Gray level co-occurrence matrix and wavelet analyses reveal discrete changes in proximal tubule cell nuclei after mild acute kidney injury

doi:10.21203/rs.3.rs-1762202/v1

Download PDF

Article

Gray level co-occurrence matrix and wavelet analyses reveal discrete changes in proximal tubule cell nuclei after mild acute kidney injury

https://doi.org/10.21203/rs.3.rs-1762202/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 10 Mar, 2023

Read the published version in Scientific Reports →

You are reading this latest preprint version

Acute kidney injury relates to an abrupt reduction in renal function resulting from numerous conditions. Morbidity, mortality, and treatment costs related to AKI are relatively high. This condition is strongly associated with damage to proximal tubule cells (PTCs), generating distinct patterns of transcriptional and epigenetic alterations that result in structural changes in the nuclei of this epithelium. To this date, AKI-related nuclear chromatin redistribution in PTCs is poorly understood, and it is unclear whether changes in PTC chromatin patterns can be detected using conventional microscopy during mild AKI, which can progress to more debilitating forms of injury. In recent years, gray level co-occurrence matrix (GLCM) analysis and discrete wavelet transform (DWT) have emerged as potentially valuable methods for identifying discrete structural changes in nuclear chromatin architecture that are not visible during the conventional histopathological exam. Here we present findings indicating that GLCM and DWT methods can be successfully used in nephrology to detect subtle morphological alterations associated with mild tissue injury demonstrated in rodents by inducing a mild form of AKI through ischemia-reperfusion injury. This rodent model allowed us to show that pre-renal mild AKI is associated with the significant reduction of textural homogeneity of PTC nuclei, indirectly assessed by GLCM indicators and DWT energy coefficients.

Acute kidney injury denotes an abrupt and rapid reduction in the kidney’s ability to perform excretory functions and other roles that support homeostasis¹. There are numerous causes of AKI, such as blood hypoperfusion, dehydration, sepsis, effects of various toxins, as well as urinary tract obstructions. These conditions can, in turn, be classified as pre-renal, intrinsic, and post-renal forms of injury. Many AKI symptoms and signs are nonspecific and include confusion, drowsiness, nausea, diarrhea, dehydration, and decreased urine output. Also, some AKIs do not present with obvious symptoms and are very difficult to diagnose. Morbidity and mortality due to AKI are relatively high, and so is the cost of treatment. According to some data, AKI remains one of the biggest financial burdens on healthcare systems, considering the number of hospitalizations and the funding needed for each hospital stay^{2, 3}.

It is well known that AKI is strongly associated with damage to the proximal tubule and that the dysfunction or death of proximal tubule cells (PTCs) are often the main consequence of AKI⁴. Cells within this epithelium often undergo programmed cell death or necrosis depending on the severity of the injury, while in some very mild AKI cases, the PTC dysfunction is barely noticeable and reversible without intervention⁵. Dysfunction of PTCs is often followed by distinct patterns of transcriptional and epigenetic alterations^6–8. Even during the very early stages of AKI, the expression of numerous genes is increased or decreased, which may or may not result in structural changes in PTC nuclei. To this date, AKI-related nuclear chromatin redistribution in PTCs is poorly understood. It is unclear whether changes in PTC chromatin patterns can be detected using conventional microscopy approaches during mild AKI. Also, it is unknown if structural changes in PTC nuclei in mild or moderate AKI have any diagnostic or research value in pathology and nephrology.

In microscopy, various ways exist to evaluate and quantify structural changes in cell nuclei. In recent years, two contemporary and innovative computational techniques have emerged as potentially valuable methods for identifying discrete morphological changes in nuclear chromatin architecture that are not visible during the conventional histopathological exam. The first technique is based on gray level co-occurrence matrix (GLCM) analysis, and it is commonly used to quantify nuclear textural features such as uniformity, homogeneity, and entropy^{9, 10}. The second technique is based on the discrete wavelet transform (DWT), a mathematical approach to the texture frequently applied for two-dimensional signal analysis as an addition to GLCM. Previous research has shown that both methods are potentially valuable tools in pathology for differentiating damaged and intact cells^{11, 12}. Also, both methods can be used to train and develop artificial intelligence (AI) machine learning models, such as those based on decision trees, logistic regression, or artificial neural networks.

