diff --git a/paper/paper.bib b/paper/paper.bib index 8685b54..552bd46 100644 --- a/paper/paper.bib +++ b/paper/paper.bib @@ -28,6 +28,21 @@ @article{barroqueiro_bridging_2021 urldate = {2023-10-30} } +@article{bazin_computational_2014, + title = {A Computational Framework for Ultra-High Resolution Cortical Segmentation at 7 ~ {{Tesla}}}, + author = {Bazin, Pierre-Louis and Weiss, Marcel and Dinse, Juliane and Schäfer, Andreas and Trampel, Robert and Turner, Robert}, + date = {2014-06-01}, + journaltitle = {NeuroImage}, + shortjournal = {NeuroImage}, + series = {In-Vivo {{Brodmann Mapping}} of the {{Human Brain}}}, + volume = {93}, + pages = {201--209}, + issn = {1053-8119}, + doi = {10.1016/j.neuroimage.2013.03.077}, + url = {https://www.sciencedirect.com/science/article/pii/S1053811913003327}, + urldate = {2024-08-02} +} + @software{brett_nipy/nibabel_2019, title = {{{NiBabel}}}, shorttitle = {{{NiBabel}}}, @@ -75,6 +90,23 @@ @software{dawson-haggerty_trimesh_2023 version = {4.0.0} } +@article{fischl_freesurfer_2012, + title = {{{FreeSurfer}}}, + author = {Fischl, Bruce}, + date = {2012-08-15}, + journaltitle = {NeuroImage}, + shortjournal = {Neuroimage}, + volume = {62}, + number = {2}, + eprint = {22248573}, + eprinttype = {pmid}, + pages = {774--781}, + issn = {1095-9572}, + doi = {10.1016/j.neuroimage.2012.01.021}, + langid = {english}, + pmcid = {PMC3685476} +} + @article{francis_magnetic_2023, title = {Magnetic {{Resonance Imaging}} to {{Evaluate Kidney Structure}}, {{Function}}, and {{Pathology}}: {{Moving Towards Clinical Application}}}, shorttitle = {Magnetic {{Resonance Imaging}} to {{Evaluate Kidney Structure}}, {{Function}}, and {{Pathology}}}, @@ -89,6 +121,22 @@ @article{francis_magnetic_2023 langid = {english} } +@article{goebel_brainvoyager_2012, + title = {{{BrainVoyager}} — {{Past}}, Present, Future}, + author = {Goebel, Rainer}, + date = {2012-08-15}, + journaltitle = {NeuroImage}, + shortjournal = {NeuroImage}, + series = {20 {{YEARS OF fMRI}}}, + volume = {62}, + number = {2}, + pages = {748--756}, + issn = {1053-8119}, + doi = {10.1016/j.neuroimage.2012.01.083}, + url = {https://www.sciencedirect.com/science/article/pii/S1053811912001000}, + urldate = {2024-08-02} +} + @book{hall_guyton_2015, title = {Guyton and {{Hall Textbook}} of {{Medical Physiology}}}, author = {Hall, John E.}, @@ -101,6 +149,71 @@ @book{hall_guyton_2015 pagetotal = {1171} } +@article{huntenburg_laminar_2017, + title = {Laminar {{Python}}: Tools for Cortical Depth-Resolved Analysis of High-Resolution Brain Imaging Data in {{Python}}}, + shorttitle = {Laminar {{Python}}}, + author = {Huntenburg, Julia and Wagstyl, Konrad and Steele, Christopher and Funck, Thomas and Bethlehem, Richard and Foubet, Ophélie and Larrat, Benoit and Borrell, Victor and Bazin, Pierre-Louis}, + date = {2017-02-23}, + journaltitle = {Research Ideas and Outcomes}, + volume = {3}, + pages = {e12346}, + publisher = {Pensoft Publishers}, + issn = {2367-7163}, + doi = {10.3897/rio.3.e12346}, + url = {https://riojournal.com/article/12346/}, + urldate = {2024-08-02}, + langid = {english} +} + +@article{huntenburg_nighres_2018, + title = {Nighres: Processing Tools for High-Resolution Neuroimaging}, + shorttitle = {Nighres}, + author = {Huntenburg, Julia M and Steele, Christopher J and Bazin, Pierre-Louis}, + date = {2018-07-01}, + journaltitle = {GigaScience}, + shortjournal = {GigaScience}, + volume = {7}, + number = {7}, + pages = {giy082}, + issn = {2047-217X}, + doi = {10.1093/gigascience/giy082}, + url = {https://doi.org/10.1093/gigascience/giy082}, + urldate = {2024-08-02} +} + +@article{ishikawa_framework_2022, + title = {Framework for Estimating Renal Function Using Magnetic Resonance Imaging}, + author = {Ishikawa, Masahiro and Inoue, Tsutomu and Kozawa, Eito and Okada, Hirokazu and Kobayashi, Naoki}, + date = {2022-03}, + journaltitle = {Journal of Medical Imaging}, + shortjournal = {JMI}, + volume = {9}, + number = {2}, + pages = {024501}, + publisher = {SPIE}, + issn = {2329-4302, 2329-4310}, + doi = {10.1117/1.JMI.9.2.024501}, + url = {https://www.spiedigitallibrary.org/journals/journal-of-medical-imaging/volume-9/issue-2/024501/Framework-for-estimating-renal-function-using-magnetic-resonance-imaging/10.1117/1.JMI.9.2.024501.full}, + urldate = {2024-04-10} +} + +@article{jenkinson_fsl_2012, + title = {{{FSL}}}, + author = {Jenkinson, Mark and Beckmann, Christian F. and Behrens, Timothy E. J. and Woolrich, Mark W. and Smith, Stephen M.