Kidney Filtration Model Investigation Based On Three Dimensional Medical Images

Evaluation of renal function plays an important role in kidney disease diagnosis and treatment process. The commonly used clinical methods for renal function evaluation are biochemical test and radionuclide scanning. The disadvantage of biochemical test method is that we can not obtain the unilateral renal function, which is quite important in many cases. The image resolution of radionuclide scanning is low, besides, the patient has to receive radiation. This paper studied the renal function evaluation model and automated analyzing method based on multi-phase CT and MR images. Through this method, we can obtain not only unilateral renal function, but also high resolution anatomical images. Obviously, it will have a very good prospect in clinical application.Our research focused on kidney filtration model and automated Glomerular Filtration Rate (GFR) measurement method based on3D medical images. We fully analyzed the core principles and methods of existing models, and performed data experiments through the cooperation with the hospitals. In CT image-based GFR model study, we analyzed the Patlak2-compartment model, and finished the patients experiment. Then we discussed the3-compartment model concerning the interstitial space in the kidney. Furthermore, we studied the primary models based on MR images, and deeply analyzed the Impulse Retention Function (IRF) based3-compartment model, and evaluated the measurement result with our model pig data. By comprehensive analysis of the advantage and disadvantage among the existing models, we proposed an improved GFR measurement model, which introduced the IRF of vascular compartment based on Patlak2-compartment model. In this model, there will be a more accurate description of the contrast agent concentration in vascular compartment of kidney, and the accuracy of the model will be improved accordingly.To implement the GFR measurement model, we proposed an automated measurement method based on CT and MR images, which included the automatic kidney positioning and segmentation, as well as the extraction of cortex and medulla of the kidney. The automated measurement method simplifies the operation of doctors’, making it possible to apply the measurement model in clinical practice.During the study, concerning the requirement from doctors’ clinical diagnosis, this paper proposed a fusion method to display the region of interest (such as the kidney or stone) which is extracted from series scan on the CT scout. Since CT scout can not show the kidney clealy, this method makes up the lack and generates a new image.25patients with CT scan were included in this study. We employed Patlak2-compartment model to calculate GFR. The measurement results from automated method and manual method were compared with the result from Cystatin C measurement. Results showed that, the correlation between automated method and cystatin C method in25patients was0.8029, and the correlation between the automated measurement and manual measurement was0.9518.6model pigs with12kidneys from MR scan were also included in this study. We employed the Patlak2-compartment model, IRF based3-compartment model and our improved model to calculate GFR respectively, and compared the results with that from SPECT. Results showed that the correlation were0.8136,0.8392and0.8168.The automated CT GFR measurement method can be potentially used for the highly improved efficiency. Moreover, it would be helpful to predict residual GFR following nephrectomy, and could also be used to predict the residual GFR for partial nephrectomy for tumor or stone surgery.Theoretical analysis and experiments showed that the improved GFR measurement model proposed in this paper can reduce the error caused by the substitute of contrast agent time concentration curve from aorta to renal blood vessels due to introducing the IRF of vascular compartment. Meanwhile, the improved model was simple to implement without the accurate segmentation of kidney cortex and medulla, which can reduce the impact from the segmentation errors; Comparing with the IRF based three-compartmental model, this model has fewer parameters and is easy to solve.