renal cystic burden was mild at 14 wk of age, with mostly small, isolated cysts localized to both the medulla and cortex. In contrast, in Cys1^cpk/cpk mice, by 21 days of postnatal age, nearly all renal tissue parenchyma was replaced by cystically dilated tubules across the cortex and medulla. Scale bar = 1 mm.

We also estimated TKV using EM (18), which is commonly used in clinical practice as an alternative approach to measure TKV (18). Specifically, in the central image slice of the kidney region, the longest line was drawn in the kidney, and another line perpendicular to the longest line was drawn (Fig. 1). To ensure the two lines were perpendicular, the angle of each line was measured. The lengths of the two lines were defined as D₁ and D₂, and kidney volume was calculated using the following ellipsoid formula: D₁ × D₂ × D₃ × π/6, where D₃ was defined as the number of image slices multiplied by the slice thickness. D₃ was used as the maximum longitudinal length of the kidney, which corresponded with LM [used in patients with ADPKD (9) and PKD models (13)], which we also explored in this study.

As an alternative to these methods, we developed SAM based on template images specific to each animal model (Fig. 1 and Fig. 2). Such an approach addressed the morphological differences in renal cystic disease across the studied PKD models. A total of 10 MRI images were used per animal model to create the template image and were not used in subsequent analyses. The binary images of kidneys were resampled and coregistered using similarity transformation. After that, both the left and right kidney regions were manually segmented, and the binary images of kidneys were coregistered using similarity transformation (15). Finally, all coregistered binary images were averaged and saved as the template image. Image coregistration and template image creation were automatically implemented using a laboratory-made software package in LabView (National Instrument, Austin, TX).

Figure 2.

The front panel of a software package for semiautomated segmentation of a three-dimensional left kidney region. A: kidney images at axial, coronal, and sagittal views overlapped with the reference template image immediately after the template reference image was loaded. The boundary of the template reference is indicated with a white dotted line at each view, whereas that of the measured kidney is indicated with a red dotted line. The reference image was transformed to match the boundary of the kidney image more accurately by conducting translation (B), rotation (C), and scaling (D) to match with the actual kidney region. Specifically, an adjustment in the axial plane (B) moved the kidney image into the center of the template reference image (see the gray overlay). Adjustment in the coronal plane rotated the reference image to overlay the kidney image (C) even more accurately (as indicated by the gray overlays). Finally, an adjustment in the sagittal plane (D) scaled the template reference image to further enhance the match of the kidney image overlay to the measured kidney shape (as indicated by the gray overlays in all three planes). Adjustments made in each plane can be seen on the scale bars below each image and are highlighted by a red outline and arrow.

Image processing for the template-based semiautomatic segmentation of a 3-D kidney image was done as follows. First, the template image and kidney magnetic resonance image were loaded into the LabView program. After that, the boundaries of the template and kidney images loaded into the software were indicated with dotted lines: white for the template reference and red for the kidney (Fig. 2A). First, the template image was translated using three slide bars via X, Y, and/or Z directions (Fig. 2B). After that, the template image was rotated in the axial, coronal, and sagittal views (Fig. 2C). Finally, the size of the template image was adjusted, overlaying with the kidney image (Fig. 2D). TKV was calculated by summing all voxels within the segmented region and multiplying them by unit voxel volume. TKV was calculated automatically using LabView-based software made in our laboratory.

To assess the reproducibility of the studied methods, two observers were trained to analyze TKV from the T2-weighted MRI images with MM, EM, SAM (Fig. 1), and LM. Subsequently, they independently applied these methods to analyze all studied measurements. The results of two observers were used for statistical analysis.

Assessment of Processing Time for TKV Measurement Methods

To compare the time necessary to calculate TKV with each of the studied methods, the amount of time required to determine TKV was recorded for MM, EM, LM, and SAM in each of the three studied PKD models. This recorded time included the time required to determine TKV of both (left and right) kidneys; these data were collected from each of the 10 animals studied in each model. For MM, the time included loading the MRI images, outlining the kidney, and calculating TKV based on these measurements. Similarly, EM time included image loading, determination of the middle MRI slice, drawing of perpendicular lines, and calculation of TKV. For LM, the time included loading the images and counting the number of MRI images with visible kidney tissue. For our SAM, time to complete included loading images, adjusting the template image to the desired size and shape, and calculating TKV. The average time to complete each method with each model was calculated, and all methods were compared.