In this work, we present findings indicating that GLCM and DWT methods can be successfully used in nephrology to detect subtle morphological alterations associated with mild tissue injury demonstrated in rodents. A well-established bilateral renal pedicle clamp rat model induced a mild form of AKI through ischemia-reperfusion injury (IRI)^{4, 13–15}. Mild IRI is a pathological process that leads to an acute tubulointerstitial injury that can be followed by a repair process that restores renal function, as outlined by serum creatinine levels, urine production, and the need for renal replacement therapy¹⁶. This rodent model allowed us to show that pre-renal mild AKI is associated with the reduction of textural homogeneity of PTC nuclei quantified by GLCM and the increase of nuclear structural heterogeneity indirectly assessed with DWT energy coefficients. To the best of our knowledge, this is the first study to quantify GLCM and DWT indicators of PTC nuclear structure after mild AKI and the first to show that these indicators can be used to separate cells from damaged and intact tissue. In this article, we also propose creating two hypothetical AI models based on random forest and support vector machine algorithms that theoretically have high accuracy and discriminatory power in classifying PTC nuclear structural patterns associated with prominent renal injuries. Such AI-based solutions can pave the way for digital pathology-based platforms that support enhanced quantitative histopathologic assessments that allow us to explore and extract information beyond human visual perception¹⁷.

Serum Creatinine and Histological Analyses

We monitored serum creatinine levels, SCr, in both groups of animals. The levels observed among the control (sham) rats showed a minor and non-significant (p = 0.068) fluctuation in the serum creatinine concentration from the 0-hour to the 24-hour mark post-procedure. Moreover, these creatinine values remained within normal baseline levels (0.3 ≤ SCr ≤ 0.6 mg/dl) compared to the measurements obtained from animals with mild IRI. In contrast, the creatinine levels in the acutely injured animals rose abruptly and significantly (p < 0.01) above their original range at the 0-hour mark during the first 24 hours after injury (0.9 ≤ SCr ≤ 1.2 mg/dl). However, these levels gradually returned to baseline within the 7-day measurement period (Fig. 1C). Serum creatinine measurements differed significantly across the 7 days (p < 0.05) from the levels measured in sham animals. Likewise, brightfield images collected from cortical sections in the sham group revealed subtle alterations to the renal microarchitecture (Fig. 1A), and more pronounced disruptions were observed after inducing the mild form of IRI (Fig. 1B). Altogether, these changes to normal creatinine levels and the renal ultrastructure correlate well with standard biochemical and histological findings.

GLCM Analysis

The average values of the angular second moment and inverse difference moment in the controls were 0.036 ± 0.003 and 0.53 ± 0.01, respectively, while in the AKI group, the values were decreased and equaled 0.018 ± 0.001 and 0.41 ± 0.01, respectively (Fig. 2). Differences between AKI and controls were statistically highly significant for both GLCM indicators (p < 0.01). This result implied that both local textural homogeneity and uniformity of PTC nuclei decrease due to AKI. Conversely, the mean values of textural contrast increased after AKI from 5.02 ± 0.52 (controls) to 8.73 ± 0.66 (p < 0.01), and a similar, although the less pronounced increase was observed in the GLCM correlation feature (0.806 ± 0.013 to 0.841 ± 0.006, p < 0.01). The mean values of textural Sum Average and textural Sum Variance indicators also increased, and the change was highly significant (p < 0.01). In the control group, these values equaled 44.21 ± 0.53 and 68.36 ± 11.98, respectively, while in the AKI group they equaled 51.21 ± 0.86 and 149.47 ± 8.36, respectively.

The mean values of discrete wavelet transform coefficient energies increased in AKI, which is in accordance with the previous GLCM results showing the reduction of textural homogeneity and uniformity. The wavelet coefficient energy, EnLH, obtained with the use of a combination of low-pass and high-pass filleters showed an increase from 16.38 ± 1.80 in the controls to 30.37 ± 1.75 in AKI (p < 0.01), while a similar increase was observed for its EnHL counterpart (16.98 ± 1.83 to 30.27 ± 2.39), (p < 0.01). The wavelet coefficient energy, EnHH, in the control group equaled 2.69 ± 0.23, and in the AKI group, it equaled 4.39 ± 0.37 (p < 0.01). Figure 3 shows the mean values and standard deviations for DWT indicators.

In this work, we propose creating three hypothetical machine learning models based on logistic regression ¹⁸, support vector machine (SVM), and random forest (RF). Of the three trained and tested models, the highest accuracy, equaling 0.79 was determined for the random forest algorithm. The model had excellent discriminatory power in terms of potential ROI classification, and the area under the receiver operating characteristics curve equaled 0.86. The ROC curves for each model are presented in Fig. 4. The support vector machine model had an estimated accuracy of 0.73 and the area under the ROC curve of 0.79. The logistic regression model presented the lowest accuracy of 0.69 with the area under the ROC curve of 0.78. These results indicate that the random forest classifier has the greatest potential in future development of computer-aided sensing systems for the identification of acute kidney injury.