}, + date = {2012-08-15}, + journaltitle = {NeuroImage}, + shortjournal = {NeuroImage}, + series = {20 {{YEARS OF fMRI}}}, + volume = {62}, + number = {2}, + pages = {782--790}, + issn = {1053-8119}, + doi = {10.1016/j.neuroimage.2011.09.015}, + url = {http://www.sciencedirect.com/science/article/pii/S1053811911010603}, + urldate = {2020-12-10}, + langid = {english} +} + @article{li_renal_2020, title = {Renal {{BOLD MRI}} in Patients with Chronic Kidney Disease: Comparison of the Semi-Automated Twelve Layer Concentric Objects ({{TLCO}}) and Manual {{ROI}} Methods}, shorttitle = {Renal {{BOLD MRI}} in Patients with Chronic Kidney Disease}, @@ -192,6 +305,21 @@ @article{selby_assessment_2024 langid = {english} } +@article{stevens_assessing_2006, + title = {Assessing {{Kidney Function}} — {{Measured}} and {{Estimated Glomerular Filtration Rate}}}, + author = {Stevens, Lesley A. and Coresh, Josef and Greene, Tom and Levey, Andrew S.}, + date = {2006-06-08}, + journaltitle = {New England Journal of Medicine}, + volume = {354}, + number = {23}, + eprint = {16760447}, + eprinttype = {pmid}, + pages = {2473--2483}, + issn = {0028-4793}, + doi = {10.1056/NEJMra054415}, + url = {http://dx.doi.org/10.1056/NEJMra054415} +} + @article{yamashita_value_2015, title = {Value of Renal Cortical Thickness as a Predictor of Renal Function Impairment in Chronic Renal Disease Patients}, author = {Yamashita, Samia Rafael and family=Atzingen, given=Augusto Castelli, prefix=von, useprefix=false and Iared, Wagner and Bezerra, Alexandre Sérgio de Araújo and Ammirati, Adriano Luiz and Canziani, Maria Eugênia Fernandes and D'Ippolito, Giuseppe}, diff --git a/paper/paper.md b/paper/paper.md index ff8a29a..593851d 100644 --- a/paper/paper.md +++ b/paper/paper.md @@ -25,47 +25,52 @@ bibliography: paper.bib --- # Summary -Informative measurements of the kidneys structure and function can be performed using quantitative Magnetic Resonance Imaging (MRI) where each voxel of the image is a measurement of the physical properties of the tissue being image. Traditionally, analysis of these images is performed by segmenting the kidney into its constituent tissue types and calculating the average of each measurement for each tissue type. The process of segmenting renal tissue types is time consuming and inaccurate. -An alternative to tissue segmentation proposed by Pruijm _et al_ involves dividing the kidney into layers based on the distance of each voxel between the outer and inner surface of the kidney, a method known as Twelve Layer Concentric Objects (TLCO) [@piskunowicz_new_2015; @milani_reduction_2017; @li_renal_2020]. Layer based analysis only requires segmenting the whole kidney rather than the tissues within and is therefore quicker and more repeatable. The TLCO method does however have some limitation, it can only be performed on a single slice image, requires the image to be acquired at a specific angle to the kidneys, requires manual delineation of the outside and inside surface of the kidney, divides the kidney into the same number of layers irrespective of the size of the kidney and the software itself is closed source. -These limitations are addressed by `3DQLayers`, an open source Python package to automatically define 3D, multi-slice, renal layers of known thickness. +Quantitative Magnetic Resonance Imaging (qMRI) provides informative measurements of structure and function of an organ where each volumetric pixel (voxel) provides a measure of the physical properties of the underlying tissue. Traditionally, analysis of MR images is performed by first segmenting the organ and its constituent