Statistical Analyses

The accuracy of TKV estimation by each method was determined using the Pearson correlation coefficient (r); TKV determined manually using the MM approach was used as a reference (19). We also used the interclass correlation coefficient (ICC) to determine the data agreement between the method under test and MM (20). In addition, for analyses of image processing time (time required for image analysis of both left and right kidneys for each animal), we used a one-way ANOVA to determine the statistical difference between the groups (21). A P value lower than 0.05 was considered significant statistically. These analyses were performed using SAS (v. 9.4, SAS Institute, Cary, NC). Bland–Altman plots were generated using R (v. 3.6.3).

RESULTS

MRI Was Safe Across Studied Mouse and Rat PKD Models Representing Different Degrees of Renal Cystic Burden

We used MRI to assess TKV in three murine PKD models. Pkd1^RC/RC mice (14 wk old) and Pkhd1^pck/pck rats (10 wk old) represented models with mild-moderate renal cystic burden and Cys1^cpk/cpk mice (21 days old) represented rapidly progressing models with severe renal cystic phenotype with nearly homogenous kidney enlargement due to severe ectatic dilatation of mostly collecting ducts (see examples in Fig. 1B). In our experiments of 10 animals from each model, we did not observe any short-term or long-term adverse effects that can be attributed to the MRI procedure across the two animal species, animals’ age (as noted above), body size/weight, and phenotypic manifestations.

Accuracy of MRI-Based TKV Measurement Methods

In Cys1^cpk/cpk mice, both SAM and EM yielded excellent intraclass correlation (i.e., data agreement) with MM for the TKV assessment (ICC = 0.97 and 0.94, respectively; see Table 1 for details). However, in Pkd1^RC/RC mice, SAM was superior to EM (excellent ICC = 0.87 vs. moderately good ICC = 0.74, respectively). In Pkhd1^pck/pck rats, SAM was also superior to EM (moderately good ICC = 0.59 vs. ICC < 0.10, which indicates no reliability among raters). LM had ICC < 0.10 for all three studied models. Similarly strong correlations for SAM and EM (vs. MM) were found in the Pkd1^RC/RC mouse model (r = 0.98 and 0.97, both P < 0.001). However, in Pkhd1^pck/pck rats, SAM was superior to EM (strong r = 0.81 with P < 0.001 vs. moderately strong r = 0.63 with P = 0.003). Compared with SAM and EM, the LM-based correlation with MM was lower for Cys1^cpk/cpk and Pkd1^RC/RC mice (r = 0.71 and 0.89, both P < 0.001). In Pkhd1^pck/pck rats, LM values correlated poorly with those obtained by MM (r = 0.30, P = 0.202). The comparison of agreement between TKVs obtained with EM versus MM and SAM versus MM for each of the three studied PKD models was also depicted with Bland–Altman plots (Fig. 3); plots for left versus right kidney volumes showed similar patterns (Supplemental Figs. S1 and S2).

Table 1.

Comparison of three methods for total kidney volume assessment in three polycystic kidney disease animal models

Model	EM vs. MM			LM vs. MM			SAM vs. MM
	ICC	r	P	ICC	r	P	ICC	r	P
Cys1cpk/cpk	0.97	0.99	<0.001	<0.10	0.71	<0.001	0.94	0.97	<0.001
Pkd1RC/RC	0.74	0.97	<0.001	<0.10	0.89	<0.001	0.87	0.98	<0.001
Pkhd1pck/pck	<0.10	0.63	0.003	<0.10	0.30	0.202	0.59	0.81	<0.001

EM, ellipsoid formula method; ICC, interclass correlation coefficient; LM, length method; MM, manual method; SAM, semiautomated method.

Figure 3.

Bland–Altman plots for the agreement between the ellipsoid formula method (EM) vs. manual method (MM) and semiautomated method (SAM) vs. MM for total kidney volume (TKV). In all graphs, the x-axis represents mean TKV (combined volumes or left and right kidneys) obtained with EM vs. the MM reference (left) and SAM vs. the MM reference (right) in mm³. The y-axis represents the difference between TKVs obtained with EM vs. MM (left) and SAM vs. MM (right) in mm³. The horizontal solid line shows the mean bias indicating the average under- or overestimation of EM or SAM vs. the MM reference. The lower and upper horizontal dotted lines represent the limits of agreement (average difference ± 1.96 SD). Finally, the regression line of difference is represented by a dashed line.