Even though AKI is highly variable in its clinical presentation, damage to proximal tubule epithelium is a significant hallmark of the condition that can manifest through various mechanisms that elicit pre-renal, intrinsic, and post-renal injuries. For instance, the degree of injury, clinical severity, and progression of AKI are associated with the duration of IRI. The current understanding of AKI has been developed through extensive research using animal models, particularly mouse and rat species, and the ability to rapidly and reliably induce mild, moderate, and severe damage in the PCTs by administering nephrotoxins in various doses¹⁹ and modulating the duration needed to induce IRI²⁰. Based on these facts, in this work, we demonstrate the ability of gray level co-occurrence matrix and discrete wavelet transform methods to detect subtle structural changes in PTC nuclei after mild pre-renal AKI.

The gold standard in the biochemical evaluation of AKI relies on serum creatinine levels. These peak serum creatinine elevations 24 hours post-ischemia corresponds to peak injury and is a hallmark of this model²¹. Using this approach, we verified the induction of mild AKI using a mild form of IRI, evidenced by the two-fold rise in this biomarker 24 hours after injury and its gradual return to baseline within a week. Moreover, biopsies also revealed that the bilateral pedicle clamp model supported mild tubular disruptions evidenced by histology The examination of images collected from sham and mildly injured animals allowed us to show that the most critical nuclear GLCM indicators such as angular second moment and inverse difference moment significantly change as the result of mild AKI, which indicates the rise of nuclear textural heterogeneity due to the injury-associated redistribution of euchromatin and heterochromatin. To our knowledge, this is the first study to combine GLCM indicators and DWT coefficient energies to reveal discrete AKI-related alterations in PTC nuclear architecture. We also propose hypothetical machine learning models based on support vector machines, random forest, and logistic regression, which might in the future be used as a part of accurate computational AI sensing systems for diagnostic purposes.

In our previous articles, we applied similar GLCM approaches to analyze alterations to the renal vascular architecture²², highlighting its potential application in the characterization of whole organ scaffolds²³ that can be generated for bioartificial kidney development²⁴. Using this technique, We also examined cell nuclei after the damage induced by exposure to a sublethal toxic dose of ethanol¹². On an experimental model of saccharomyces cerevisiae, we calculated angular second moment, inverse difference moment, textural contrast, GLCM correlation, and variance and demonstrated that these features significantly change after alcohol treatment. This type of cell damage was also associated with the reduction of textural homogeneity and uniformity, leading us to believe that these changes in nuclear textural patterns are generally related to cell damage. Similarly, as in the present study, we proposed several machine learning models, such as the ones based on logistic regression, decision trees, and artificial neural networks¹². Despite the apparent differences in the methodological approach and experimental protocol between the two works, this assumption is worthy of investigation in future research.

One of the earliest research articles on applying GLCM in the histological evaluation of kidney tissue was published in 2013. Indicators such as the angular second moment and inverse difference moment were quantified to assess chromatin architecture in macula densa cells during mice postnatal development and aging²⁵. Textural features were evaluated in conjunction with fractal dimension and lacunarity as indicators of complexity. Although no statistically significant changes were detected in ASM and IDM values, the research nevertheless holds some value since it was the first to show that GLCM analysis is not only possible in kidney tissue but also applicable for scoring various aspects of textural heterogeneity in individual kidney cell nuclei.

Our current study is also not the first to use the GLCM computational method for the detection of AKI. Previously, textural features such as GLCM contrast and GLCM correlation were quantified in the kidney medulla after inducing IRI in rats by clamping both renal vascular pedicles and subsequent reperfusion with saline²⁶. It was shown that both CON and COR features had excellent discriminatory power in separating injured from control medullar tissue, with the area under the receiver operating characteristic curve in both cases higher than 85%. The results identified both fractal and GLCM parameters as suitable candidates for the future development of computational biosensors in nephropathology.

