Image Processing Techniques for Characterization of Renal Calculi Using Ultrasound Images

Abstract

Kidney is the main organ of the urinary system. It is the processing plant for waste disposal. If the waste to be disposed is more and the fluid volume flushing it is less, there are more chances of clogging. The kidney stone is formed out of accumulated toxic waste. The composition of kidney stones may vary from person to person. Its size may be as small as a grain of sand or as large as ping-pong ball. Some of them are smooth, some have horns. The medical name for kidney stone is renal calculi. Urinary stone disease is an extremely painful condition and it can be formed with calculus, found at number of locations, types and shapes in the urinary system, principal sites being the lower calyx pelvis of the kidney, ureter and the bladder. Nine different types of renal calculi have been identified based on their chemical compositions namely Calcium Oxalate Monohydrate(CaC2O4H2O), Calcium Oxalate Dihydrate(CaC2O42H2O), Magnesium Ammonium Phosphate Hexahydrate (struvite)(MgNH4PO46H2O), Hydroxyapatite(Ca10(PO4)5(OH)2), Calcium Hydroxyapatite Phosphate Dihydrate (Brushite) (CaHPO4.2H2O), Tricalcium Phosphate (Whitlockite)(Ca3(PO4)2), Uric acid(C5H4N4O3), Cystine[-SCH2CHNH2COOH]2 and Xanthine(C5H4N4O2). Detection of renal calculus presence in the early stages of growth is beneficial in the appropriate management of the condition that can take place, with options which include removal of calculus. The investigation on presence of renal calculi can be done using appropriate imaging modality including Ultrasound Scan or Computer Tomography Scan or Magnetic Resonance Imaging. Computer Tomography uses x-radiation, which is an ionizing radiation. Magnetic Resonance Imaging uses powerful magnetic field. But, out of these three imaging modality, ultrasound has been most preferred imaging modality as it is not harmful radiation and also it is cheaper. Limitation of existing ultrasound system is that, the images acquired are of low resolution and do not reveal meaningful information about type, location and shape. Hence to decide the mechanism for removal of renal calculi, radiologist have to refer CT Scan or Magnetic Resonance Imaging scan of the region, which reveals more information to infer about the type of the stone or hardness of the stone based on CT number. This information is required to decide the most appropriate medication required for removal of stone or number of shocks required for fragmenting the stone in micro granules during the procedure of Extracorporeal Shock Wave Lithotripsy (ESWL).At present there is absence of the system to support physician‘s decision about the type of renal calculi directly from ultrasound imaging. This research work focuses on an approach to characterize renal calculi directly from ultrasound images in grey scale form (8-bit depth) acquired using ultrasound scanner to assist physician for detecting type of renal calculi and deciding most appropriate procedure for removal. The prime objective of this research is to find set of parameters for texture characterization of renal calculi, with various compositions. The main objective of this research is to present an approach towards development of system which will use images captured by ultrasound scanner, perform image processing and assist radiologist to support his/her conclusion about type of renal calculi directly from ultrasound images. To fulfil this objective, the researcher has collected 112 real renal calculi, which are removed after surgery from patients from three (03) hospitals namely kidney hospital, Nadiad, LIONS hospital, Mehsana and Bombay Hospital, Mumbai, which are also geographically located at distance. Image acquisition set up used is at Muljibhai Patel Urological Hospital (Kidney Hospital) under supervision of Dr. Lalit Panchal, Head- Department of Radiology. HONDA make ultrasound scanner (Model HS 2000) is used to acquire image using 3.5 MHz probe. The acquired image is firstly stored in memory of scanner and then transferred to Desktop computer where Matlab and ImageJ software are installed for image processing. The renal calculi are then removed from balloon and sent for FTIR (Fourier Transform Infrared) analysis to METROPOLIS Laboratory, Mumbai. The report generated by this FTIR analysis instrument is used for comparison with the type identified by the proposed method to verify sensitivity and specificity of the approach. To begin with, the research is started with evaluation of first order statistical parameters (from 8-bit depth grey scale images after cropping the image with Region of Interest-ROI) like range of intensity i.e. minimum and maximum value of intensity, mean, skewness which is a measure of degree of asymmetry, kurtosis which is a measure of peakedness of distribution. Observation based on numerical results obtained from all 112 images shows that there is a good discrimination in both skewness and kurtosis for different types of renal calculi. To enhance the above observation about discrimination among different types of renal calculi, it is required to evaluate second order statistical parameters using Grey Level Cooccurance Matrix (GLCM) method, which is introduced by Haralick etal. The GLCM indicates how often different combinations of grey level occur in given image. The features, evaluated using GLCM are; (1) Entropy (2) Contrast (3) Correlation (4) Energy and (5) Homogeneity. These features are used quantitatively to characterize texture of the image. Observation based on numerical results obtained from all 112 images shows that there is a good discrimination using entropy and energy for different types of renal calculi. To investigate texture patterns further, Fast Fourier Transform (FFT) is used. FFT is essential to find intensity distribution between two neighbouring pixels i.e. frequency distribution in the whole image. All stones are analysed to find out distribution of low and high frequency distribution in the whole image. The values of low and high frequency power are also evaluated from the frequency spectrum. Observations based on numerical results obtained from all images have shown that distribution of frequency (High and Low both) is different for different types of renal calculi. Wavelet transform has good time- frequency characteristics, which is effective in extracting information from the image. Three wavelets namely Daubechies, Symlet and Coiflet are applied on all images and from that approximation and detailed coefficients are evaluated. Observation based on numerical results obtained from all images shows that there is a good discrimination using both coefficients for different types of renal calculi. Next, a novel concept of non-separable wavelet namely quincunx is applied, which is a significant tool in texture characterization. Observation based on numerical results obtained from all images shows that there is a good discrimination after decomposition of image spectrum in Low and high frequency components. Interestingly when HH images (image reconstructed after high frequency components) are displayed on screen, different patterns are also observed visually for different types of renal calculi. After evaluating all parameters, Dual Simplex Method of Liner Programming is used where objective function is set to minimize the weighted sum of symptoms within the specified threshold, with the constraint for sensitivity and specificity of training cases. By optimization algorithm weight factors for different stone categories are calculated which help in classification of any unknown stone in to corresponding category. Objective function is defined as; Maximize or Minimize Z = W1 X1 + W2 X2 + .......+ Wn Xn Where; Wj = Weight factor for parameter j, for specific type of stone Xj= Multiplying factor indicating presence of symptom by 1 and absence by 0 The sensitivity and specificity for six types of renal calculi is evaluated from samples collected. It is observed from the results that use of evaluated parameters can be used to judge the type of renal calculi with 100% sensitivity (selecting the correct type of renal calculi from results) and 100% specificity (rejecting the other type of renal calculi) with the group of sample stones considered for this research.