tissue types, which for the kidneys involves separating them into the cortex and medulla, before calculating the average measurement within each voxel. The process of segmenting renal tissue types is typically manual, making it time consuming and prone to inaccuracies. + +An alternative to voxel-based analysis in MRI is the layer model which divides the organ into ordered surfaces. For the kidney, this involves generating layers based on the distance of each voxel between the outer and inner surface of the kidney. From this, the gradient of change in qMRI measures between the cortex and medulla of the kidney can be computed to evaluate pathological and physiological aspects of the kidney. Here, `3DQLayers` an open-source Python software package to automatically define and interrogate 3D renal layers is presented. # Statement of need ## Background -The kidneys are a pair of structurally and functionally complex organs in the lower abdomen that participate in the control of bodily fluids by regulating the balance of electrolytes, excreting waste products of metabolism and excess water from blood to urine [@lote_principles_2012]. The each kidney is separated into two tissue types; cortical tissue located towards the outside of each organ, and medullary tissue arranged in small pyramids towards the centre of the organ [@hall_guyton_2015], as shown in \autoref{fig:renal_structure}. Quantitative MRI is the process of taking measurements where the value of each voxel has numerical significance, in physical units, based on the tissues underlying properties rather than simply representing signal intensity in arbitrary units. Example quantitative measurements include how readily water can diffuse through the tissue and the rate at which blood perfuses into the tissue. To help interpret quantitative images, regions of interest (ROI) are defined and statistical measures taken of the voxels within each region. +The kidneys are structurally and functionally complex organs in the abdomen responsible for the removal of waste products and excess fluid from the blood to produce urine [@lote_principles_2012]. Each kidney is separated into the cortex which forms the outer layer of the kidney and the medulla in the inner part which is arranged in a series of small pyramids [@hall_guyton_2015], as shown in \autoref{fig:renal_structure}. The kidney maintains homeostasis through filtration, reabsorption, secretion, and maintenance of the cortico-medullary gradient (CMG), meaning a method to assess changes in physiology from the cortex to the medulla is a key. + +Quantitative MRI (qMRI) goes beyond conventional MRI which primarily assesses signal intensity in a voxel in arbitrary units, by instead providing voxel-wise measurements with numerical significance, in physical units, based on the tissues underlying properties. For example, qMRI of relaxation times with parameters which carry information about the local microstructural, or those of how readily water can diffuse through the tissue and the rate at which blood perfuses the tissue. To interpret quantitative images, regions of interest (ROIs) for the renal cortex and medulla are defined in the tissue and statistical analysis performed on the voxels within each. Segmenting such ROIs manually is time consuming, and prone to intra- and inter-reader variation. + +The group of Pruijm proposed an alternative to tissue ROI based analysis termed the Twelve Layer Concentric Object (TLCO) method [@piskunowicz_new_2015; @milani_reduction_2017; @li_renal_2020] where users delineate the inner and outer boundaries of the kidney to generate twelve equidistant layers between the renal pelvis and the surface of the kidney. The outer layers represent the cortex and the inner layers the medulla, the gradient across central layers (the CMG) can also be computed. Since this layer-based analysis only requires segmenting the boundaries of the kidney rather than the cortex and medulla within it is quicker and more repeatable. An analogy to this is the development of layer-based