automatic segmentation

The MOCCA tagger was based on a classification ap- proach, which introduces a post-processing step after the actual alignment. The general setup of the experiments was as follows: test data consisting of the speech signal and the corresponding transcript were processed by the S&L sys- tem Munich AUtomatic Segmentation System (MAUS) de- scribed in Schiel (1999). Based on features derived from the MAUS decoding process, MOCCA tagged each word of the input transcript as to whether it matched the speech signal or not (experiment 1) and at the same time estimated the degree of overlap (OvR) between the calculated seg- mentation and the ground truth segmentation (experiment 2). The estimation of whether a word label is correct is a two-class classification problem, while the prediction of the OvR is a regression task; for both tasks, classification and prediction, we tested a SVM, which was reported to give good results in Zhang and Rudnicky (2001), and a RF. 2.2. S&L System MAUS

Abstract—Colorectal cancer is the third most common type of cancer worldwide. However, this disease can be prevented by detection and removal of precursor adenomatous polyps during optical colonoscopy (OC). During OC, the endoscopist looks for colon polyps. While hyperplastic polyps are benign lesions, adenomatous polyps are likely to become cancerous. Hence it is common practice to remove all identified polyps and send them to subsequent histological analysis. But removal of hyperplastic polyps poses unnecessary risk to patients and incurs unnecessary costs for histological analysis. In this paper, we develop the first part of a novel optical biopsy application based on narrow- band imaging (NBI). A barrier to an automatic system is that polyp classification algorithms require manual segmentations of the polyps, so we automatically segment polyps in colonoscopic NBI data. We propose an algorithm, Shape-UCM, which is an extension of the gPb-OWT-UCM algorithm, a state of the art algorithm for boundary detection and segmentation. Shape-UCM solves the intrinsic scale selection problem of gPb-OWT-UCM by including prior knowledge about the shape of the polyps. Shape-UCM outperforms previous methods with a specificity of 92%, a sensitivity of 71% and an accuracy of 88% for automatic segmentation of a test set of 87 images.

The goal of this paper is to create an extended rule corpus with approximately 2300 phonetic rules which model segmental variation of regional variants of German. The phonetic rules express at a broad-phonetic level phenomena of phonetic reduction in German that occurs within words and across word boundaries. In order to get an improvement in automatic segmentation of regional speech variants, these rules are clustered and implemented depending on regional specification in the Munich Automatic Segmentation System.

In this paper, a novel method for automatic segmentation of SSP tendon from ultra- sound image is proposed. The method involves image enhancement and feature extrac- tion from ultrasound image. The image was contrast enhanced using statistically adaptive method followed by speckle removal using anisotropic diffusion method. The image was then decomposed using curvelet transform. The energy analysis of decom- position was performed to select the amount of curvelet features needed for mask gen- eration. It was found, that 6.5% of curvelet features, at scale 2 and 16 orientations, provides best mask for segmentation. Images were reconstructed using extracted curve- let features and geodesic morphological operations were used to extract edges and re- move outliers. Connected component analysis and area filtering were applied to remove the remaining false areas and perform accurate detection. There is a trade-off between selecting curvelet features and removal of false areas. High percentage of cur- velet features results in increase of false positives. The polynomial curve fitting is used to smooth the area of SSP tendon as per radiologist’s recommendations. The seg- mented SSP tendon will assist the radiologist for focused and accurate diagnosis of ab- normalities in the tendon. The quantitative assessment performed for segmentation and results of diagnosis for pathological conditions suggests the effectiveness of pro- posed algorithm. Also the computation time for algorithm shows the capability of the algorithm to be made available for real time diagnosis of pathologies in SSP tendon.

