Algorithms for Breast Cancer Decision Phase - Coursework Example

THE MOST PERFORMING ALGORITHMS FOR BREAST CANCER DECISION PHASE THE MOST PERFORMING ALGORITHMS FOR BREAST CANCER DECISION PHASE Introduction The detection and diagnosis of breast cancer in its very early stage boosts the chance for total recovery and successful treatment of the patient. The best presently available radiological technique for the early detection of breast cancer is screening mammography. Ongoing improvements and advances in imaging technologies have enhanced the sensitivity of breast cancer diagnosis and detection; however every modality is mainly beneficial when utilized according to traits of the individual such as breast density age, and risk (2). Mammography is considered to be the gold standard in the assessment of the breast lesions from an imaging viewpoint. Magnetic resonance imaging and Ultrasound examination are being offered as adjuncts to the pre and postoperative workup and as diagnostic techniques. Regardless of all of these advances, no particular imaging modality is capable of characterizing and identifying all breast abnormalities. It is an x-ray test of the breasts in asymptomatic a woman. The diagnostic mammography assessment is performed on symptomatic women who have an irregularity found during screening mammography (5). Currently, in a good number of hospitals the screen film mammography is being substituted with digital mammography. Using digital mammography a special electronic x-ray detector that changes the image into a digital mammogram is used to capture the image of the breast for viewing or storing on a computer monitor. Every breast is imaged independently in mediolateral-oblique (MLO) view and craniocaudal (CC) view. Provisions have been developed to explain lesion classification, breast density and lesion features. Screening mammography facilitates detection of early signs of breast cancer such as bilateral asymmetry, masses, architectural distortion and calcifications. A mass is the space that occupies lesion and is seen in at least two diverse projections Masses have diverse density, different margins and shape. Oval and round shaped masses with margins that are smooth and bounded usually signify benign changes. A malignant mass on the other hand, generally has a speculated, blurry and rough boundary. Nonetheless, there is a typical case of speculated or macrolobulated benign masses, as well as well-circumscribed or microlobulated malignant masses. Deposits of calcium in breast tissue are known as Calcifications. Calcifications that are detected on a mammogram are a significant indicator of malignant breast disease however; they are also present in several benign changes. The Benign calcifications are generally coarser and larger with smooth and round contours (1). Architectural distortion is known as deformation of the normal architecture with no specific mass visible, as well as spiculations radiating from a position and focal distortion or retraction at the edge of the parenchyma. Architectural distortion of breast tissue might indicate malignant changes particularly when integrated with noticeable lesions such as calcifications, mass or asymmetry. Architectural distortion may be classified as benign once there is a scar and damage of a soft-tissue due to trauma. The breast parenchyma asymmetry between the two sides is a helpful sign for detecting early breast cancer. Bilateral asymmetries of alarm are the ones that are new, changing or enlarging, also the ones that are blatant and the ones linked with other findings, such as architectural distortion or Microcalcifications. If a clear mass or thickening corresponds to an asymmetric density, this density is observed with a bigger degree of suspicion for malignancy. Breast lesions have various features that are indicative of malignant transformations, but can also be a fraction of benign alterations (4). They are now and then impossible to differentiate from the surrounding tissue this makes the diagnosis and detection of breast cancer more complicated. Therefore, it is crucial to come up with a system that can aid in the decision between biopsy and follow-up. Using computers to process and analyze biomedical images allows a more precise diagnosis by a radiologist. Human beings are prone to committing errors and their analysis is more often than not qualitative and subjective. Application of computers to biomedical image analysis facilitates quantitative and objective analysis and leads to a more precise diagnostic assessment by the physician. Computer-aided diagnosis (CADx) and computer-aided systems detection (CADe) may improve the outcome of mammography screening programs and reduce number of fake positive cases (3). The majority of image processing algorithms consist of a few distinctive steps. The screen film mammographic images have to be digitized before processing the image. This is one of the advancements of digital mammography because an image can be processed directly. Preprocessing step is the first step in image processing and to reduce the noise it must be