There are several potential explanations for the AKI-related changes in PTC nuclear textural patterns, which were detected in our current study. First, it is possible that AKI led to the redistribution of euchromatin and heterochromatin in PTCs, and that the redistribution was the result of either direct damage to the cell or activation of a signaling pathway. AKI is indeed associated with sometimes profound epigenetic changes, as explained in detail by other authors⁶. Some of the genes that are upregulated during AKI may also influence chromatin integrity and remodeling on higher scales. Although these phenomena are generally not noticeable during the standard histopathological evaluation, the subsequent changes in textural patterns may have been detected with GLCM and DWT. Also, it should be considered that, sometimes, euchromatin and heterochromatin, at least in the ultrastructural sense, have different levels of fractal complexity²⁷, and these differences in complexity may have reflected on GLCM and DWT features in this experimental setting as well.

Another possibility is that the mild AKI in the renal cortex is sometimes associated with programmed cell death. Indeed, PTCs are highly susceptible to apoptosis, as discussed earlier²⁸, and this type of cell death contributes to the loss of kidney functionality during AKI. On the other hand, some previous works have suggested that GLCM indicators such as angular second moment and inverse difference moment significantly decrease after cell treatment with proapoptotic substances²⁹. During the early stages of programmed cell death, phenomena such as (initial) condensation and marginalization of chromatin may perhaps lead to the increased textural heterogeneity, detectable using both GLCM and DWT. However, additional research is needed to confirm this assumption, particularly on PTCs and other cell populations in the renal cortex.

In the future, it will be possible to broaden this type of research by developing AI models based on artificial neural networks. This approach could include relatively simple perceptron networks, but also more complex recurrent (RNNs) and convolutional neural networks (CNNs). The input layer of neurons in these models could receive DWT and GLCM data, but also the data from a three-dimensional matrix of values based on red, green, and blue light intensities. Inclusion of various other image analysis quantifications, such as fractal dimension, lacunarity, and granularity might further benefit the network’s ability to distinguish damaged from intact cells. Convolutional neural networks are of particular interest since previously, they have been successfully applied for image classification on numerous occasions^{30, 31}. Creating a complex CNN that combines DWT and GLCM with other input parameters could lead to developing a sensitive, accurate, and affordable computer-aided diagnostic system that might serve as an essential addition to current nephropathology practices.

Our study had several important limitations that need to be discussed and considered when conducting future research in this scientific area. First, GLCM and DWT methods are relatively new in terms of their applications in pathology and histology, so there is a lack of data on their quality assurance and validity. Different software platforms often produce different results, and based on our previous experience, indicators such as angular second moment and inverse difference moment can significantly vary depending on the software settings and parameters during micrograph acquisition. Second, one needs to stress that the AI models proposed in this research are only hypothetical since they were trained and tested on a very limited number of nuclear ROIs. To increase validity and test this approach's diagnostic value, one would need to develop the machine learning models on a very large sample of micrographs with one ROI corresponding to one individual micrograph or even one individual animal. Finally, also from our previous experience, values obtained through GLCM, and DWT analyses greatly depend on the histological staining applied to the tissue. In the future, one might consider repeating the experiments and using other techniques such as periodic acid–Schiff, Sirius Red, Feulgen, or Toluidine Blue. Only then would we be able to have full insight into the true scientific value of GLCM and DWT computational methods.

In conclusion, we present evidence that GLCM and DWT computational methods can detect subtle structural alterations in PTC nuclei associated with mild AKI. After quantifying textural features such as the angular second moment and inverse difference moment of nuclear architecture, we conclude that this form of injury leads to the rise of nuclear textural heterogeneity. This change is not clearly visible during a conventional histopathological evaluation. Since this syndrome rarely has a sole and distinct and is frequent among patients without critical illness, it is essential that health care professionals, especially those without specialization in renal disorders, detect it easily³². Thus, we propose creating AI models that use GLCM and DWT indicators as input data, capable of AKI identification and PTC classification with great accuracy and discriminatory power. The obtained results present a potentially valuable foundation for future research in AI applications in pathology, nephrology, and related disciplines and support current regimens used to address AKI.

Mild acute kidney injury

The study was conducted in accordance with the Indiana University School of Medicine Institutional Animal Care and Use Committee, the Animal Research Oversight Committee at Khalifa University of Science and Technology, and the ARRIVE guidelines. All procedures were approved by these organizations and carried out using these relevant regulations and guidelines on 200 to 400 gm male Sprague Dawley rats (Harlan Laboratories, Indianapolis, IN) to ensure that animals were treated ethically and humanely. The animals were anesthetized using 5% isoflurane delivered in oxygen (Webster Veterinary Supply, Devens, MA) and then given intraperitoneal injections of 50 mg/kg of pentobarbital (Hospira, Inc., Lake Forest, IL). Each rat was then placed on a heating pad to maintain normal physiological temperature, and intraperitoneal incisions were made to expose both renal pedicles for the injured and sham groups. Bilateral micro-serrefine clamps with delicate, atraumatic serrations (Fine Science Tools, Foster City, CA, USA) were applied to occlude blood flow for 15–20 minutes (animals in the sham group was not subjected to the bilateral clamps), as shown in Fig. 5. After this period, the clamps were removed to reinstate renal blood flow, and the animals were allowed to recover fully post ischemia-reperfusion and sham injuries. It should also be noted that, throughout our studies, the rats were given free access to standard rat chow and water.