The classification process, as described in [21], is used exclusively in the sagittal images to label the closed con- tours as either intervertebral disks or background. In short, the classification step allows us to eliminate back- ground regions that are falsely detected as intervertebral disks in the automatic segmentation step. The super- vised k-Nearest Neighbours (k-NN) classifier is used with four statistical and four spectral texture features to label each region as either intervertebral disk or back- ground in the sagittal segmented images. The statistical texture features are based on histogram of the closed contour (mean, standard deviation, skewness and entropy). All four spectral texture features are based on the energy Fourier spectrum of the closed region. By using the Fourier spectrum, we have information about the orientation and the frequency of intensity variation of the closed region. To facilitate interpretation, the spectrum is expressed in polar coordinates (r,Θ). Hence, the 4 descriptors of this function used as spectral texture features are the angle θmax at which the spectrum is maximal, the value S(θ)max of the spectrum at θmax, the variance of S(θ) and the difference between S(θ)max and S(θ)mean [21].

In this paper, a new technique is presented which combines thresholding and probabilistic Relaxation Labelling Process (RLP) for automatic segmentation.. of EM images.[r]

However, the automatic segmentation models in prior work were developed for predicting top- level topic segments. In addition, compared to read speech and two-party dialogue, multi-party dialogues typically exhibit a considerably higher word error rate (WER) (Morgan et al., 2003). We expect that incorrectly recognized words will impair the robustness of lexical cohesion-based approaches and extraction of conversation-based discourse cues and other features. Past research on broadcast news story segmentation using ASR transcription has shown performance degradation from 5% to 38% using different evaluation metrics (van Mulbregt et al., 1999; Shriberg et al., 2000; Blei and Moreno, 2001). However, no prior study has reported directly on the extent of this degra- dation on the performance of a more subtle topic segmentation task and in spontaneous multiparty dialogue. In this paper, we extend prior work by

Background: T2-weighted cardiovascular magnetic resonance (CMR) has been shown to be a promising technique for determination of ischemic myocardium, referred to as myocardium at risk (MaR), after an acute coronary event. Quantification of MaR in T2-weighted CMR has been proposed to be performed by manual delineation or the threshold methods of two standard deviations from remote (2SD), full width half maximum intensity (FWHM) or Otsu. However, manual delineation is subjective and threshold methods have inherent limitations related to threshold definition and lack of a priori information about cardiac anatomy and physiology. Therefore, the aim of this study was to develop an automatic segmentation algorithm for quantification of MaR using anatomical a priori information.

Fuzzy entropy based optimization of clusters for the segmentation of lungs in CT scanned images proposed by M. Arfan Jaffar , Ayyaz Hussain and Anwar Majid Mirza [11]. This method utilizes Fuzzy Entropy and Morphology segmentation methods. Fuzzy entropy is used for determining dynamic and adaptive optimal threshold. Histogram based background removal operator is proposed for removing the background areas. Xiangrong Zhou et.al [12] discussed automatic segmentation and recognition of anatomical lung structures from high-resolution chest CT images. This method recognize lung anatomical structures in chest by segmenting the different chest internal organ and tissue regions sequentially from high-resolution chest CT images. A sequential region-splitting process is used to segment lungs, airway of bronchus, lung lobes and fissures based on the anatomical structures and statistical intensity distributions in CT images. Jiantao Pu et.al [13] proposed technique for lung segmentation Adaptive border marching algorithm. It smoothes the lung border in a geometric way and can be used to reliably include juxtapleural nodules while minimizing over segmentation of adjacent regions such as the abdomen and mediastinum.

The observation of chromosomes has been crucial for our understanding of their structure, function, organization, and evolution of genes and genomes [1] as well as morphological changes during mitotic and meiotic divisions [2]. In this work, we present an automatic algorithm for the segmentation of centromeres and foci of DNA processing proteins, as well as the delineation of convoluted chromosomes. The algorithm is fully automatic and does not require tuning of parameters. Statistical measurements of numbers, areas distance and lengths are provided by the algorithm. The work is preliminary as this algorithm has not been tested on a large database nor used to differentiate between populations, however, it is considered that given it is fully automatic and fast it should be a useful tool for the analysis of chromosomes.

As the lung is actually a bags filled with air inside the body it is appeared as dark region in CT scans. Lung surrounding tissues and image intensities are contrasted. So the most of the segmentation structures are based on gray level thresholding which also includes histogram thresholding, iterative or automated 3D thresholding, multiple 2D thesholding. Affecting lung edges that takes place in diffuse parenchymal lung disease. We use these methods so that they reach to limit in the presence of pathologies. This occurs because image intensities alters in pathological regions where grey levels closer to the bone, fat or muscle. To beaten this thresholding methods associate with other techniques which are based on mathematical morphology or rolling ball operation, region growing and anatomical knowledge. Statistical approach of using 3D active shape models was chosen by Li and Reinhardth in order to produce relative segmentation. Stochastic categorization methods are advanced for lung segmentation and are based on texture examination or Markor-Gibbs model.

