Description:Field of the Invention
With its vast application arena for numerous clinical problems, medical image processing is an interdisciplinary research area that has shown tremendous growth in the past few decades. It is a compilation of knowledge from several fields, including science, biology, engineering, statistics, physics, and computer science. Computerized systems that draw on the aforementioned scientific disciplines have been developed through the use of Medical Image Processing to address medical diagnostic issues. People all over the globe use the transformed photographs after having been saved in one of the aforementioned formats. Image segmentation entails dividing a picture into several areas with shared characteristics. Separating a picture into its component parts is an essential first step in automated image analysis. Before enhancing medical photos, segmentation can be done to make the enhancement process faster and more accurate. Delineating one or more structures of interest within an image is called image segmentation. Avoiding the tedious task of manually contouring the structures calls for automated techniques. Using lung cancers as an example, the procedure becomes even more challenging. While the range of tumor intensities does overlap with that of healthy tissues, the majority of tumors exhibit heterogeneity in appearance.
Objective of the Invention
The objective of the invention includes designing a feature extraction-integrated classification method capable of handling massive datasets. The median filter is proposed to get rid of the background noise. The reconstruction operators are used to recreate the gradient image after introducing a morphological gradient in the watershed segmentation process.
Background of the Invention
Ensuring that input photos are of excellent quality is crucial for medical image processing algorithms to work well. The medical photos are marred by artifacts, mistakes, and noise. The ability to discern anatomical features is severely impaired in low Signal-to-Noise-Ratio (SNR) or contrast images. These problems are addressed by the filtering technique, which eliminates background noise and other undesired signals resulting from imaging (Shuqian Luo 2001). In (Arnold et al. 2006), the Low Dose CT images were filtered using the Noise Variance (NOVA) algorithm. It was the overarching structure for nonlinear filtering, which makes use of a spatially dependent noise variance estimate in images. Iteratively, the NOVA filter filters the image while estimating the local noise. This filter was designed with the particular goal of quantifying emphysema in low-dose CT images by utilizing earlier data. The NOVA filter requires multiple iterations and has many parameters. Iterative schemes are adjusted using parameter settings when no information about the photographed object is known. In two steps, we determined the kind of noise that was present in the medical picture. Two steps were taken: first, a criterion was employed to identify the impulsive noise; second, the intensity variation between pixels was decreased. The median filter alters the average intensity value of the picture if the distribution of spatial noise is not uniform within the viewing frame. You can think of the adaptive median filter as a hybrid of the median and mean filters.
The process of image segmentation involves dividing a picture into areas that represent distinct items or things' components. A connected region is formed by pixels in the same category that have similar grayscale multivariate values. and the values of neighboring pixels in various categories are different. What follows is a summary of some prior research on picture segmentation. Using a levelset-based segmentation technique, the authors of the study by Mohammad et al. (2011) were able to isolate the relevant areas of several medical photographs. To begin, we need to determine the input image's threshold so that every pixel falls below it and the remaining values are used as the original image. With this method, all values, with the exception of those pixels that aren't relevant to the analysis, remain unchanged from the original image, preserving its original features. After that, a morphological technique is employed to eliminate a few minor sections that can be ignored. The segmentation process is completed by employing the variational level set method. When applied to big image sets, the thresholding-based segmentation method's low accuracy and excessive computing time became apparent. In order to determine the contour line of the desired area, the edge-based method first finds the object's contour points and then, using a predetermined concept, links them. There are two distinct phases of region-based segmentation. The first is identifying the image's primary edges by topological gradient restoration. Step two involves running the watershed method using the topological gradient as input rather than the typical morphological gradient. This is done after the first step has been completed. Xing et al. (2011) detailed a technique for segmenting CT scans of liver tumors.Using a 6-neighborhood technique, the CT volume was subdivided into smaller areas using the watershed transform. Compared to the total number of voxels, the number of regions was reduced by a factor of two. Next, tumors from liver parenchyma are extracted using a support vector machine (SVM) classifier trained on user-selected seed points. A feature vector is then generated for training and prediction using each small region formed by the watershed transform. To enhance the segmentation outcome of SVM classification, morphological procedures are subsequently applied to the entire segmented binary volume. Finding gradients and extracting features have been challenges with this strategy. Specifically, the authors Benmazou et al. (2012) created a Cellular Automaton (CA) and a Moore neighborhood to address the issue of picture segmentation. Each agent will have access to two states, and cellular automata are varied using two transition functions. The utilization of the NetLogo platform for mammography validation was employed. A group of cells is controlled by a local rule known as the rule of transition, which governs how the cells behave. One of the characteristics of CA is its symmetry, which means that all cells follow the same rules and have the same topology. Another characteristic is its ability to describe both space and time.