Serum Creatinine Measurements

Blood samples were collected from the injured and sham rats daily across a week, in 1 ml Eppendorf heparin-treated tubes after making small incisions on their tails. These samples were centrifuged at 100,000-130,000 rpm for 10 minutes. The supernatants were then stored at 4°C. A quantitative determination of creatine kinase activity in serum was then estimated with Pointe Scientific CK (Liquid) Reagents (Point Scientific, Inc., Canton, MI, USA). Measurements were performed with a Beckman Creatinine Analyzer 2 (Beckman Instruments, Brea, CA) according to the manufacturer’s specifications and reported values in milligrams per deciliter (mg/dl). Approximately, 10 µL of each serum sample was added to the working reagent (1000 µL) and the absorbance was immediately measured using a microplate reader.

Histology

Euthanasia was performed on day 7 to obtain whole kidneys from the two groups of animals. For this process, the animals were again anesthetized with pentobarbital, and once fully sedated, whole kidneys were acquired after each renal pedicle was clamped. Once removed, the kidneys were fixed with 4% paraformaldehyde for 24 hours at 4°C. The samples were then immersed in 10% neutral buffered formalin or 4% phosphate-buffered formalin, again for a minimum of 24 hours at room temperature. The specimens were then rinsed in distilled H₂O and stored in 70% ethanol. For infiltration, the specimens were dehydrated through a graded series of ethanol (70%; 80%, 95%, and 100%; two changes each under vacuum for 45 minutes at room temperature). The specimens were cleared in two changes of xylene (under vacuum at room temperature for 45 minutes each), infiltrated with 4 changes of paraffin (under vacuum at 59°C; 45 minutes each), and embedded in fresh paraffin. After which, 4–5 µm thick sections were cut using a Reichert-Jung 820 microtome (Depew, NY). The sections were flattened on a warm water bath and mounted on coated and charged glass slides. After drying, the sections were deparaffinized, rehydrated, and stained with hematoxylin and eosin (H&E). Images were then acquired with a Nikon Microphot SA Upright Microscope equipped with a sensitive Diagnostic Instruments SPOT RT Slider color camera (Nikon, Tokyo, Japan) and a 20X objective.

Computational Analysis

We created digital micrographs of the renal cortex with dimensions of 1600 (width) x 1200 (height) resolution units in uncompressed TIF format and RGB photometric interpretation. The micrographs had both horizontal and vertical resolution of 96 dpi and a bit depth of 48. They were later converted to BMP format with the same dimensions and a bit depth of 24 and analyzed in a grayscale setup (Fig. 6). For GLCM and DWT analyses we used Mazda software previously developed for the needs of COST B11 European project "Quantitative Analysis of Magnetic Resonance Image Texture" (1998–2002) and COST B21 European project "Physiological modeling of MR Image formation"^33–36. A similar micrograph analysis protocol was used in the previous publications¹¹. Briefly, a total of 1400 regions of interest of PTC nuclei were analyzed: 700 from animals with mild AKI (100 per animal) and another 700 from control animals. For each ROI, 6 GLCM indicators were quantified: angular second moment (ASM), inverse difference moment (IDM), GLCM contrast (CON), GLCM correlation (COR), GLCM sum average (SA), and textural sum of variance (SVAR).

The angular second moment is a textural feature that quantifies the uniformity of the two-dimensional signal and can be calculated from the probability (p) values for changes between gray levels i and j:

$$\text{ASM=}\sum _{\text{i}}\sum _{\text{j}}{\left\{\text{p}\left(\text{i, j}\right)\right\}}^{\text{2}}$$

Here, the resolution units are assigned values taking into account their gray level intensities. Another similar feature that quantifies textural local homogeneity of the signal is inverse difference moment, which is determined as:

$$\text{IDM }\text{=}\sum _{\text{i}}\sum _{\text{j}}\frac{\text{1}}{{\text{1+}\left(\text{i-j}\right)}^{\text{2}}}\text{ p(i,j)}$$

