Region Descriptors to Extricate Renal Calculi by Discerning Artifacts from Ultra Sound Images

The study presents a new technique to identify kidney stones and differentiate it from artifacts from an ultra sound scanned image. Image analysis plays vital role in medical diagnosis. Analysing texture is a major source of discrimination in image analysis. There are situations when artifacts are mistaken for kidney stones in Medical scan using Ultra sound technique. We are proposing a new approach wherein the kidney stones are differentiated from artifacts using various shape descriptive parameters. Various descriptors like Compactness, Dispersion, Smoothness and third moment are used in our method to differentiate artifacts and stones from ultrasound images. Basic regional descriptor characterizes the geometric properties of ROI (Region of Interest) whereas Invariant moment concentrates on the density of ROI. Initially the US image is polished using a Combinational approach of noise removal and this image is taken as input for our further analysis. Pre-processing is done to minimize the occurrence of artifacts. The potential calculi are extracted and the validity is tested using the above specified parameters. Test results show convincing difference in values between Calculi and Artifact. The suggested method efficiently helped in identifying Renal Calculi from Ultra sound Images.


INTRODUCTION
Kidney stones otherwise called as Renal Calculi are becoming one of the common ailments.Calcium that is not used by the bones and muscles goes to the kidneys.In most people, the kidneys flush out the extra calcium with the rest of the urine.People who have calcium stones keep the calcium in their kidneys.Kidney stones are formed by crystal nucleation, growth and aggregation processes.
Small stones pass through the urine without any pain.But this stone becomes significant when the calcium, uric acid deposits are accumulated and when it attains a significant size.The patient will be aware of this only when this stone passes through urethra and gets stuck in the urinary tract which causes extreme pain.The blockage caused if the stone is not passed into the bladder by the peristaltic (spasmodic) action of the surrounding musculature will result in interruption of urine flow and subsequent renal damage (Coe et al., 1992).There are different types of kidney stones depending on its mineral components.The most common ones are Calcium oxalate, calcium phosphate, Struvite and Cystine stones.These stones may not be in its pure form.It will be the combination of many minerals.Struvite stones are found in individuals who have developed infection in urinary tract.These stones are called infection stones.It has been proposed that urine naturally contains substances that inhibit the development of large calcium oxalate crystals.Calcium oxalate kidney stones result when a normal crystallization-inhibitory system goes awry.Consistent with this assumption, studies have identified various urine components that can inhibit calcium oxalate crystal formation and/or aggregation (Nakagawa et al., 1983;Asplin et al., 1991;Shiraga et al., 1992;Atmani et al., 1996;Yamaguchi et al., 1993).Cystine stones are very rare and this stone formation runs in families.

PROBLEM STATEMENT
Ultrasound (US) scan is a popular scanning tool for any diagnosis.This technique is regarded as the safest and most adaptable method within the reach of common man in terms of cost.US scan is the first scan done by radiologists for any type of ailment.The complex physical interactions of the ultrasound beam with the human tissues results in some unexpected results in the ultrasound scanned images.Though US images are adaptable and comparatively safe, there are some acoustic interferences and artifacts in images.As a result of variations in the propagation speed of sound in different tissues and due to the bending and vibrations encountered by ultrasound energy as it traverses complex anatomical structures or due to the improper registration and display of information on the image due to limitations of the transducer, receiver or other components of the imaging system, some artificial stone like foreign bodies appear in US scanned image.The magnitude of the electromagnetic energy captured in a digital image is represented by positive digital numbers.Shadowing artifact is caused by partial or total reflection or absorption of the sound energy.A much weaker signal returns from behind a strong reflector like air or sound -absorbing structure like gallstone, kidney stone etc results in shadowing, which confuses the radiologists for a stone as it is a common belief that stones cast shadows (Kamaya et al., 2003;Sumarsono, Year;Vincent and Anahi, 2011;Behnam et al., 2010;Tsuicheng et al., 2013).Emphysematous Pyelonephritis is a rare and serious complication of urinary tract infection with extensive necrosis and gas formation in the renal parenchyma.Most patients (90%) have diabetes mellitus and 20% have associated urinary tract obstruction (Patel et al., 1992;Lipset et al., 1997;Joseph et al., 1996).In Patients suffering from Pyelonephritis, tendency of air formation are more and these air bubbles can cast shadow which appears as Stones in US images.All these problems, complicates the identification of renal calculi by the radiologists from Ultra sound scanned images.

