Use this URL to cite or link to this record in EThOS:
Title: Automatic breast cancer classification using novel feature extraction for magnetic resonance imaging and image processing technique
Author: Alshanbari, Hanan Saad J.
Awarding Body: Coventry University
Current Institution: Coventry University
Date of Award: 2013
Availability of Full Text:
Access from EThOS:
Access from Institution:
Determining the appropriate methodologies for the early detection of breast cancer is still an open research problem in the scientific community. Breast cancer continues to be a significant problem in the contemporary world. After 40 years of age, people tend to be more able to developing cancer, and nearly 25% of cancers are detected in women between the age of 40-49 years (Health Quality Ontario, 2007). Early detection and treatment are currently the only proven means to reduce breast cancer's related mortality rates. Computer-aided detection is suggested as an adjunct to screening mammogram to decrease perception-based errors. Medical image processing tools were demonstrated to be effective methods for helping radiologists identify suspicious tissues in different medical imaging modalities such as Mammograms, Magnetic Resonance Imaging (MRI), and ultrasound. Since there are several types of abnormalities in the breast, they require special focus for detection; however, even trained specialists are frequently unable to detect them. Moreover, medical experts might make mistakes, as they are only human; they may be over-worked or may make common errors, which can result in even bigger issues (or translate into death) for the patient. Hence, to lessen the burden on those physicians who face these problems, as well as higher workloads, it is imperative to facilitate the diagnostic process and to also train sufficient numbers of residents to interpret mammograms, (MRI), or other imaging modalities in the future. MRI-based imaging provides far superior clarity and resolution when compared to other imaging modalities. The clear and precise information offered by MR imaging serves as the basis for correctly detecting cancers, while also identifying their specific type. Various imaging modalities (including MRI) provide outputs that do not give clear information or that do not clarify hidden information associated with breast cancer; in fact, it is often not possible for people to detect these unclear outputs. A specialist may find it difficult to correctly predict the cancer type, which could lead to the wrong diagnosis. This prediction can be improved by computer-aided technologies, which also minimizes human intervention. In this study, we tried provide an automatic detection breast cancer detection system using samples of MRI breast images. One of the fundamental issues in the design of a detection system is to determine the efficient features that should be used together to improve the accuracy of said system. Depending on the type of feature (bad to best) that is used in system design, the detection scheme can provide classification accuracy results (0 to 100%). The use of a good classifier is also necessary to deal with non-linear classification. This research work proposes pre-processing methods for MR images, as well as novel methods to detect cancer in those images. This work also proposed new methods to retrieve discriminative features from suspicious MR images, and also utilizes the neural network classifiers on them to create an automatic decision making system. An extensive test is conducted on the classifier to assess its ability to provide false-positive and false-negative readings, and also evaluated its accuracy rates. The proposed research starts with a detailed study to understand the suspicious patterns observed in MR images. In order to overcome some of the bottlenecks in the existing methods, this study tries to improve the suspicious MRI pattern detection by devising novel techniques (novel features). In this investigation, this study further offers an exploration of the current theoretical approaches to segmentation, and then aim to assess the impact of a watershed transform algorithm on magnetic resonance (MR) image quality in the early detection of breast cancer to confirm the efficiency of this auto segmentation method. Moreover, five features are tested to classify the tumours. Further, ANN-based classifiers are used on these features to improve detection and to generate correct cancer classifications. This study further incorporates the support vector machine (SVM) classifier and test correct classification ability of proposed system. Different kernels of SVM are tested to find out the best results. SVM outperforms the ANN classifier in terms of accuracy by 98.52%. This study successfully classifies the type of a tumour with high accuracy using MRI of breast images. MRI data for this study is collected from King Abdullah medical city (KAMC) (
Supervisor: Amin, Saad Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available