Processing Times for the Measurement of TKV

Processing times for measuring TKV using MM, the gold standard organ volume measurement, in Cys1^cpk/cpk, Pkd1^RC/RC, and Pkhd1^pck/pck models were on average 66.1 ± 7.3, 38.3 ± 7.5, and 29.2 ± 3.5 min (means ± SD, respectively; see details in Table 2). In comparison, TKV measurement using our newly developed SAM was 3.6 ± 0.6, 3.1 ± 0.4, and 3.7 ± 0.8 min, respectively. This corresponds to the reduction of processing times using SAM (vs. MM) by 95% in the Cys1^cpk/cpk mouse model, 92% in the Pkd1^RC/RC mouse model, and 87% in the Pkhd1^pck/pck rat model (all P < 0.001). Processing times for obtaining kidney length, width, and depth for estimating TKV using EM in Cys1^cpk/cpk, Pkd1^RC/RC, and Pkhd1^pck/pck models were, on average, 4.4 ± 0.7, 7.1 ± 2.6, and 3.2 ± 0.5 min, respectively. In addition, EM reduced processing times (vs. MM) by 93% in the Cys1^cpk/cpk mouse model, 81% in the Pkd1^RC/RC mouse model (both P < 0.001), and 87% in the Pkhd1^pck/pck rat model (all P < 0.001). The time required for assessing kidney length only using LM was even shorter: 0.9 ± 0.1, 0.9 ± 0.2, and 0.8 ± 0.1 min for the measurement of each kidney from the Cys1^cpk/cpk, Pkd1^RC/RC, and Pkhd1^pck/pck model, respectively, although the accuracy of LM was substantially inferior to SAM and EM.

Table 2.

Comparison of kidney segmentation methods based on processing time

Model	Time, min				P Value
	MM	EM	LM	SAM	MM vs. EM	MM vs. LM	MM vs. SAM	EM vs. LM	EM vs. SAM	LM vs. SAM
Cys1cpk/cpk	66.1 ± 7.3	4.4 ± 0.7	0.9 ± 0.1	3.6 ± 0.6	<0.001	<0.001	<0.001	<0.001	0.017	<0.001
Pkd1RC/RC	38.3 ± 7.5	7.1 ± 2.6	0.9 ± 0.2	3.1 ± 0.4	<0.001	<0.001	<0.001	<0.001	<0.001	<0.001
Pkhd1pck/pck	29.2 ± 3.5	3.2 ± 0.5	0.8 ± 0.1	3.7 ± 0.8	<0.001	<0.001	<0.001	<0.001	0.876	<0.001

EM, ellipsoid formula method; LM, length method; MM, manual method; SAM, semiautomated method.

DISCUSSION

To our knowledge, the present study provides the first comprehensive assessment of the reliability and accuracy of several methods for TKV assessment in animal models. The conventional MM approach that we used as a reference for the other TKV assessment methods is broadly recognized as the most accurate. However, the time- and labor-intensive nature of this approach contributes excessively to the overall cost of preclinical PKD studies relying on TKV.

The EM-based TKV estimation (18), a commonly used approach in a clinical setting (3), also performed relatively well in our study. Its accuracy and reproducibility were good in the Cys1^cpk/cpk and Pkd1^RC/RC mouse models. However, its accuracy was only moderate and its reproducibility was poor in the Pkhd1^pck/pck rat model, partly due to the asymmetric kidney shape in this model. Therefore, it seems that although EM might be used as a faster alternative to MM in mouse PKD models, its use in rat PKD models might not be reliable [similarly, EM had limited performance vs. MM in patients with ADPKD (22)]. LM, used in some clinical practices to assess eligibility for ADPKD therapy, required the least time. However, its accuracy was inferior to the other methods reported in this study. Whether LM is more appropriate for large animal PKD models remains to be determined.

SAM, which we developed to address the MM limitations, outperformed MM alternatives on nearly all levels. Its accuracy was outstanding in Cys1^cpk/cpk and Pkd1^RC/RC mice, and it performed best in Pkhd1^pck/pck rats among the studied methods. SAM also outperformed EM in processing time in Cys1^cpk/cpk and Pkd1^RC/RC mice. SAM was more accurate than EM because it uses a template image reference adjusted to fit the kidney shape, accounting for PKD model-specific asymmetries and shape variations. SAM thus emerged from this study as the best, fastest alternative to MM in mouse and rat models of PKD. Compared with MM, the improved time efficiency of SAM-based TKV assessment might reduce the image-analysis cost (potentially allowing evaluation of additional animals or timepoints). Compared with EM, the improved TKV assessment accuracy offered by SAM might also decrease the study cost by reducing the number of animals needed to achieve statistical significance or allowing earlier detection of TKV changes.