done on digitized images and it also improves the image quality (10). Most of the digital mammographic images are very high quality images. In the mammographic image analysis case, the results that are produced by a definite method may be presented in a few ways. The interpretation maybe mostly used is the (1) confusion matrix or just by the number of (TPs) true positives and (FPs) false positives. The confusion matrix consists of (TN) true negative, (FP) false positive, (FN) false negative and (TP) true positive. C = FP TN FN TP Algorithms that are presented for the detection of the two mainly common signs of breast cancer are masses and calcifications. The Algorithms for bilateral asymmetry detection and architectural distortion detection are regularly part of the mass detection algorithms (7). Mass Detection Algorithms A few researchers have mainly focused on the detection of spiculated masses due to their high probability of malignancy. The algorithms for detection of breast mass in digital mammography generally consist of a number of steps: one is segmentation; the second one is feature extraction, then feature selection and lastly classification. The intention of the segmentation is to extract Regions of Interests containing all the masses and to locate the suspicious mass candidates from the Regions of Interest. Segmentation of the regions that are suspicious on a mammographic image is intended to have an extremely high sensitivity and a great number of false positives are tolerable since they are expected to be removed in later on stage of the algorithm (12). With the speedy progress of computer and physic technology, detection of calcifications and masses on mammograms has been a burning field of research in early breast cancer detection. Masses are the main signs of breast cancer on mammograms. For this reason, there is a series of innovative methods to automatically detect masses. An innovative image enhancement method is proposed based on the morphological analysis; it can efficiently suppress the background and improve the features of masses on mammograms at the same time. After that the plump seed regions may be removed from the images using various features such as contrast or gray in the original and enhanced images. A new method stimulated by fuzzy set, is established to extend the blurry region growing to a fuzzy version, which guarantees the totality as well as stability of the growing region. A hierarchical detecting method is developed and it can detect the not easily seen masses successfully. Due to a great deal of false positives in regions of interest, the SVM classifier is designed to differentiate masses from areas that are normal with excellent detected result. To increase the true positives while reducing false positives, the significance feedback is introduced to sort the amount of false positives (8). Microcalcification Detection Algorithms Calcium deposits inside the breast are referred to as calcifications. They can be generally divided in two main groups i.e. Microcalcification and Macrocalcifications. Macrocalcifications are mainly large calcium deposits, while on the other hand Microcalcifications are small calcium deposits. Macrocalcifications are generally not linked with the growth of breast cancer and that is the main reason why no exceptional attention is devoted to them. However, detection of Microcalcifications is very vital for the detection of early breast cancer. Microcalcifications are frequently linked with additional cell activity in the breast tissue. This extra cell activity doesn’t have to be cancerous and it generally isn’t, but if the Microcalcifications are grouped in clusters, it can be an indication of a malignant tumor that is developing. Spread Microcalcifications are typically a part of a breast tissue that is benign. In mammograms calcifications they are viewed as different sized bright dots. The accurate position and number of Microcalcifications cannot be predicted (12). Even though Microcalcifications may be grouped in clusters, more often than not they are found to stand alone. Microcalcifications detection is a very difficult task for computer-aided detection software as well as for radiologists. High-quality spatial resolution is very significant for Microcalcification detection since their actual size may be as small as 100 μm. Digital mammography is mostly used in the present day since it displays with high resolution and is essential for delivering images that are sharper and richer with facts to radiologists. Computer-aided detection software (CADe ) makes the diagnosis process easier and almost automatic. In application of mammography, one of the main important responsibilities of CADe is to identify the presence of Microcalcifications, particularly clustered ones, since they can be a likely sign of early cancer. Since Microcalcifications are tiny and scattered randomly in breast tissue it is quite possible for a radiologist to fail to notice them (1). Due to this reason CADe software must give good results by producing a smaller amount of false negative results (FN). This