In this work, we also quantified textural correlation (gray-level linear dependency) and textural contrast (local intensity variation) as additional features which supplement angular second moment and inverse difference moment:

$\text{C}\text{O}\text{N}=\sum _{\text{i}}\sum _{\text{j}}{\left(\text{i}-\text{j}\right)}^{\text{k}}{\text{P}}_{\text{d}}{[\text{i},\text{j}]}^{\text{n}}$

$\text{C}\text{O}\text{R}=\frac{\sum _{\text{i}}\sum _{\text{j}}\left(\text{i}\text{j}\right)\text{p}\left(\text{i}, \text{j}\right)- {{\mu }}_{\text{x}}{{\mu }}_{\text{y}}}{{{\sigma }}_{\text{x}}{{\sigma }}_{\text{y}}}$

where in the normalized GLCM, µ is the mean of rows x and y, and σ the standard deviation^37–39.

Sum average of the GLCM as the mean of the gray level sum distribution of the image and Sum variance as the indicator of gray level textural dispersion around the mean³⁷ were quantified as:

$$\text{S}\text{V}\text{A}\text{R}=\sum _{ }\text{i}{\text{p}}_{\text{x}+\text{y}}\left(\text{i}\right)$$

$$\text{S}\text{V}\text{A}\text{R}={\sum }_{ }{\left[\text{i}-\sum _{ }\text{i}{\text{p}}_{\text{x}+\text{y}}\left(\text{i}\right)\right]}^{2}$$

In addition to the quantification of GLCM textural features, we also applied discrete wavelet transform analysis³³ by performing linear transformation of data vectors with the vectors having a length of an integer power of two. The vectors were transformed into numerical vectors that had the same length. A special filtering cascade was applied after the separations of rows and columns of data which included the use of high (H) and low-pass (L) filters. We quantified 3 DWT wavelet coefficients (d) energies for the specific combination of filters for the first subband at successive scales: EnLH, EnHL, and EnHH. Considering the subband location (x,y), the scale, and the ROI number of resolution units (n), the energies can be determined as:

$$\text{E}=\frac{{\sum }_{\text{x},\text{y}\in \text{R}\text{O}\text{I}}{\left({\text{d}}_{\text{x},\text{y}}^{\text{s}\text{u}\text{b}\text{b}\text{a}\text{n}\text{d}}\right)}^{2}}{\text{n}}$$

In this study, we also proposed the creation of 3 hypothetical AI machine learning models: support vector machine (SVM), random forest (RF), and a model based on the binomial logistic regression¹⁸. The models were created in scikit-learn open source, commercially usable machine learning library for the Python programming language⁴⁰. The models were trained using the GLCM and wavelet indicators as inputs on 80% of the ROIs while the remaining data were used for testing. We determined the average accuracy and discriminatory power (area under the receiver operating curve) in terms of the ability of the model to assign the ROI to the AKI or the control kidney tissue. The data were analyzed using Pandas, an open source data analysis and manipulation Python library, as well as IBM SPSS 25.0 statistical analysis software. The Mann-Whitney U test was used to determine whether the increases in serum creatinine observed across the first 24 hours after injury was significant. All variables are expressed as mean ± standard deviation, and for all evaluations, a p value of less than 0.05 was considered statistically significant.

DATA AVAILABILITY

The datasets generated during and/or analyzed during the current study are available from the corresponding author on reasonable requests.

ACKNOWLEDGEMENTS

The authors would like to thank Ms. Raheema Khan and Ms. Meklit G. Shibru for proofreading the manuscript and help preparing Fig. 5.

FUNDING

Prof. Igor Pantic received financial support from the Science Fund of the Republic of Serbia, grant #7739645 “Automated sensing system based on fractal, textural and wavelet computational methods for detection of low-level cellular damage,” SensoFracTW. The Indiana Center for Biological Microscopy also supported this study, School of Medicine, Indiana University, through an NIH P-30 O’Brien Center (DK 079312) award and funds from Khalifa University of Science and Technology under Award No. RC2-2018-022 (HEIC) and Research Fund FSU-2020-25 were granted to Prof. Peter R. Corridon.

COMPETING INTERESTS

The authors declare no competing interests.

AUTHOR CONTRIBUTIONS

P.R.C and I.P. conceived and designed the project. P.R.C. performed all experiments, analyzed the associated data, and interpreted the results of the experiments. I.P., J.C., and S.D. performed all computation analyses. The authors drafted, edited, and approved the final version of the manuscript.