analysis tools applied in the brain for neuroimaging including BrainVoyager [@goebel_brainvoyager_2012], CBSTools/Nighres [@bazin_computational_2014; @huntenburg_laminar_2017; @huntenburg_nighres_2018], FreeSurfer [@fischl_freesurfer_2012], and FSL [@jenkinson_fsl_2012]. -Segmenting ROI for the renal cortex and medulla manually is time consuming, difficult and prone to intra- and inter-reader variation thus decreasing the repeatability of measurements. Pruijm _et al_ proposed an alternative to tissue ROI based analysis in the Twelve Layer Concentric Object (TLCO) method [@piskunowicz_new_2015; @milani_reduction_2017; @li_renal_2020] where users delineate the inner and outer boundaries of the kidney to generate twelve equidistant layers between the renal pelvis and the surface of the kidney. The outer layers are analogous to the cortex; inner layers, the medulla; and gradient of the central layers, the cortico-medullary ratio. +However, the TLCO software is closed source and has some limitations. It requires manual delineation of the outside and inside surfaces of the kidney, divides the kidney into the same number of layers irrespective of the size of the kidney, and can only be performed on a single slice cutting through the kidneys on their longest axis (coronal-oblique) which is not always desirable [@bane_consensus-based_2020]. Due to the spatial distribution of kidney pathology, researchers prefer to acquire multi-slice images for full 3D coverage of the kidney to increase the number of voxels sampled and gain a better understanding of the heterogeneity of the kidney. An automated-TLCO method has been proposed [@ishikawa_framework_2022], however this work highlights difficulties analysing small kidneys due to the limited layer thickness and is only applied to water images from a Dixon protocol. -TLCO requires the MR image to be a single slice cutting through the kidneys on their longest axis (coronal-oblique) however, this is not always desirable [@bane_consensus-based_2020]. Often researchers prefer to acquire multi-slice images to increase the number of voxels in the image and gain a better understanding of the heterogeneity of the kidney. Additionally flexibility in the orientation images are acquired at is highly desirable. These limitations of TLCO were the motivation for the development of `3DQLayers`, a volumetric, quantitative-depth based analysis method for renal MRI data. +The motivation of `3DQLayers` was to address these limitations of TLCO to provide an open-source Python package to automatically define 3D, multi-slice, layers in the kidney of known thickness for quantitative-depth based analysis of kidney MRI data, enabling its use in large studies. ## Methods -`3DQLayers` is an open-source Python package that aims to build upon the premise of TLCO and allow layer based analysis to be fully automated for use in large studies. `3DQLayers` fundamentally differs from TLCO in that layers are defined based on each voxels distance from the surface of the kidney in millimetres rather than the proportion through the kidney. As such, the input to `3DQLayers` is a whole kidney ROI; this can be automatically generated from a structural image [@daniel_automated_2021; @daniel_renal_2024]. +`3DQLayers` is an open-source Python package building on the ideas within TLCO, with the fundamental difference that the layers are defined based a voxels’ distance from the surface of the kidney in millimetres rather than the proportion of the kidney for 3D analysis. As such, the input to `3DQLayers` is a whole kidney mask, which can be automatically generated from a structural image (e.g. using a U-net applied to T~2~-weighted images [@daniel_automated_2021; @daniel_renal_2024]). -The pipeline by which layers are defined is outlined in \autoref{fig:flowchart}. Pre-processing steps fill in holes in the ROI caused by cysts as the surface of a cyst is not representative of the surface of the kidney. Next the voxel-based representation of the ROI is converted to a smoothed mesh-based representation