also lays a major problem for the CADe software, because radiologists can probably fail to notice some microcalcifications due to trusting the accuracy of the software detection too much, which can again give FN results. The general microcalcification detection process involves: , After image enhancement, (ROI) region of interest ought to be detected. The next two steps are Feature extraction and selection. Lastly, the decision algorithm provides detection based on selected features. Another element of the preprocessing step is the removal of the background area and if an image is MLO views the removal of the pectoral muscle from the breast area. The segmentation step intends to uncover suspicious regions of interest (ROIs) that contain abnormalities. In the step of feature extraction features are calculated from the region of interest’s characteristics. The feature selection step is the significant issue in algorithm design it is where the finest set of features are selected for the elimination of false positives and classification of lesion types. Feature selection is selecting a tiny feature subset which leads to the biggest value of some classifier performance function. Lastly, on the base of selected features the false positive lesion and reduction classification are carried out in the classification step (9). Wavelet based methods are being regularly used in the microcalcification detection process. The proposed methods give acceptable results and work more or less in a similar way. Microcalcifications are very tiny objects and to be able to detect them it is crucial to extract high frequency components. Wavelet transform provides the spatial information of the detected object and this is the major reason why it is consequently successful in this area. Moreover some other methods have been proposed for microcalcification detection but their number is considerably smaller. Characterization of Microcalcifications In addition to detecting Microcalcifications, another demanding task is automatic characterization of Microcalcifications. Characterization gives the answer on whether microcalcifications are malignant or benign. Characterization of microcalcification clusters in mammograms is very important in daily clinical practice. Currently, characterization of microcalcification clusters mostly depends upon the knowledge of experts; this increases the workload and complexity (6). This process can be divided into three phases: a) The detection stage where clusters of microcalcifications are identified b) The feature extraction stage in this stage the important features of each cluster are computed. c) The classification stage it provides final characterization. For characterization purposes, many categories have been used. Some frequently used characterization methods are: K-nearest neighbor classifiers, neural networks, supports vector machine, different decision trees and Bayesian characterization. The categories for solo microcalcification evaluate these features: local contrast, compactness, border gradient strength and roughness. The characterization of the clusters of microcalcifications evaluates these features: centre mass of the cluster, mass density of the cluster, and the average mass of the microcalcifications, the standard deviation of distance between microcalcifications and standard deviation of the masses of the microcalcifications and center of mass cluster. Some characteristic microcalcification shapes of different possibility to be malignant and form are shown. Techniques for decision making and diagnosis In the last decade, Breast imaging has been able to make great advances, and together with current techniques of primary diagnosis of breast cancer, several new methods are in use and they show potential in detecting isolated metastasis, recurring disease and evaluating the reaction to treatment. The Full-field digital mammography makes use of the lesion-background contrast and provides better compassion, and by altering computer windows it is likely to see through all the dense tissues; this can be mostly helpful in young women with breasts that are dense (13). Through this method the requirement for repeat imaging is decreased, with the additional advantage of reduction of the radiation dose given to the patients. The computer-aided detection systems on the other hand can assist the radiologist in interpreting both digital and conventional mammograms. An MRI also has a great role in the screening of women who are at who are at a higher risk of developing breast cancer. It aids in management of cancer by assessing the reaction to treatment and may also help in making a decision on the suitable surgery by providing precise facts on the size of the tumor. Modern techniques for diagnosis such as optical imaging, molecular diagnostic and sestamibi scan techniques look quit promising; however more investigation on their use is needed. Their functions will become clearer in future years, and they might possibly turn out to be helpful in additional investigating lesions which are undetermined on normal imaging. Additional