Song, H.H.G., Rumma, R.T., Ozaki, C.K., Edelman, E.R. & Chen, C.S. Vascular Tissue Engineering: Progress, Challenges, and Clinical Promise. Cell Stem Cell 22, 340-354 (2018).
Torio, C.M. & Andrews, R.M. in Healthcare Cost and Utilization Project (HCUP) Statistical Briefs (Agency for Healthcare Research and Quality (US), Rockville (MD); 2006).
Silver, S.A. & Chertow, G.M. The Economic Consequences of Acute Kidney Injury. Nephron 137, 297-301 (2017).
Corridon, P.R., Karam, S.H., Khraibi, A.A., Khan, A.A. & Alhashmi, M.A. Intravital imaging of real-time endogenous actin dysregulation in proximal and distal tubules at the onset of severe ischemia-reperfusion injury. Sci Rep 11, 8280 (2021).
Corridon, P.R. et al. A method to facilitate and monitor expression of exogenous genes in the rat kidney using plasmid and viral vectors. Am J Physiol Renal Physiol 304, F1217-1229 (2013).
Bonventre, J.V. & Yang, L. Cellular pathophysiology of ischemic acute kidney injury. The Journal of clinical investigation 121, 4210-4221 (2011).
Sato, Y., Takahashi, M. & Yanagita, M. Pathophysiology of AKI to CKD progression. Semin Nephrol 40, 206-215 (2020).
Sharifian, R. et al. Distinct patterns of transcriptional and epigenetic alterations characterize acute and chronic kidney injury. Sci Rep 8, 17870 (2018).
Gopinath, B. & Shanthi, N. Computer-aided diagnosis system for classifying benign and malignant thyroid nodules in multi-stained FNAB cytological images. Australas Phys Eng Sci Med 36, 219-230 (2013).
Cai, X., Li, X., Razmjooy, N. & Ghadimi, N. Breast Cancer Diagnosis by Convolutional Neural Network and Advanced Thermal Exchange Optimization Algorithm. Comput Math Methods Med 2021, 5595180 (2021).
Paunovic, J. et al. Effects of Iron Oxide Nanoparticles on Structural Organization of Hepatocyte Chromatin: Gray Level Co-occurrence Matrix Analysis. Microscopy and Microanalysis 27, 889-896 (2021).
Davidovic, L.M. et al. Gray-Level Co-occurrence Matrix Analysis for the Detection of Discrete, Ethanol-Induced, Structural Changes in Cell Nuclei: An Artificial Intelligence Approach. Microscopy and microanalysis : the official journal of Microscopy Society of America, Microbeam Analysis Society, Microscopical Society of Canada 28, 265-271 (2022).
Collett, J.A. et al. Hydrodynamic Isotonic Fluid Delivery Ameliorates Moderate-to-Severe Ischemia-Reperfusion Injury in Rat Kidneys. J Am Soc Nephrol 28, 2081-2092 (2017).
Kolb, A.L. et al. Exogenous Gene Transmission of Isocitrate Dehydrogenase 2 Mimics Ischemic Preconditioning Protection. J Am Soc Nephrol 29, 1154-1164 (2018).
Ashworth, S.L., Sandoval, R.M., Hosford, M., Bamburg, J.R. & Molitoris, B.A. Ischemic injury induces ADF relocalization to the apical domain of rat proximal tubule cells. Am J Physiol Renal Physiol 280, F886-894 (2001).
Katayama, S. et al. Factors predicting successful discontinuation of continuous renal replacement therapy. Anaesthesia and intensive care 44, 453-457 (2016).
Baxi, V., Edwards, R., Montalto, M. & Saha, S. Digital pathology and artificial intelligence in translational medicine and clinical practice. Modern Pathology 35, 23-32 (2022).
Shaya, J. et al. Design, photophysical properties, and applications of fluorene-based fluorophores in two-photon fluorescence bioimaging: A review. Journal of Photochemistry and Photobiology C: Photochemistry Reviews 52, 100529 (2022).
Hall, A.M., Rhodes, G.J., Sandoval, R.M., Corridon, P.R. & Molitoris, B.A. In vivo multiphoton imaging of mitochondrial structure and function during acute kidney injury. Kidney Int 83, 72-83 (2013).
Dong, Y. et al. Ischemic Duration and Frequency Determines AKI-to-CKD Progression Monitored by Dynamic Changes of Tubular Biomarkers in IRI Mice. Front Physiol 10, 153 (2019).
Williams, P. et al. Characterization of renal ischemia-reperfusion injury in rats. J Pharmacol Toxicol Methods 37, 1-7 (1997).