Watershed transformation was demonstrated by Mehena et al. (2015) for the purpose of segmenting and extracting brain tumors. Using MR brain imaging, this process entails finding the tumor, locating it, following it, improving it, and finally recognizing it. In this case, the MR brain image is sharpened using the sobel edge emphasizing filter. Using the initial image's gradient, the watershed transform was calculated. The watershed transform is a segmentation algorithm with two drawbacks: over segmentation and an inability to discern important sections with low contrast borders. Effective pathology knowledge and comprehension of the MRI image's intensity and form are prerequisites for the FCM and K-means clustering methods described in (Kauser et. al., 2016) for brain tumor segmentation. The fact that every tumor is unique in terms of size, location, shape, and severity presents the biggest challenge to segmentation. Feature extraction was also impacted by this algorithm's primary testing domain: brain pictures. For the purpose of segmenting CT images of lung nodules, Ezhi et al. (2016) created a region-based active contour model with the Fuzzy C-Means method. To begin, the nodules in the lungs are segmented from the rebuilt CT lung image. Clustering was utilized for nodule segmentation, and Selective Binary and Gaussian Filtering (SBGF) with Signed Pressure Force function (SPF) was used for parenchyma reconstruction. This approach does a poor job of detecting the edges of medical images. The preprocessing stage involves enhancing and normalizing the images to the same scale in order to automate the segmentation of brain images containing glioma tumors (Kadkhodaei et al., 2016). These images can vary greatly in size, shape, and appearance. 3D super-voxels are used for intensity-based image segmentation after the enhancements. The saliency map was used to improve the tumor borders, and an edge-aware filtering technique was used to align the edges of the original image with it. Afterwards, a collection of strong texture features is retrieved from super-voxels for the purpose of tumor classification in brain imaging. Identifying the borders or pixels between distinct areas from which intensity is extracted is the first step in picture segmentation, which aids in both decision-making and the resolution of image segmentation problems.
Summary of the Invention
Despite the proposal of multiple techniques for tumor classification, the most important problem in medical image processing remains tumor classification. The volume is rapidly increasing as a result of medical data being digitalized. When presented with massive amounts of data, current categorization systems perform poorly. Current classifiers necessitate an independent feature extraction methodology, the creators of which dictated the minimal amount of features. Therefore, a feature extraction-integrated classification method capable of handling massive datasets is required. There is a higher chance of noise in the capture and transmission of pictures acquired by electronic devices, such as CT scanners. Because many different kinds of noise can damage images acquired after transmission, de-noising is a necessary step before performing tasks like segmentation, feature extraction, and classification. To get rid of the background noise, the median filter is applied here. Once the noise has been removed from an image, it can be divided into regions using the morphological watershed transformation process. The reconstruction operators are used to recreate the gradient image after introducing a morphological gradient in the watershed segmentation process. Next, we eliminate the low-value gradient pixels while keeping the high-value ones.
Brief Description of Drawings
Figure 1: (a) Catchment Basins (b) Watersheds
Figure 2. Watershed segmentation procedure
Detailed Description of the Invention
Segmenting a picture into its component parts by clustering nearby pixels is known as image segmentation. If you want to put it another way, segmentation is a method for pixel classification that lets you create image sections with similarities. Medical photographs, for example, use segmentation techniques to isolate specific anatomical or functional features. The goal of this process is to reduce the overall picture size while still preserving relevant semantic information. The segmented image clearly shows the boundaries of all the items in the original image. Pixels belonging to the same class create a contiguous region with similar grayscale multivariate values; this process is called segmentation. The values of adjacent pixels in different categories will be different. After the image has been filtered, a segmentation method should be applied to it. It is possible to enhance the outcomes of segmentation by applying mathematical morphological methods. Quantitative data will be retrieved from the pictures using the segmentation outcomes. There are a variety of approaches to picture segmentation, such as algorithms that rely on thresholds, edges, regions, level sets, clustering, or graphs. In terms of computational complexity and quality, these segmentation approaches are distinct from one another.
The idea of homogeneity is central to region-based segmentation, which takes into account the fact that pixels within a region tend to cluster around shared traits while pixels outside of it tend to exhibit more diverse patterns. Similarity checks are performed on each pixel by comparing it to its nearby pixels in terms of characteristics including shape, color, texture, and grey level. The zone is expanded by adding that specific pixel to the existing one if the result is positive. To understand how this method works, consider the whole area that an image occupies as R. Then, to segment an image, think of segmentation as a process that divides R into n smaller parts, R1, R2,..., Rn. Assuming they add up to a connected set, the two areas and are considered neighboring. Every pixel needs to be in a region in the first stage, and then the points in that region need to be connected in some predefined sense. Once the regions have been separated, the following step is to specify the properties a segmentation region's pixels must meet. So long as the brightness of each pixel is uniform. The basic challenge of segmentation is to divide a picture into areas that meet the aforementioned criteria. There are two main types of region-based methods: region-growing and region-split and merge. Picking out a specific pixel in the image is the first step. Due to its focus on overall segmentation quality, making the right option is of the utmost importance in this approach. What follows is a general approach for an algorithm that grows regions. To begin with To begin the segmentation process, choose a pixel inside the image. Choose criteria for expanding the region, and then add a pixel that is 8 pixels in size and connected to at least one other pixel in the region. After you've tested every pixel for allocation, label all the regions. Combine areas that have the same label. Lastly, if the difference value is below a given threshold, the pixel will be allocated to a specific region, let's call it Si. Otherwise, the pixel will be allocated to a new region, Sj.