forthcoming techniques are tomosynthesis and contrast enhanced mammography. These may provide supplementary information in undefined lesions, and as shown in recent studies they assist in reducing the recall rates when used in screening. Computed tomography/PET has a part in discovering local distant metastasis and disease recurrence in patients with breast cancer. (SVM) Support vector machine is a type of machine learning algorithm. It is derived directly from the theory of statistical learning. SVM is based on the standard minimization of risk that is conducted by the generalization error minimization (4). Generalization error is brought about by the learning machine on a test data set that is dissimilar to the training data set and it has no overlapping. For SVM to properly function the following should be done; avoid over fitting, the decision boundary must not match up to the training data set. To avoid the likely over fitting some exceptional user defined parameters are presented. An additional approach in microcalcification characterization uses the content-based image retrieval technique. This proposed method involves of two steps: 1) Using learning based similarity measure to retrieve similar mammogram images from a database 2) Characterizing the query mammogram image based on the retrieved results (retrieval-driven characterization). Conclusion Breast cancer is one of the major causes of death among women. Digital mammography screening programs can facilitate early detection and diagnose of the breast cancer this reduces the mortality and raises the chances of full recovery. Screening programs create a great amount of mammographic images which radiologists have to interpret. Because of the wide range of abnormalities features in the breast some abnormalities may be misinterpreted or missed. There is also a huge number of false positive findings and for that reason a lot of unnecessary biopsies. Diagnosis algorithms and Computer-aided detection have been developed to assist radiologists in giving an accurate diagnosis and to decrease the number of false positives (2). There are lots of algorithms that have been developed for the detection of calcifications and masses. Over time there has been advancement in the detection algorithms however their performance is still not perfect. There are still lots of false positive outputs and the likely reason for such a performance may possibly be due to the characteristics of breast abnormalities. Calcifications and Masses are occasionally hidden and superimposed in the dense tissue this makes the segmentation of correct regions of interest difficult. Another matter is extracting and selecting suitable features that will give the most excellent classification results. Whats more, the choice of a classifier greatly influences the final result and classifying abnormalities as malignant or benign is a complicated task even for expert radiologists. Additional developments in each algorithm step are necessary to improve the general performance of diagnosis algorithms and computer aided detection. References: (1). B. Acha and R.M. Rangayyan: Recent Advances in Computer-Aided Diagnosis, Breast Imaging and Mammography, of Breast Cancer; Detection of Microcalcifications in Mammograms. SPIE, Bellingham (2006) (2). R.M Rangayyan and F.J Ayres: The Review of Computer-Aided Breast Cancer Diagnosis: the Franklin Institute Journal Vol 344(3-4), pp312–348 (2007) (3). M.P Sampat and A.C Bovik.: Computer-Aided Diagnosis and Detection in Mammography. Elsevier Academic Press, Amsterdam (2005) (4).E.S.Paredes,: The Atlas of Mammography, 3rd edition. Lippincott Wilkins & Williams, Philadelphia (2007) (5).H.D. Cheng and H.N. Du: Pattern Recognition of Detection and Classification of Masses in Mammograms. Vol 39(4), pp. 646–668 (2006) (6).R.M. Rangayyan,: Biomedical Image Analysis. CRC Press LLC, Boca Raton (2005) (7).A.K. Jain and J. Mao: Transactions on Pattern Analysis and Machine Intelligence; Statistical Pattern Recognition Vol 22(1), pp. 4–37 (2000) (8).B. Sahiner and J.Shi: Descriptors for Mammographic Mass Margin: Computer-Aided Diagnosis, vol. 6915 (2008) (9).N. Tóth and Pataki, B.: Detecting of Masses in the Mammographic Images. Transactions on Measurement and Instrumentation Vol 55(3), pp. 944–951 (2006) (10).B. Zheng and D. Gur: Improvement of resemblance of parallel Breast Masses chosen by Computer-Aided Diagnosis Schemes. Vol. 12-15, pp. 516–519 Pappas, T.N (2007) (11).S. Timp and N. Karssemeijer: Medical Image Analysis: Change Analysis to develop Computer Aided Detections in Mammography. Vol. 10, pp. 82–95 (2006) (12).S. Timp, and C.Varela: sequential Analysis of Change for Mass Lesions Characterization in Mammography. Vol. 26(7), pp. 945–953 (2007) (13).L. Hadjiiski and M.A. Helvie: Computerized Classification and Detection of Benign and Malignant Microcalcifications on Full Field Digital Mammograms. vol. 5116, pp. 336–342. Springer, Heidelberg (2008) (14).K. Geetha and A. Kumar: Feature Classification and Selection of Microcalcifications in Mammograms. ICSCN Vol. 4-6, pp. 458–463 (2008) Read More