Automated Region of Interest Detection Method in Scintigraphic Glomerular Filtration Rate Estimation
Abstract: The glomerular filtration rate (GFR) is a crucial index to measure renal function. In daily clinical practice, the GFR can be estimated using the Gates method, which requires the clinicians to define the region of interest (ROI) for the kidney and the corresponding background in dynamic renal scintigraphy. The manual placement of ROIs to estimate the GFR is subjective and labor-intensive, however, making it an undesirable and unreliable process. This work presents a fully automated ROI detection method to achieve accurate and robust GFR estimations. After image preprocessing, the ROI for each kidney was delineated using a shape prior constrained level set (spLS) algorithm and then the corresponding background ROIs were obtained according to the defined kidney ROIs. In computer simulations, the spLS method had the best performance in kidney ROI detection compared with the previous threshold method (Threshold) and the Chan–Vese level set (cvLS) method. In further clinical applications, 223 sets of 99mTcdiethylenetriaminepentaacetic acid (99mTc-DTPA) renal scintigraphic images from patients with abnormal renal function were reviewed. Compared with the former ROI detection methods (Threshold and cvLS), the GFR estimations based
on the ROIs derived by the spLS method had the highest consistency and correlations (r=0.98, p<0.001) with the reference estimated by experienced physicians. The results indicate that the proposed automated ROI detection method has great potential in automated ROI detection for accurate and robust GFR estimation in dynamic renal scintigraphy. Existing system: Segmentation process robust against misleading information resulting from noise, clutter, and occlusion. Several studies have exploited the shape prior to refine the segmentation in medical images, including CT, MRI, and ultrasound images. It is obvious that the anatomical rough shape of a low-functioning kidney would have a high similarity to a normal kidney. In this condition, the accuracy of ROI detection would be improved by the segmentation method using the intensity information of the renal scintigraph along with the shape prior of kidney. In this paper, an automated ROI detection method was proposed by introducing the shape prior into the level set algorithm to achieve the accurate scintigraphic GFR estimations. Proposed system: To calculate the GFR for kidneys via the Gates method, the accurate regions of interest (ROIs) for both the kidneys and the corresponding background must first be delineated. Although it is time-consuming and subjective to manually delineate the renal contours, most clinicians usually adopt the manual ROIs in the Gates method. Sometimes the results of the manual ROIs are hard to reproduce, even for the same operator. To address these problems, the procedures using different image processing algorithms have been applied directly to obtain the ROIs for kidneys automatically or semi-automatically. Thres holding techniques were proposed in the early days. The single-threshold method is the simplest method, but it applies only to high-contrast use. The double-threshold method is an improvement and twice executes different thresholds for segmentation based on the manually identified center of the kidney. These thres holding techniques would fail to detect kidney ROI in complicated low-contrast images from patients with low renal function. A semi-automated ROI detection method was developed to improve the accuracy of ROI detection by manually placing preliminary rectangular ROIs for each kidney and the corresponding background area between the kidneys.
Advantages: Then, for the kidney template acquisition, the Daubechies wavelet algorithm and arithmetic average fusion rule was used to combine the regional images, which were derived from the 50 cases by the modified contours of the kidneys. The level set method is a powerful tool used to perform contour evolution in image segmentation. The Chan-Vese level set (cvLS) algorithm has been proposed to segment images with an ambiguous edge by integrating the ideas of the level set and Mumford-Shah model The spLS algorithm as well as the cvLS algorithm was both applied to segment the kidney in the simulated composite images. In addition, the threshold with the edge detection method proposed by Tian (Threshold) was also used to extract the kidney for further comparison. All of these three approaches (spLS, cvLS, and Threshold) were evaluated for the kidney ROI delineation under different noise levels. Disadvantages: Sometimes the results of the manual ROIs are hard to reproduce, even for the same operator. To address these problems, the procedures using different image processing algorithms have been applied directly to obtain the ROIs for kidneys automatically or semi-automatically. The double-threshold method is an improvement and twice executes different thresholds for segmentation based on the manually identified center of the kidney. These thres holding techniques would fail to detect kidney ROI in complicated low-contrast images from patients with low renal function. A semi-automated ROI detection method was developed to improve the accuracy of ROI detection by manually placing preliminary rectangular ROIs for each kidney and the corresponding background area between the kidneys. Modules: Glomerular filtration rate: The glomerular filtration rate (GFR) refers to the amount of filtrate generated by a kidney per unit of time; it is an important and regular index used to evaluate renal