A ridge that separates land areas drained by several river systems is the physical basis for the topographical notions of watersheds and catchment basins. Rivers and reservoirs receive their water from an area called a catchment basin. By bringing these concepts to grayscale image processing, the watershed transform is able to address a wide range of picture segmentation issues. The grayscale picture is transformed into a topological surface with the values of f (x, y) represented as heights using the watershed transform. The watershed transform uses a grayscale picture to identify the catchment basins and ridge lines. A new image is created from the original, with the items or places to be identified serving as catchment basins. You can think of a watershed as the boundary between two or more catchments. The watershed marks the limits of nearby catchments. As seen in Figure 1, the minimum is used as a marker for watershed zones and can be seen as contours. Watersheds are defined by the boundaries of the multiple catchment basins that are created by the process. When it comes to processing speed and precision of edge detection, the watershed transforms are the most efficient. You can use the watershed method on a distance modified image, an edge enhanced image, or even on the original image itself. A topographic surface is used to comprehend the watershed transform of a grayscale image.
Mathematical morphology is an efficient non-linear method for extracting object components from images by identifying its boundaries and framework. When you use morphological operations to generate a tiny form as a picture structural element, you can then place that shape wherever you choose on the image and use querying to compare each pixel's neighborhood to the rest of the image. In order to choose the neighborhood's size and shape, the morphological operation is defined; this shape selection inside the input image is quite sensitive. A morphological technique involving the structure elements disk and radius is used to extract the image's backdrop. It is dilation and erosion that are the most typical morphological processes. Grayscale photographs can be subjected to dilation, a morphological technique that enlarges objects by stretching their boundaries. The already-miniscule openings in the expanding areas get even more so. Grayscale images can be treated using erosion in the same way as dilatation. As erosion takes place, pixel size decreases and hole size increases in the affected area.
The term "segmentation" refers to the method of extracting and expressing data from images in order to cluster pixels according to their shared characteristics. The segmentation of CT lung images using the watershed morphology technique is shown in Figure 2. Following is the process flow of the watershed morphology algorithm. Initialize the location of the image to the segmentation using the input image received from the median filter. Following the detection and extraction of vertical and horizontal edges, the tumor regions are located using the gradient approach. An method for morphological watersheds Set up the starting point for the segmentation using the scalar function. Locate the picture's borders Use the Gradient Application Technique Locations of Minimum and Maximum, Laplacian Regions, and Other Areas Aerial Watershed Retrieved Results Phase of preprocessing. Topographic relief is produced after locating an image's gradient, which allows for the determination of its minima and maxima regions. And lastly, the segmented picture is delivered. One of the primary goals of segmentation is to identify the main edges of the damaged image; the second is to calculate the topological watershed of the identified gradient. After being divided into several areas and edges, the CT lung picture can be viewed as a topographical surface. To begin, the position’s likelihood of falling into a single minima zone is highest when it is in a position to fall, and the position’s likelihood of falling into more than one such region is highest when it is in a position to fall. At the second stage, known as the watershed basin of the minimum, a set of positions must be satisfied. Once the third step is met, a summit line, also called a watershed line, is drawn on the topographic surface. , Claims:The scope of the invention is defined by the following claims:

Claim:
1. The System/Method for Image segmentation of Lung Tumor using morphological Watershed Segmentation Algorithm comprising the steps of:
a) A method is designed to remove impulse or speckle noise from the image.
b) A method is designed to identify boundaries and framework of the image.
c) A method is designed to detect the tumor regions from the image.
2. The System/Method for Image segmentation of Lung Tumor using morphological Watershed Segmentation Algorithm as claimed in claim1, led to the design of a Median filtering process to remove impulse or speckle noise from the images.
3. The System/Method for Image segmentation of Lung Tumor using morphological Watershed Segmentation Algorithm as claimed in claim1, morphological operations are used to identify boundaries and framework of the images.
4. The System/Method for Image segmentation of Lung Tumor using morphological Watershed Segmentation Algorithm as claimed in claim1, watershed segmentation algorithm is designed to find the minima and maxima tumor regions from the images.