Koichi Kamura

Skeletonization of renal cysts of autosomal dominant polycystic kidney disease using magnetic resonance imaging

Autosomal dominant polycystic kidney disease (ADPKD) is a condition in which numerous cysts develop in the renal tubules and grows over the lifetime, resulting in compression of renal parenchyma and worsens renal function. Conventionally, total kidney volume (TKV) is bluntly estimated as a time cross sectional parameter of disease status. In assumption that cyst initiation rate and growth speed of cysts would be another prognostic marker for renal function, and regulate the morphological feature of the enlarged kidney, we attempted to extract a morphological feature of the renal cystic region quantitatively from MRI T2-weighted images. Skeletonization algorithm was applied to the binarized cystic region after extracting cystic regions by discriminant analysis. Then, morphological feature of the renal cystic region was converted to distribution pattern in number and size of cysts, and “branch” of adjacent cysts. The number of “branches” corresponded with the number of cysts, and the cumulative probability curves of “branch” length shifted according to cyst size distribution. The proposed method successfully quantified morphological feature of cystic region objectively in semi-automatic manner. The method would contribute to manage ADPKD patients in deciding time to start therapies after affirmation for consistency with cyst initiation rate, growth speed of cysts and renal function. Introduction Autosomal dominant polycystic kidney disease (ADPKD) is a hereditary, progressive disorder that is characterized by a gradual increase of cysts and kidney volume with subsequent renal insufficiency. The Consortium for Radiologic Imaging Studies of Polycystic Kidney Disease (CRISP) study has shown that total kidney volume (TKV) can predict the progression of increasing TKV and subsequent renal insufficiency. However, the good correlation between TKV and the prognosis of this disease is confined to only when kidney enlargement advances obviously. While many patients with ADPKD require hemodialysis by the fifth to seventh decade of life, some patients do not develop renal insufficiency in their lifetime. There are substantial individual differences in the long-term progression of the disease that also make a prediction of the likelihood of renal insufficiency in the early stage difficult, and should be solved to reduce burden of therapies in both physical and economical aspects. In the early stage of the disease, the renal cysts are considered to develop and enlarge without the increasing of TKV. In the previous report, a growth speed of cysts was indicated to be a factor responsible for increase of TKV. In contrast, another report revealed that growth speed of cysts was rather uniform, and earlier initiation of cysts was potent prognostic factor for developing renal insufficiency in relation to cytogenetical findings. However, the methods to estimate such prognostic factors are limited to cytogenetical investigation or use of mathematical models that require manual counting of numerous cysts. It could be assumed that multifocal and constant sprout of cyst would determine morphological appearance of the kidney afterward, the affinity of cyst initiation in renal tubules would be diagnosed by analyzing temporal images of the enlarged kidney, especially using magnetic resonance (MR) T2-weighed imaging. Since the most cysts are agglutinated and their boundary are often unclear, shape, number of each cyst and agglutinated condition should be converted to numerical parameter before analysis. The skeletonization algorithm is used to convert shapes into a group of lines for computational recognition and tracking of objects, and the morphological analysis using a thinning and skeletonization algorithm has been widely used for medical applications-. The purpose of this study is to develop a software program that contains a semi-automatic binarization of renal cyst regions in an MRI T2 image and a skeletonization algorithm, and validated acquired numerical parameters using cyst-like geometrical models and MR images of patients with ADPKD. Materials and Methods The MRI were performed as routine clinical practice using MAGNETOM Symphony (Siemens, Erlangen, Germany), and the images were provided for this study under approval by the local ethics committee of Chiba East Hospital and Chiba University Hospital. The imaging protocol was T2-weighted imaging (Half-Fourier single-shot turbo spin echo: HASTE), with coronal cross sections, slice thickness of 4 10 mm, and spatial resolution of 1.67 mm/pixel. One image for processing was selected from each image series containing the renal pedicle, and the pixel intensity was compressed into an 8-bit value range. The algorithm of image processing was as follows and is shown in Figs. 1 and 2. The first step was image binarization of the cystic region. Since the intensity of cystic region varies widely and cause over and under extraction in the binarization process, a look up table (LUT) was applied as contrast enhancement for the mid-range values. The LUT curve was generated using a sigmoid function as follows, ς(x) = 256 1+e−0.05(x−b) (1) where ς(x) is the output value against the input intensity value x, and b is a value that determines the inflection point of the sigmoid curve. The adjustment of b is mentioned in the next section. To determine a proper threshold value for cyst region extraction, outlines of both kidneys were pre-segmented manually by a urologist as two polygonal regions. Then a threshold value was determined to divide renal parenchyma and cyst into two classes using a discriminant analysis according to Otsu’s method. In the second step, the morphological analysis was processed to the binarized cystic area to quantify the features of the cystic region, such as density and size distribution of the renal cysts. In this study, a skeletonization algorithm was applied to the binary image that would show the morphological characteristics of the cysts without recognizing and counting individual cysts.