SEGMENTATION AND INTERPRETATION OF ULTRASOUND MEDICAL IMAGES

Abstract

Diagnostic imaging is recognized as an important adjunct to clinical examination in the case of patients with many common illnesses and provides effective and non invasive mapping of the anatomy and physiology of human body. Different body tissue pathological conditions produce different image patterns, and the patterns exhibited by the biological tissues through their images have been in routine clinical use for medical diagnosis purposes. The role and contribution of radiology to medical diagnosis has expanded tremendously due to advances in images quality compliance regulations, image detector systems, and computer technology. The recent development and proliferation of medical imaging technologies has revolutionized the medical field and the role of medical imaging has been expanded beyond the simple visualization and inspection of anatomical structure. The different imaging technologies provide exceptional views of internal anatomy and the trained radiologists quantify and analyze the embedded structures. However, the manual analysis of images is a time consuming process besides being susceptible to human errors depending upon one's experience and expertise. To gain scientific knowledge of diseases and their effect on anatomical structures in-vivo for diagnostic and treatment planning, the medical image analysis community has become preoccupied with the challenging problem of extracting clinically useful information of anatomic structures from images obtained through CT, MR, PET, ultrasound(US) and other modalities. The choice of the best imaging technique to help to solve any particular clinical problem is based on the factors such as resolution, contrast mechanism, speed, convenience, acceptability and safety. The cost effectiveness and portability of the modality is particularly important in countries like India to allow widespread access to sophisticated medical imaging in a timely manner. Due to its ability to visualize human tissue without deleterious effects, ultrasound B-scan imaging has become the most widely used method of imaging soft tissue such as the lungs, liver, rectum, prostate, uterus, and neonatal brain, spleen, kidney etc. However, ultrasound images are of a relatively poor quality and its analysis in general is complex due to its data composition. Images obtained from echo US imaging systems, have interference patterns called speckle. From a technical point of view, speckle is an undesirable

interference effect that covers fine details in the image, like lesions with faint grey value transitions, and small details. The situation is completely different from medical point of view. In the medical approach speckle is considered important diagnostic information and the clinicians considers the original image as the most reliable since that contain all speckle. The major steps in the diagnosis process of ultrasound modality include the characterization of the sensors, image formation, image restoration, quantification, feature extraction, and interpretation. The present research has been focused on the understanding of ultrasound medical images, in which main emphasis has been given to the ultrasound medical image enhancement, delineation of region of interest from its background, and the interpretation of the extracted region of interest. The entire work has been divided into four parts, viz. 1) image enhancement (speckle suppression) for image segmentation 2) image enhancement for image interpretation 3) automated image interpretation for liver disease classification and 4) automated image segmentation with the aim of foetal growth evaluations. Under the image enhancement for segmentation, the emphasis has been on speckle suppression. Speckle is a term used for the granular pattern that appears in Bscan images; and is the consequence of unresolved scatters, which are caused by tissueultrasound interaction, and not an image of the scatters. Speckle tends to mask the presence of low contrast lesions and reduces the ability of a radiologist to resolve the actual information. Speckle occurs especially in images of body organs like liver and kidney whose underlying structures are too small to be resolved by large wavelength ultrasound. Unlike CT and MR images, ultrasonic images cannot generally be easily segmented and classified into separate anatomical or structural elements. The presence of speckle noise affects not only the human interpretation of the images but also the accuracy of computer-assisted diagnostic techniques. The conventional filters such as median filter, Wiener adaptive filter, Lee/Frost/Kuan/Gamma filter do not enhance the edge features, rather they inhibit the smoothing perpendicular to edges i.e., the high contrast speckle/noise is retained in the neighborhood of edge points. An examination of literature reveals a number of local statistical based adaptive filters such as adaptive weighted median filter (AWMF), adaptive speckle suppression