LITERATURE REVIEW
There are few literatures which describe the significant work that have been carried out in this area.Ioannis et al. (2006) proposed an algorithm for renal stone detection based on two types of image features contrast and target shape.Statistical characteristics of these features are considered based on the images of kidney recorded during optimization of Extracorporeal Shock Wave Lithotripsy (ESWL) (Ioannis et al., 2006).Sridhar (2012) has suggested an algorithm wherein Calculi are segmented and identified using intensity profile.Algorithm detects calculus based on the shadow it casts in ultrasound image.The potential region of the shadow is detected to a smaller portion of a polygon.Tamilselvi andThangaraj (2011, 2012) have used seeded region growing method that performs segmentation and classification of kidney images with stone sizes using ultrasound kidney image for the analysis of stone and its early detection.The images have been classified as normal, stone and early stone stage by employing intensity threshold difference on segmented portions of the images to ease identification of multiple classes.Saurin et al. (2010) has used various content descriptive features like first and second order statistical parameters to classify different types of stones.Tamilselvi and Thangaraj (2012) had also suggested a Morphology based segmentation technique to extract the calculi present in ultrasound images.The above literatures have not discussed much on artifacts which is very crucial for the identification of stones.

PROBLEM SOLUTION
In our approach, we are introducing a new method to identify Kidney stones or Gall Stones and thereby differentiate artifacts and Calculi.The steps followed for our study is shown in Fig. 1.
Images for our study are collected from various Diagnostic centres.The raw images extracted from US Scan are initially pre-processed using a Combinational Approach of Noise Removing (CANR).Nasira and Ranjitha (2014a) had proposed a new technique of noise removing and smoothing (CANR) to remove the speckle noise and artifacts.Using this approach, maximum artifacts are removed and a clear image with only the possible calculi is obtained.In case of images with low contrast, Histogram equalization is performed to enhance the image (Nasira and Ranjitha, 2013).While pre-processing, we have ensured that the information for objective analysis is not disturbed as it is supposed to provide some imperative information for decision support and decision making.Since we have concentrated more on distinguishing Stones and Artifacts, we have extracted the ROI (Region of Interest) using cropping for image analysis.The intention of using multiple filters is to refine the image; thereby the image is populated with maximum occurrence of Calculi and least artifacts (Nasira and Ranjitha, 2014a).This is very much clear from the given Fig. 2.After thresholding, the stones and left out artifacts are displayed in Fig. 3. Our new findings are based on various regional descriptors to distinguish Calculi and Artifacts from an ultra sound medical image.This new approach helps the radiologists to reassure the results of their diagnosis.Any type of stone will have a rough surface and artifacts will be generally smooth.A region preserves the scalar measures based on its geometric properties.The simplest property is its size or area.The area of a region in the plane is defined as: where, I(x, y) = 1 if the pixel is within a shape, (x, y) ∈ S and 0 otherwise.In practice, integrals are approximated by summations.That is: where, ∆A is the area of one pixel.Thus, if ∆A = 1, then the area is measured in pixels.Area varies with changes in scale.One simplest property is to measure the perimeter of a region.If x (t) and y (t) denotes the parametric coordinates of a curve enclosing a region S, then the perimeter of the region is defined as: This equation represents the sums of extremely small arcs which defines the curve.x(t) and y(t) defines the set of pixels in the image.The above equation can be computed as: where, x i and y i are the co ordinates of i th pixel which forms the curve.Based on the above perimeter and area it is possible to describe the compactness of a region.Compactness can be measured as the ratio of perimeter to area: Which can be rewritten as: Here, the denominator represents the area of a circle whose perimeter is P(S).Thus, compactness measures the ratio between the area of the shape and the area of the circle that can be traced with the same perimeter.That is, compactness measures the efficiency with which a boundary encloses the area.In mathematics, it is known as the isoperimetric quotient.For a perfectly circular region compactness = 1, which represents the maximum compactness value.Low values of compactness are associated with convolved regions and for elongated shapes.We are taking the advantage of this property as calculi can never be a perfect circle.Another measure that can be used to distinguish regions is dispersion.Dispersion which is otherwise called as irregularity has been measured as the ratio of major chord length to area (Chen et al., 1995): where, ) , ( y x denotes the coordinates of the centre of mass of the region.Numerator actually defines the area of maximum circle enclosing the region, which defines the density of the region.Another approach to measure dispersion is to calculate the ratio of maximum to minimum radius which is in fact an alternative form of irregularity (Chen et al., 1995): The measure increases as the region spreads.Thus the irregularity of a circle is unity and that of a square is 2 Thus irregularity increases for irregular shapes and compactness decreases for irregular shapes.In image analysis, statistical moments plays a vital role in analyzing the texture of an image.Moments are used to portray a shape's sketch that is the arrangement of its pixels, like combining area, compactness, irregularity and higher order descriptions together.Moments are global description of a shape, accruing this same advantage.Basic regional descriptors characterises the geometric properties of the region whereas moments focus on density of the region.In our study we have taken the third order moment M3, which is an invariant descriptor and increases its value with increasing shape roughness.It reflects the asymmetry and uniformity of the histogram.
There are seven invariant moments, defined by Hu (1962), which is independent of rotation and is computed over shape boundary and its inner region (Keyes and Winstanley, 2001).
Relative smoothness, R is a measure of gray level contrast that can be used to establish descriptors of relative smoothness: where, σ is the standard deviation of ROI.Swift variations in brightness levels reflects high spatial frequency that is roughness and crouch variations in brightness levels are related to low spatial frequencies that is smooth areas.We have exploited only those properties which show some significant differences in values for our area of study.Image analysis with its various stages of output is shown in the Fig. 2. Figure 2 is the original ultrasound scanned image of kidney.This image is filtered using CANR (Fig. 3).The possible Stones are extracted using a threshold (Fig. 4).In Fig. 4, the thick arrow denotes a stone and dotted arrow points to an artifact.
Only the possible Stone/Artifact are segmented (ROI) from the image (Fig. 3) and the above region descriptor values (Eq. 1 to 9) are computed.We have found significant changes in values of density, Compactness, dispersion, smoothness and third moment in case of stones and artifacts which is shown in Table 1 and 2 in below section.We have already proposed in our earlier work that the density of ROI also plays a vital role in the analysis of stone and artifact (Nasira and Ranjitha, 2014b).In our earlier work we had used a statistical approach to weigh each pixel intensity according to its likelihood of being a stone or an artifact.The preprocessed image is segmented and each segment is tested for the density of the pixels in that ROI (Region of Interest) using mean.It has been observed that Mean of the probable stone is comparatively higher than the artifacts.Measuring the average value of the area gave the density of the pixels in that ROI.The Size of the stone is also computed as the total number of the pixels present in the area and multiplying the number of pixels with the dimension of one pixel (Zhou et al., 2001;Yang et al., 2004;Salman et al., 2005).