The present study also offers insights into the potential benefits and limitations of these methods in additional mouse and rat PKD models. For example, our experiments in the Cys1^cpk/cpk mouse may be applicable to other models in which cystic dilation occurs in most nephrons and cysts account for most of TKV, giving the kidney a symmetrical shape [e.g., conditional mouse models with an early-induced defect in cystogenic genes such as polycystin 1 (Pkd1), polycystin 1 (Pkd2), or intraflagellar transport 88 (Ift88)]. A quick TKV assessment by SAM or EM in these models might be especially beneficial in the pretreatment evaluation of rapidly progressing models with early mortality.

In contrast, our analyses demonstrated that SAM performed better than EM in models with the oblong, more physiological, kidney shapes since EM requires additional time to accurately measure the dimensions of the kidney and this extends the time to analysis. Therefore, SAM may be the superior approach to the measurement of TKV for PKD models in which the contribution of cyst volume to the TKV is less prominent, including the Pkhd1^pck/pck rat (a model used in the development of tolvaptan, the only FDA-approved therapeutic for ADPKD), Pkd1^RC/RC mouse (one of the most commonly used orthologous PKD models), and PKD mouse models with a slower pace of renal cystic disease progression triggered by a late induction of PKD using a conditional defect in cystogenic genes. When the shape of the kidney is less spherical and more irregular, the application of the 3-D template may allow for a more accurate TKV estimate. In contrast, when the kidney shape is more spherical, even EM can estimate TKV accurately.

Also, SAM might be up to 50% faster (vs. EM, as suggested by our analyses in Pkd1^RC/RC mice) because the template image used by SAM is based on the kidney shape of the studied model itself.

Follow-up studies might include comparing TKV image-analysis methods in life animals versus ex vivo (externalized) kidneys when there are no confounding issues with motion. Also, the present study has to be validated on females since it was based only on males, and the male sex is associated with more severe renal manifestations across several PKD models. Similarly, the present study assessed TKVs at only one timepoint. Outcomes might differ in younger or older animals, as renal cystic disease typically progresses over time.

Although our study points to the need to standardize how images are analyzed in preclinical studies, image acquisition strategies and hardware can also play a significant role in the quality and accuracy of the subsequent measurements. How much these measured differences between TKV measurements from these techniques depend on image acquisition parameters remains to be determined. A possible solution to further standardize such analyses across different MRI scanners and acquisition parameters would be using a physical imaging reference or standard, such as a point-of-care fantom that can be imaged together with tested subjects in MRI scanners.

In summary, the most accurate method for TKV measurement is to manually outline areas of kidney tissue in each image slice (MM). However, this approach is also the most time-consuming. In the present study, we demonstrated that our newly developed template-based SAM could determine TKV with comparable or better accuracy to manual segmentation in significantly less time. The reproducibility of SAM was outstanding in the results of mouse PKD models and moderately good in the Pkhd1^pck/pck model, exceeding the performance of the other studied methods. We propose that standardization of methods for the analysis of magnetic resonance images to determine TKV should be based on the animal models being used in the study.

DATA AVAILABILITY

Data will be made available upon reasonable request.

SUPPLEMENTAL DATA

Supplemental Figs. S1 and S2: https://doi.org/10.6084/m9.figshare.21950933.v1.

GRANTS

This work was supported by the National Institutes of Health-funded PKD Research Resource Consortium (Grant U54DK126087), the Office of Research and Development, Medical Research Service, Department of Veterans Affairs (Grant 1-I01-BX004232-01A2), and the Detraz Endowed Research Fund in Polycystic Kidney Disease (to M.M.).

DISCLOSURES

M.M. reports grants and consulting fees outside the submitted work from Otsuka Pharmaceuticals, Sanofi, Palladio Biosciences, Reata, Natera, Chinook Therapeutics, Goldilocks Therapeutics, and Carraway Therapeutics. D.P.W. has grant support from Calico Laboratories. None of the other authors has any conflicts of interest, financial or otherwise, to disclose.

AUTHOR CONTRIBUTIONS

M.C.D., B.K.Y., H.K., and M.M. conceived and designed research; M.C.D., S.M., R.R., J.Z., and P.C. performed experiments; M.C.D., E.M., B.K.Y., H.K., and M.M. analyzed data; M.C.D., E.M., D.P.W., F.Q., P.C.H., B.K.Y., H.K., and M.M. interpreted results of experiments; M.C.D., E.M., B.K.Y., H.K., and M.M. prepared figures; M.C.D., B.K.Y., H.K., and M.M. drafted manuscript; M.C.D., E.M., D.P.W., F.Q., P.C.H., B.K.Y., H.K., and M.M. edited and revised manuscript; M.C.D., S.M., R.R., J.Z., P.C., E.M., D.P.W., F.Q., P.C.H., B.K.Y., H.K., and M.M. approved final version of manuscript.