This paper presents an accurate heart sound segmentation algorithm that combines time-domain, frequency-domain and time–frequency-domain analysis. Compared to existing studies, this method is applicable to a wide range of heart sounds, from nor- mal to those containing S3, S4 and various murmurs. To verify this method, quantita- tive experiments were performed using the University of Michigan’s Heart Sound & Murmur Library, an authoritative open database. The experimental materials incor- porated two types of normal heart sounds and 14 types of abnormal heart sounds. The results show that the boundary localization has an average Se of 100%, an average PPV of 99.3% and an average Acc of 99.93%. Moreover, the Se, PPV and Acc of the component identification reach 98.63%, 99.86% and 98.49%, respectively, indicating outstanding performance of the proposed method. There are still some shortcomings of this work. For example, the component identification relies on the success of car- diac cycle calculation; therefore, this method cannot be applied to the heart sound with severe arrhythmia because of the failure to achieving accurate cardiac cycle by using UACF. The study of segmentation provides a good basis for extracting signifi- cant features of heart sounds. Therefore, the further study will focus on the classifica- tion of heart sounds.

In this paper we have a tendency to have a propensity to face stay playing the lowest fact and fuzzy segmentation. Through exercising the ones techniques that the membrane vessel picture its miles cut up proper into a try of factors i.e. thick and thin vessels. The enter images want to be forced to have loads of clarity, sharpness, evaluation, and reputation for the right segmentation. The various normalized pictures the thick vessels location unit detected with the beneficial useful resource of version community thresholding.

In the data preparation test, two different factors are investigated, the slice arrangement and the image contrast. The slice arrangement defines the number of slices that are used for training the network. Instead of using a single slice, the stacked slice is introduced to provide 3D context from volume image. Using the dice metric, the network that trained using stacked slice achieve a higher score than the network with a single slice. However, there is a maximum number of slices that can be added. After passing this limit, the dice score stops showing an increasing trend and start decreasing gradually. The second factor that needs to be investigated is the influence of image contrast on the segmentation result. An opposite result has been observed by comparing two different enhancement techniques that based on pixel intensity and histogram data. The method that works based on the pixel intensity information shows a slight improvement to the dice score, while method based on image histogram lead to a sharp drop in the dice score value.

To achieve this goal the image segmented to many parts after converting color image to binary image based on Y component from the YCbCr color space.. Color factors determined for[r]

In clinical practice, radiologists make a comprehen- sive diagnosis of glioma on the basis of the characteris- tics of various MRIs. Commonly used sequences generally include at least four types: T1-weighted im- aging (T1), T2-weighted imaging (T2), fluid-attenuated inversion recovery (FLAIR) imaging, and contrast-en- hanced T1-weighted (CET1) imaging [4, 5]. CET1 can reflect the blood flow information of a lesion, T1 pro- vides anatomical information, FLAIR imaging can help distinguish the cerebrospinal fluid of the edema area, and T2 is sensitive to the edema area and can provide such information as tumor boundary and edema degree [6]. In these sequences, T2 images can considerably re- flect the morphological information of tumors and are often used in clinical segmentation of gliomas. More- over, using the segmentation result of the edema area as ROI of each sequence can provide information on all types of regions of tumors because such an area often contains the real and necrotic areas of tumors. There- fore, the design of an automatic segmentation algo- rithm for T2 sequence has superior clinical value.