of the kidneys, the distance from the centre of each voxel to the surface of the mesh can then be calculated producing a depth map [@dawson-haggerty_trimesh_2023]. As tissue adjacent to the renal pelvis is not representative of the medulla, it is excluded from layer-based analysis. This is achieved by automatically segmenting the pelvis then calculating the distance from each voxel to the pelvis as above. Voxels closer than a specified threshold, typically 10 mm, are excluded from the depth map. Finally, a layer image is generated by quantising the depth map to a desired layer thickness, typically 1 mm. +The pipeline for defining the layers from the whole kidney mask is outlined in \autoref{fig:flowchart}. Pre-processing steps fill in holes in the kidney mask caused by cysts as the surface of a cyst is not characteristic of the surface of the kidney. Next the voxel-based representation of the ROI is converted to a smoothed mesh-based representation of the kidneys, the distance from the centre of each voxel to the surface of the mesh is then calculated to produce a depth map [@dawson-haggerty_trimesh_2023]. Tissue adjacent to the renal pelvis that is not representative of the medulla is then excluded from layer-based analysis. This is achieved by automatically segmenting the pelvis then calculating the distance from each voxel to the pelvis using the method described as above. Voxels closer than a specified threshold, typically 10 mm, are then excluded from the depth map. Finally, a layer image is generated by quantising the depth map to a desired layer thickness, typically 1 mm. -The layer image and quantitative images are resampled to the same resolution using `NiBabel` [@brett_nipy/nibabel_2019], this allows each layer to be used as an ROI with statistical measures of the quantitative image e.g. median, standard deviation and kurtosis, calculated as a function of depth through the kidney. The gradient of the central layers can be calculated, additionally, if tissue ROI are available the distribution of tissue types with depth can be explored. As the layers are generated from a structural image rather than the quantitative map, using `3DQLayers` stipulates no requirements on quantitative map acquisition, unlike TLCO. +The layer image and quantitative images are resampled to the same spatial resolution using `NiBabel` [@brett_nipy/nibabel_2019], to allow each layer to be used as an ROI to interrogate each qMR image with statistical measures (e.g. median, standard deviation and kurtosis) across the depth of the kidney. The gradient of the central layers can be calculated to estimate the CMG in qMRI metrics. Additionally, if cortex and medulla ROI are available the distribution of tissue types across layer depth can be explored. The left and right kidney can be analysed separately or combined. As the layers are generated from a structural image rather than the quantitative map, using `3DQLayers` stipulates no requirements on quantitative map acquisition, unlike TLCO. -An object oriented interface makes it easy for users to generate layers and use them to analyse quantitative images. [Documentation](https://qlayers.readthedocs.io/) is provided to guide users through instillation via `PyPI`, `conda` or from [source code on GitHub](https://github.com/alexdaniel654/qlayers) and includes tutorials and an API reference. An automated test suite with high coverage gives users confidence in the stability of `3DQLayers` and that there will be no unexpected changes to results unless highlighted in the change-log. +An object-oriented interface is provided to allow end users to simply generate layers and apply these to quantitative MR images. [Documentation](https://qlayers.readthedocs.io/) is provided to guide users through installation via `PyPI`, `conda` or from [source code on GitHub](https://github.com/alexdaniel654/qlayers) and includes tutorials and an API reference. An automated test suite with high coverage provides users with confidence in the stability of `3DQLayers` and that there will