RESULTS AND DISCUSSION
This section presents the experimental results that are conducted in a MATLAB environment with around 30 Ultrasound images.The comparison of region descriptors like Compactness, dispersion, smoothness and third moment is shown in Fig. 5 to 8. Table 1 and 2 shows the sample test results of ten images of our experiment.It has been observed that Compactness of Artifact lies between 0.001 and 0.005 whereas for stone the range is between 0.0005 and 0.001.Low values of compactness are associated with irregular shape which is far from a perfect circle (Nixon and Aguado, 2008).Similarly, Dispersion for artifact is always less than 100 whereas for stone it was greater than 100 which entail that as irregularity (far from circle) increases dispersion also increases.Stones are not perfect circles, it has peaks and valleys whereas an artifact when compared to stones will be smooth and hence the dispersion values are less.
Likewise smoothness is less than 0.025 for artifacts whereas for stone it was greater than 0.025.The third moment values for artifact were within the range 300-3000 whereas it is greater than 3000 for stones.This invariant moment increases its value as the roughness increases and decreases for smooth images.It also specifies the density of ROI.Density of Stone is more than artifact.A comparison of all these parameters is presented in Fig. 5 to 8.

CONCLUSION
Objects are nothing but collection of pixels in a region which can be grouped according to some properties.From these properties, we can compare and recognize objects in an image against the values of known objects.We have introduced a new method to extract Renal/Gall stone from ultra sound Images.The proposed method also differentiates stones from artifacts by putting these two entities into two separate region categories.It can also serve as a method to distinguish the type of Stone from US images under the common assumption that Calculi stones have spikes whereas Uric acid stones are smooth.This method is based on the regional descriptors which are generally used to distinguish shapes.Depending upon the value of Compactness, dispersion, Smoothness and third moment, we have arrived at the result with fair amount of accuracy.

Fig. 4 :
Fig. 4: Thresholded image-the thick arrow denotes a stone and dotted arrow an artifact

Table 1 :
Experimental result of sample artifacts of US image

Table 2 :
Experimental results of sample calculi of US image