ABSTRACT: Extraction of image-based biomarkers, such as the presence, visibility or thickness of a certain layer, from 3D optical coherence tomography data provides relevant clinical information. We present a method to simultaneously determine the number of visible layers in the outer retina and segment them. The method is based on a model selection approach with special attention given to the balance between the quality of a fit and model complexity. This will ensure that a more complex model is selected only if this is sufficiently supported by the data. The performance of the method was evaluated on healthy and retinitis pigmentosa (RP) affected eyes. Additionally, the reproducibility of automatic method and manual annotations was evaluated on healthy eyes. A good agreement between the segmentation performed manually by a medical doctor and results obtained from the automatic segmentation was found. The mean unsigned deviation for all outer retinal layers in healthy and RP affected eyes varied between 2.6 and 4.9 µm. The reproducibility of the automatic method was similar to the reproducibility of the manual segmentation. Overall, the method provides a flexible and accurate solution for determining the visibility and location of outer retinal layers and could be used as an aid for the disease diagnosis and monitoring.

Normally brain tumor image contain large amount of information’s so manual segmentation is time consuming and complex process. In order to overcome these complication, automatic segmentations methods were introduced to detect the of brain tumor such as region growing [2], thresholding [3], artificial neural network [4] and clustering [5], etc. But still segmentation of brain tumor is still a challenging problem in image processing and analysis. The structure of the brain is complicated so it is difficult to determine the accurate segmentation of necrosis, edema and enhanced tumor. Several tissues present in the brain consist of three normal tissue [6] region, namely, gray matter (GM), white matter (WM), and Cerebrospinal Fluid (CSF), which is significant to analysis and treatment for diseases such as multiple sclerosis, Alzheimer's disease and epilepsy. These three regions are identified by the segmentation of brain image by utilizing the gray level distribution of pixels. The main goal of brain tumor segmentation [7] is to identifies the extensive and location of the tumor region [8], such as edema, active tumorous tissue and necrotic tissue. Mean-shift algorithm [9] were used to detect the brain tumor in MRI image. The most widely used automatic segmentation technique in bioinformatics application [10] is clustering. Now a days, clustering based image segmentation on pixels are used in imaging technique, which organizing a given database into a group. Significant role of clustering in MRI image is generally used to detect the brain diseases and abnormalities, to monitor, diagnose and treat disease. Several clustering technique are used in the existing work to detect the abnormalities such as fuzzy k-mean clustering [11], , adaptive fuzzy k-mean clustering [12], modified k- mean clustering [13] and fuzzy C-mean [14] clustering. The goal of these clustering is to detect the abnormalities based algorithm to minimize the objective function based on certain criteria. In this paper, we introduce a Novel center ______________________________

As a human, we can finish the tasks of object recognition and segmentation very fast, while developing a computer system that can automatically and in real time detect and segment an object is something that computer vision scientists have been working on for decades. Many researchers tried various segmentation techniques to make computers mimic human vision pro- cessing. However, we have to admit that there is still a long way to go before computer perfor- mance can compare with a human, even on a relatively simple task of foreground segmentation. In this thesis, we focus on “binary” automatic segmentation of an input image. Intuitively, given an input image, the task is to separate it into two regions, one corresponding to the fore- ground, and the other one to the background (see figure 1.1), based on the feature di ff erences in the two parts.

The objective of this research is to make reliable estimation of pleural effusion volume in CT imaging using digital image processing algorithms. In order to make reliable estimation we need to do the manual and automatic segmentation of CT images and to perform the comparison of automatic and manual segmentation for the quantification of pleural effusion on CT images which provides help in the diagnosis of the pleural disease. Pleural effusion is the collection of excess fluid in the pleural cavity. Excessive amount of fluid can impair breathing by limiting the expansion of lungs. Heart failure, cancer, cirrhosis, pneumonia, tuberculosis and many other are the causes of pleural effusion. A number of noninvasive imaging techniques such as radiography, ultrasound and computed tomography (CT) can detect the pleural effusion. The problem faced is the quantification of pleural effusion volume for the purpose of diagnosis of the pleural disease. The objective of this research is to make reliable estimation of pleural effusion volume in CT imaging using digital image processing algorithm. In order to make reliable estimation we need to do the manual and automatic segmentation of CT images and to perform the comparison of automatic and manual segmentation for the quantification of pleural effusion on CT images which provides help in diagnosis of the pleural disease. The results obtained by both the aforementioned techniques indicate that the manual segmentation is better because automated technique has less number of pixels.