be no unexpected changes to results unless highlighted in the change-log. ## Usage Examples -An estimated glomerular filtration rate (eGFR) above 90 ml/min/1.73m^2^ is considered healthy. \autoref{fig:egfr_gradients} shows `3DQLayers` being used to measure different gradients of R~2~^\*^ in volunteers with normal and impaired renal function. This replicates results shown using TLCO however `3DQLayers` controls for kidneys size and as such the gradients are measured in quantitative units of Hz/mm rather than Hz/layer as in TLCO. +\autoref{fig:egfr_gradients} shows `3DQLayers` being used to measure different gradients of R~2~^\*^ in a heathy volunteer with normal and patient with impaired renal function (an estimated glomerular filtration rate (eGFR) of above 90 ml/min/1.73m^2^ measured from blood samples is considered in the healthy range [@stevens_assessing_2006]). This replicates results shown using TLCO of a lower gradient in patients, however `3DQLayers` controls for kidney size resulting in the gradient being measured in quantitative units of Hz/mm rather than Hz/layer as in TLCO, thus increasing generalisability. -`3DQLayers` can also be used to analyse kidneys outside the body. \autoref{fig:exvivo_profiles} shows example quantitative maps acquired from a kidney removed for transplant and associated layer profiles. \autoref{fig:roi_layers_corr} compares results of ROI-based analysis and layer-based analysis in fifteen transplant kidneys. A significant correlation between outer layers and the cortex, and inner layers and the medulla was shown across all quantitative mapping techniques and a significant correlation between cortico-medullary ratio and layer gradient was shown for T~1~, T~2~, T~2~ ^\*^ and Magnetisation Transfer Ratio (MTR) mapping. +\autoref{fig:cortical_thickness} shows how `3DQLayers` can be used in combination with cortex and medulla tissue ROIs to analyse the distribution of voxel counts of each tissue as a function of layer depth of the kidney. From this, average cortical thickness can be defined as the depth at which the voxel distribution crosses from cortex to medulla. Cortical thickness has been hypothesised as a potential biomarker of renal disease [@yamashita_value_2015]. Here cortex and medulla ROIs are initially generated using a Gaussian mixture model to segment a T~1~-weighted structural image followed by manual ROI correction. -\autoref{fig:cortical_thickness} shows how `3DQLayers` can be used in combination with tissue ROI to analyse the distribution of tissues within the kidney. Average cortical thickness can be defined as the depth at which most voxels are medulla rather than cortex. Cortical thickness has been hypothesised as a potential biomarker [@yamashita_value_2015]. +`3DQLayers` can also be used to analyse ex-vivo kidneys outside the body. \autoref{fig:exvivo_profiles} shows example quantitative maps acquired from a kidney removed for transplant but subsequently deemed unsuitable and the associated layer profiles. \autoref{fig:roi_layers_corr} compares the results of tissue ROI based analysis and layer-based analysis in fifteen transplant kidneys. A significant correlation between outer layers and the cortex, and inner layers and the medulla was shown across all quantitative mapping techniques and a significant correlation between cortico-medullary ratio and layer gradient was shown for T~1~, T~2~, T~2~ ^\*^ and Magnetisation Transfer Ratio (MTR) mapping. # Figures -![a) A schematic of the kidneys showing the renal cortex and medullary pyramids. b) An anatomical MR Image of the abdomen showing the kidneys with the renal cortex appearing as a light band towards the outside of the kidneys and medullary pyramids as darker patches towards the centre of the kidneys. \label{fig:renal_structure}](kidney_overview.png){ width=90% } +![a) A schematic of the kidneys showing the renal cortex and medullary pyramids. b) A T~1~-weighted structural MR image of the abdomen showing the kidneys with the renal cortex appearing as a light band on the outer edge