Early diagnosis of CKD by imaging is crucial as there is not apparent clinical sign or finding in early stages
1. CT TA can have the potential in the evaluation of renal function before biochemical or anatomical changes occur. The expectations about texture analysis in our study is to put forth the microstructural differences quantitatively, in patients with chronic renal disease by using non enhanced CT images before human eye perceives. In this manner by using statistical based technique (by MATLAB), via placing ROI on cortex and medulla of the kidney; CT histogram method which extract 13 texture features, a total of 7 results managed to discriminate patients with high creatinine levels from the normal ones with a sensitivity from %70 to %52,4 and specifity from %56,4 to %50.0 And the features including kidney medulla mean maximum (r =-0,16, p =0.05), median , size % Land entropy values as well as size %U, size %M density from kidney cortex showed significant correlations with creatinine levels.
Kidneys possess high blood flow for water transport function. GFR which decreases during the development of chronic renal failure, is the result of glomerular loss and fibrosis histopathologically. In this process, structural deterioration that occurs is insidious; typically, kidneys with reduced size with thin parenchyma are seen in endstage renal disease patients, while there is no CT finding indicating decreased GFR in the early period. This study was designed by predicting that the attenuation value of the collimated X-ray beam in the area of interest can reflect histopathological changes before visible changes, thanks to histogram features. To the best of our knowledge, this is the first study that investigated CT histogram analysis to reveal such a relationship.
Previously, via DWI which detects the free Brownian motion of water molecules within a voxel of targeted tissue; relationship between DWI, eGFR and ADC values was examined 11. Although low eGFR patients tended to show with low ADC values, no significant correlation was found. However, in another study comparing the ADC values according to the stages, the ADC values were found to be significantly higher in stage 3, 4, 5 compared to the normal population 12.
In a comprehensive texture analysis study conducted according to the stages of liver fibrosis in the literature, whom had studied texture-based analysis on contrast-enhanced CT images suggested that texture features might have the potential to reveal liver fibrosis non-invasively 10. They also used MATLAB program like we did and found out 7 histogram features which were able to discriminate groups of hepatic fibrosis stages. Of these, fibrosis scores; the most correlated finding was median and the highest AUC value showed ‘poor ‘ROC curve results. In our study, similar to this study, when the group with normal creatinine levels was compared with the group with high creatinine levels; the highest AUC value is ‘mean’ value from the medulla. And this value corresponded to the range of poor ROC curve results (Table 2).
Previously, for ADC measurements of kidneys made from DWIs, where molecular diffusion was calculated for this purpose, ROI was placed at the corticomedullary junction to include the cortex and medulla from the middle part of the kidneys 11. In this study, we placed two separate ROI in the cortex and medulla although, there were some who predicted that this is not certain 13 Although the borders of the cortex and medulla are not clear, a circular ROI (60 mm2) with a diameter of 5 mm and the closest (at least 2 mm distance) to the sinus was placed in the cortex closest to the capsule (at least 2 mm away) to increase the hit rate.
As a matter of fact, histogram values taken from the medulla showed more correlation with creatinine values than those taken from the cortex (total 5 values in the medulla) (mean, median, maximum, size %L and entropy); and 2 values in the cortex (size %U, and size %M). Besides this; a total of 4 distinct features from the medulla and only 1 feature of the cortex differed between the normal and reduced renal functions group. This difference overlaps with the classical knowledge that microstructural changes occur primarily in the medulla.
In this study, texture analysis was performed with a statistical based method; and images are quantified with standard markers (Table 1). The distribution of the gray color level of the pixels in the ROI or their relationship to each other has turned into quantitative figures.
In some of the studies in the literature 3,10,14, it is seen that before CT histogram measurement filtration applications were used to reduce the exposure to photon noise at the beginning; by using spatial scale factor at different levels, for example (0 mm no filtrasyon, 2 mm fine texture scale, 3, 4 and 5 mm medium scale and 6 mm coarse texture scale (on the Laplacian of Gaussian spatial bandpass scale filter to produce images of different spatial scales) were used 15. Before histogram analysis we did not apply filtration because the number of pixels in the ROI drawn semi-quantitatively was almost standard, has smaller number of pixels than the others and the CT examinations in current study were without contrast. If we had used filtration; it probably would not have much of an effect for the reasons we mentioned above.
Histogram analysis is a post processing method; which produces diverse spectrum of texture features related to vendors (TEXRAD, MATLAB etc) but not standard. In this study we have used MAtlab program dealing with 13 features related to histogram features. For instance, Daginawala et al 10 in their study investigating the early diagnosis and staging of liver fibrosis, a total of 41 features including One gray level cooccurrence, 6 gray level run length, 1 Laws level, 4 gray level gradient matrix I have been investigated, 12 of which were histograms (mean median, SD, range etc.). Studies examining the relationship between tumor heterogeneity such as colorectal cancer, esophageal cancer, prognosis or response to treatment was activated gray level cooccurrence matrix dealing with 5 features 3-5.
The ‘entropy’ value is mostly a data corresponding to spatial heterogeneity in lesions with malignant potential and the mean value of the medulla and the serum creatine level are negatively correlated. In addition, a significant difference in terms of ‘uniformity’ values in the group with normal and abnormal creatinine levels were detected; It can be thought that as a result of sclerosis and fibrosis, the kidneys turn into a more uniform structure.
Generally, decrease in density of renal pixels is observed during chronic renal failure. And negative correlation with creatinine levels and mean, maximum and median % L size values; support this knowledge statistically.
Being a retrospective study, low number of patients with high serum creatinine levels; not correlated with histopathology findings, placement of ROI s manually (cortex and medulla discrimination may not be done on each patient) were limitations of this study. Intra or inter observer agreement was not tested. Serum creatinine levels may also affected by other conditions including dehydration, edema, infection and drugs those were not further interested. Data extracted from these type of studies may be helpful in a patient who has undergone non-contrast abdominal CT imaging for any reason, elevated serum creatinine can be predicted before visible changes begin.
)
)
)
)