function in clinical routines. The GFR can be accurately determined in vitro using blood-sampling techniques. These in vitro approaches, however, are usually tedious and time-consuming for high-throughput clinical applications. Moreover, blood-sampling techniques are invasive and cannot provide separate renal function information for each kidney. Conversely, the GFR also can be measured in vivo using medical imaging techniques, such as nuclear medicine imaging, computed tomography (CT) imaging, and magnetic resonance imaging (MRI) . In light of its ability to provide functional information, nuclear medicine imaging using 99mTcdiethylenetriaminepentaacetic acid (99mTc-DTPA) is the most commonly used method to determine the GFR in clinical routines. It was found that 99mTc-DTPA plasma clearance was significantly associated with the inul in clearance rate that is recognized as the gold standard to measure the GFR. In this condition, the accumulation of 99mTc-DTPA in each kidney at 2- to 3-min time intervals after tracer administration is proportional to the GFR. According to this proportional relationship, the Gates method is routinely used to estimate the GFR. Regions of interest: To calculate the GFR for kidneys via the Gates method, the accurate regions of interest (ROIs) for both the kidneys and the corresponding background must first be delineated. Although it is time-consuming and subjective to manually delineate the renal contours, most clinicians usually adopt the manual ROIs in the Gates method. Sometimes the results of the manual ROIs are hard to reproduce, even for the same operator. To address these problems, the procedures using different image processing algorithms have been applied directly to obtain the ROIs for kidneys automatically or semi-automatically. Thres holding techniques were proposed in the early days. The single-threshold method is the simplest method, but it applies only to high-contrast use. The double-threshold method is an improvement and twice executes different thresholds for segmentation based on the manually identified center of the kidney. These thres holding techniques would fail to detect kidney ROI in complicated low-contrast images from patients with low renal function. A semi-automated ROI detection method was developed to improve the accuracy of ROI detection by manually placing preliminary rectangular ROIs for each kidney and the corresponding background area between the kidneys. Acquisition of the Kidney Template:
99mTc-DTPA renal scintigraphy with normal or relatively good renal function has sharper boundaries for the kidneys, which contains the prior information of shape for kidney ROI detection. According to the required image quality, 50 cases of 99mTc-DTPA renal scintigraphy were selected by an experienced physician to extract the kidney template with the shape prior. In all 50 cases, data were first acquired by performing image preprocessing and derived the rough contours of the kidneys by the level set method automatically. The obtained rough contours of the kidneys were then modified by an experienced physician. By using the standard scanning protocol and the same scanner (the scanning produces would be descripted in the following section about clinical data acquisitions) in this study, the position of the kidneys would maintain a relatively constant position in the renal scintigraphic images. The sizes of kidneys could be considered as in the comparable size in light of the same field of views of the images and the similar body sizes of patients. Then, for the kidney template acquisition, the Daubechies wavelet algorithm and arithmetic average fusion rule was used to combine the regional images, which were derived from the 50 cases by the modified contours of the kidneys. Clinical Data Acquisition: Within the local clinical database, the clinical data were reviewed from 223 patients (93 women, with a median age of 52 years, ages ranging from 22 to ~80 years, and 130 men, with a median age of 57 years, ages ranging from 19 to ~86 years), who had at least one kidney with a single renal function (SRF) <50 mL/min/1.73 m2. Considering clinical indications, all of the patients had undergone 99mTc-DTPA renal scintigraphy for disease diagnosis. Dynamic renal scintigraphy was performed in accordance with the clinical guidelines for nuclear medicine. A 1-min pre-injection syringe count was initially determined followed by an injection of a dose of 110 MBq of 99mTc-DTPA. Then, dynamic renal scintigraphy was performed in a supine position using a dual-headed rotating camera with a low-energy collimator (Philips Precedence 6 SPECT-CT scanner, Amsterdam, Netherlands). The collimator face was as close to the patientâ&#x20AC;&#x2122;s body as possible in order to obtain the best resolution. The posterior view of both kidneys and bladder was acquired in a planar imaging sequence. The imaging sequence was composed of the perfusion phase (a series of 30 2-s frames) and the functional phase (a series of 80 15-s frames). At last, a 1-min post-injection syringe counts.
The acquired images were all recorded in a 64 Ă&#x2014; 64 matrix (pixel size 9.328 mm Ă&#x2014; 9.328 mm). Data Analysis: The acquired clinical data were analyzed using the routine procedure provided by Philips Healthcare (Andover, MA, USA; Best, Netherlands). As in the routine clinical procedure, the experienced physicians who had at least 3 years of experiences in nuclear medicine delineated manual ROIs for the kidneys and the corresponding backgrounds in the workstation and then given the official clinical report. The clinical reports were then double checked by another senior physician who had more than 10 years of experiences in renal evaluations. We considered the GFR measured by the experienced physicians to be the reference (i.e., gold standard) for the following data analysis and noted this as GFR_ref. The automatic ROIs for the kidneys and the corresponding backgrounds were first obtained by following the proposed automated ROI detection method for GFR estimation. For the comparisons, the automated ROI detection method with the original cvLS algorithm was also used to obtain the automatic ROIs in the GFR estimation. Furthermore, the previous semi-automated ROI detection method was adopted to derive the semi-automatic ROIs to measure the GFR. All of the GFR estimations were conducted by the Gates method. GFRs estimated based on the ROIs derived by the three methods (spLS, cvLS, and Threshold) were respectively noted as GFR_spLS, GFR_cvLS and GFR_thr.