Pantic, I.V., Shakeel, A., Petroianu, G.A. & Corridon, P.R. Analysis of Vascular Architecture and Parenchymal Damage Generated by Reduced Blood Perfusion in Decellularized Porcine Kidneys Using a Gray Level Co-occurrence Matrix. Front Cardiovasc Med 9, 797283 (2022).
Corridon, P.R. In vitro investigation of the impact of pulsatile blood flow on the vascular architecture of decellularized porcine kidneys. Sci Rep 11, 16965 (2021).
Corridon, P.R., Ko, I.K., Yoo, J.J. & Atala, A. Bioartificial Kidneys. Curr Stem Cell Rep 3, 68-76 (2017).
Pantic, I. et al. Complexity reduction of chromatin architecture in macula densa cells during mouse postnatal development. Nephrology (Carlton) 18, 117-124 (2013).
Pantic, I. et al. Fractal analysis and Gray level co-occurrence matrix method for evaluation of reperfusion injury in kidney medulla. J Theor Biol 397, 61-67 (2016).
Shinkai, S., Nozaki, T., Maeshima, K. & Togashi, Y. Dynamic Nucleosome Movement Provides Structural Information of Topological Chromatin Domains in Living Human Cells. PLoS Comput Biol 12, e1005136-e1005136 (2016).
Havasi, A. & Borkan, S.C. Apoptosis and acute kidney injury. Kidney Int 80, 29-40 (2011).
Nikolovski, D., Cumic, J. & Pantic, I. Application of Gray Level co-Occurrence Matrix Algorithm for Detection of Discrete Structural Changes in Cell Nuclei After Exposure to Iron Oxide Nanoparticles and 6-Hydroxydopamine. Microscopy and microanalysis : the official journal of Microscopy Society of America, Microbeam Analysis Society, Microscopical Society of Canada 25, 982-988 (2019).
Debelee, T.G., Kebede, S.R., Schwenker, F. & Shewarega, Z.M. Deep Learning in Selected Cancers' Image Analysis-A Survey. Journal of imaging 6 (2020).
Oei, R.W. et al. Convolutional neural network for cell classification using microscope images of intracellular actin networks. PLoS One 14, e0213626 (2019).
Makris, K. & Spanou, L. Acute Kidney Injury: Definition, Pathophysiology and Clinical Phenotypes. Clin Biochem Rev 37, 85-98 (2016).
Kociolek, M., Materka, A., Strzelecki, M. & Szczypiński, P. Discrete wavelet transform-derived features for digital image texture analysis. (2001).
Strzelecki, M., Szczypinski, P., Materka, A. & Klepaczko, A. A software tool for automatic classification and segmentation of 2D/3D medical images. Nuclear Instruments and Methods in Physics Research Section A: Accelerators, Spectrometers, Detectors and Associated Equipment 702, 137-140 (2013).
Szczypiński, P.M., Strzelecki, M., Materka, A. & Klepaczko, A. MaZda—A software package for image texture analysis. Computer Methods and Programs in Biomedicine 94, 66-76 (2009).
Szczypinski, P.M., Strzelecki, M. & Materka, A. in 2007 International Symposium on Information Technology Convergence (ISITC 2007) 245-249 (2007).
Mapayi, T., Viriri, S. & Tapamo, J.-R. Adaptive Thresholding Technique for Retinal Vessel Segmentation Based on GLCM-Energy Information. Computational and Mathematical Methods in Medicine 2015, 597475 (2015).
Ou, X., Pan, W. & Xiao, P. In vivo skin capacitive imaging analysis by using grey level co-occurrence matrix (GLCM). Int J Pharm 460, 28-32 (2014).
Santos, T.A. et al. MRI Texture Analysis Reveals Bulbar Abnormalities in Friedreich Ataxia. AJNR Am J Neuroradiol 36, 2214-2218 (2015).
Pedregosa, F. et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 12, 2825-2830 (2011).

No competing interests reported.

Download PDF

Journal Publication

published 10 Mar, 2023

Read the published version in Scientific Reports →

Editor invited by journal
21 Jun, 2022
Submission checks completed at journal
21 Jun, 2022
First submitted to journal
15 Jun, 2022

You are reading this latest preprint version

Gray level co-occurrence matrix and wavelet analyses reveal discrete changes in proximal tubule cell nuclei after mild acute kidney injury

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Results

Serum Creatinine and Histological Analyses

GLCM Analysis

Discussion

Materials And Methods

Mild acute kidney injury

Serum Creatinine Measurements

Histology

Computational Analysis

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1