filter (ASSF), homogeneous region growing mean filter (HRGMF) been applied to suppress speckle. These statistical based adaptive filters are sensitive to the size and shape of the filtering kernel. An improved version of the adaptive filter based on the region growing and local statistical features known as aggressive region growing filtering (ARGF) is implemented here too. This technique is used to avoid premature blockage of the filtering region growth due to speckle. It has been observed that such methods are effective enough to preserve the edge details besides suppressing speckle. However, very little is spoken about the diagnostic features looked into by the medical fraternity, while analyzing medical ultrasound images. It has been observed that the detection of high contrast targets is limited by the imaging system's spatial resolution. Thus, the choice of the filtering algorithm or image enhancement technique is dependent not only on the system capabilities or the region to be imaged but also on the parameters looked into by the medical fraternity for diagnosis and the imaging context. A detail qualitative comparative analysis in association with a team of radiologist has been performed here. The results showed that the ARGF method is able to preserve the edges and their directions while smoothing the speckles. The superiority of this method over wavelet filter has also been shown with the help of image characteristics and segmentation of filtered images. ARG filtering, while preserving all the major edges, suppresses the spurious noise signals, form regions with similar gray levels, and thus facilitates better delineation of region of interest as compared to the other filtering algorithms. Enhancement of US images for interpretation reported in the present work aims at providing an image with a high content of visual details. There are classes of clinical US B-scan images where the speckle "texture" associated with microstructure of tissue parenchyma conveys information about the region being imaged and is considered as signal to be used for diagnosis. Image textures from different tissues may have differing appearances, which can assist in clinical image interpretation. Nonlinear filtering techniques are becoming increasingly important in image enhancement for interpretation, and are often better at removing artifacts and feature preservation in ultrasound medical images. However, design and analysis of nonlinear filters are much more difficult than for linear filters. One of the techniques for designing nonlinear filters is mathematical morphology, which creates filters based on shape and size

characteristics. Though the morphological filters are limited to minimum and maximum operations that introduce bias into images, however, the bias introduced by morphological filtering is not a problem; on the contrary, it is actually helpful for enhancing ultrasound images. An attempt has also been made to qualitatively and quantitatively evaluate the standard morphological operation for the image quality enhancement of B-mode ultrasound images. Comparative evaluation of the structuring elements for the morphological image enhancement showed that the image enhancement using line-shaped structuring element with length 2 has maximum improvement. This is in confirmation to our initial hypothesis that the speckles in US images are short, slight "banana-shaped" lines. In the present proposed computer-assisted diagnosis (CAD) system, computerized image analysis of region of interest (ROI) in the liver US image is used to suggest possible malignance in the regions so that the radiologist can examine these regions more carefully. In medical imaging, the accurate diagnosis and/or assessment of disease depend on both image acquisition and image interpretation. The image interpretation process, however, has only recently begun to benefit from computer technology. Radiologists perform most interpretations of medical images; however, image interpretation by humans is limited due to the search pattern of humans, the presence of structure noise in the image, and the presentation of complex disease states requiring the integration of the vast amount of image data and clinical information. The basic concept of CAD is to provide a computer output as a "second opinion" to assist radiologists' image readings. The basic technologies involved inCAD schemes are: (1) image processing for detection and extraction of abnormalities; (2) image features extraction from candidates of abnormalities; and (3) data processing for classification of image between normal and abnormal (or benign and malignant), using features. Adhering to the norms of medical ethics and obtaining an ethical clearance, a large number of patients' image data with different liver malignancies were obtained from a hospital to develop a CAD system for liver disease classification. Although the neural-network based approaches have been quite successful and largely reported for liver disease classification, the main disadvantage of artificial neural network (ANN) is that it requires a considerable amount of time and training efforts for good performance, especially under a wide variation in patient data and operating conditions.

To overcome these shortcomings new technique for differentiating benign and malignant livers, have been presented. The technique is based on the proposed new indices that are used to define the state of the liver. Different texture features based on the statistical properties of the intensity histogram, Fourier spectrum based spectral measures and gray level dependence matrix, which describe the physical properties of tissues, were extracted initially and then the same were used to define the indices. Though, in literature different methods have been proposed for classification of diffused liver diseases using texture, but it has been found that if the diffused liver diseases are to be classified based only on the texture of the liver surface then it is possible to clearly discriminate only between the normal liver and cirrhosis liver. Accordingly, the classification based on the proposed indices, has been confined to the discrimination between the normal liver and cirrhosis liver. At the end, one of the major applications of ultrasound imaging and image segmentation, has been discussed. The estimation of pregnancy dates is important for the mother, who wants to know when to expect the birth of her baby, and for her health care providers, so they may choose thejunctures at which to perform various screening tests and assessments. With the use of ultrasound imaging, the estimation of fetal maturity and the detection of fetal growth have been established. The important and most widely used parameters for the determination of the age being biparietal diameter (BPD), fronto-occipital- diameter (FOD), head circumference (HC), and femur length (FL). This work presents a scheme for interactive measurements of these parameters from the fetal head ultrasound image, which is mainly based on level-set deformable contours and which provide better results as compared to the methods used in the existing ultrasound machines. Complete automated segmentation method, using morphological operators for the measurement of fetal femur length and thus the gestational age of a fetal has also been proposed. In summary, the present work focuses on the problems encountered by the radiologists in interpretation and diagnosis of the liver diseases. Present work if integrated to the existing ultrasound machines, may help in rapid and more reliable screening method for initial liver disease classification. Similarly, if the proposed technique is applied for rapid estimation of gestational age of the fetus, the concerned doctor can take preventive and corrective measures.