of the kidney and the medullary pyramids as darker patches on the inner portion of the kidneys. \label{fig:renal_structure}](kidney_overview.png){ width=90% } -![The mask from the T2-weighted structural scan (a i) has any cysts filled (a ii) and is converted into a smooth mesh representing the renal surface (b i and ii). The distance (in mm) from each voxel to the surface of the mesh is calculated (b iii). The renal pelvis is segmented (c i) and any tissue within 10 mm (c ii) of the pelvis is excluded from the depth map (c iii). The tissue is then grouped into layers of a desired thickness, here shown as 5 mm layers for illustrative purposes (d). \label{fig:flowchart}](flowchart.png) +![The mask automatically computed using a U-net from the T~2~-weighted structural MR image (a i) has any cysts filled (a ii) and is converted into a smooth mesh representing the renal surface (b i and ii). The distance (in mm) of each voxel to the surface of the mesh is then calculated to generate a depth map (b iii). The renal pelvis is segmented (c i) and any tissue within 10 mm (c ii) of the pelvis is excluded from the depth map (c iii). The tissue is then grouped into layers of a desired thickness, here shown as 5 mm renal layers for illustrative purposes (d). \label{fig:flowchart}](flowchart.png) -![Layers, R~2~ ^\*^ maps and layer profiles measured using `3DQLayers` for a subject with healthy renal function and a subject with impaired renal function. Shading around profiles shows the 95% confidence interval within each layer. \label{fig:egfr_gradients}](gradients.png) +![Layers, R~2~ ^\*^ maps, layer profiles, and central layer gradients for both left and right kidneys combined measured using `3DQLayers`. Examples are shown for a subject with normal renal function and a patient with impaired renal function. Shading around profiles shows the 95% confidence interval within each layer. \label{fig:egfr_gradients}](gradients.png) -![Example quantitative maps and associated layer profiles when `3DQLayers` is applied to transplant kidneys. Uncertainty shading shows the 95% confidence interval of each layer. \label{fig:exvivo_profiles}](example_profiles.png) +![Exploring the distribution of tissue types through the kidney to measure cortical thickness. \label{fig:cortical_thickness}](cortical_thickness.png){ width=50% } -![Agreement between tissue label-based analysis methods and layer-based analysis methods and the Pearsons correlation coefficient ($\rho$). * represents a _p_-value between 0.05 and 0.1, ** between 0.01 and 0.001, and *** < 0.001. \label{fig:roi_layers_corr}](roi_layers_corr.png) +![Example quantitative maps and associated layer profiles when `3DQLayers` is applied to ex-vivo transplant kidneys. Uncertainty shading shows the 95% confidence interval of each layer. \label{fig:exvivo_profiles}](example_profiles.png) -![Exploring the distribution of tissue types through the kidney to measure cortical thickness. \label{fig:cortical_thickness}](cortical_thickness.png){ width=50% } +![ Agreement between tissue ROI-based measures and analogous layer-based measures shown for fifteen ex-vivo transplant kidneys for each qMRI with the Pearsons correlation coefficient ($\rho$). * represents a _p_-value between 0.05 and 0.1, ** between 0.01 and 0.001, and *** < 0.001. a) Plots the median within each tissue ROI (cortex or medulla semi-automatically defined) against the equivalent layers (outer layers and inner layers respectively as highlighted in \autoref{fig:exvivo_profiles})b) Shows the cortico-medullary ratio (calculated by dividing the median within the cortex ROI by the median within the medullary ROI) against central layer gradient profiles calculated using `3DQLayers`. \label{fig:roi_layers_corr}](roi_layers_corr.png) # Acknowledgements +We acknowledge the funding support of Kidney Research UK (KS_RP_002_20210111), Medical Research Council (MR/R02264X/1), NIHR (NIHR128494), and NIHR Nottingham Biomedical Research Centre during the genesis of this project. # References \ No newline at end of file