Need For Breast Dce Mri Biology Essay

Published: Last Edited:

This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.

The high occurrence of breast cancer in women has increased considerably in the recent years and it is one of the main causes of death among women. The prevention of breast cancer seems impossible since the causes of this disease are still remaining unidentified. Early detection of breast cancer is one of the most significant factors in determining diagnosis for women with malignant tumors. An improvement of early diagnostic techniques is crucial for women's quality of life. Even with remarkable advances in modern imaging technology, both early detection and accurate analysis of breast cancer are still unresolved challenges. The efficiency and usefulness of medical imaging are constantly progressing, which leads to better patient care and more dependence on imaging techniques, which in turn leads to increase in image data that can be attributed to a number of factors including: increased image resolution, increased bits per volume image element (voxel), sampling in an increased number of dimensions and an increased amount of tissues imaged [1]. Breast DCE-MRI is a highly precise imaging tool in the detection and accurate characterization of breast cancer [2]. The ability to improve diagnostic information present in MR images can be enhanced by designing computer-assisted evaluation (CAE) systems with efficient and intelligent computer processing algorithms, which have the potential to assist radiologists in the early detection of cancer.

This thesis presents a novel computer-assisted evaluation (CAE) system developed using set of intelligent image analysis tools, for the purpose of assisting radiologists with the task of detecting and characterizing breast lesions in dynamic contrast enhanced magnetic resonance imaging (DCE-MRI).


The American cancer society recommends that MRI screening to be done annually in addition to mammography for women at high risk starting at age 30 [5]. MRI is increasingly used in the clinical setting in addition to x-ray mammography and ultrasound. Of these imaging modalities, MRI has many advantages such as high contrast between soft tissues, high spatial resolution and inherent 3D nature, thus has gained wide clinical applications. Breast MRI, has been widely investigated in the past decade in the detection and diagnosis of breast cancer as a complementary [3]. MRI is the most reliable method for assessing the tumor size and extent compared to the gold standard histopathology. It also shows great promise for the improved screening of younger women (with denser, more radio opaque breasts) and for women at high risk.

In order to be effective, breast MRI requires the use of a paramagnetic contrast agent. When used in combination with the contrast agent, the examination is called dynamic contrast enhanced - MRI (DCE-MRI). The contrast-agent has no contraindications and is tolerated better than the radiations absorbed during a single mammography [4].

Computer-Assisted Evaluation (CAE)

Developing computer-assisted evaluation and diagnosis schemes has been attracting rapidly growing interest in biomedical imaging research field during the last two decades. Aiming to assist radiologists and pathologists more accurately, who are consistently reading and interpreting biomedical images in the busy clinical environment. Computer-assisted evaluation (CAE) systems are continually being developed aiming to help the physicians in early detection of breast cancer. These tools may call the physician's attention to areas in the magnetic resonance imaging that may contain radiological findings.

The framework of the present work is to develop a computer-assisted evaluation (CAE) system comprising of efficient and innovative algorithms for identification and diagnosis of abnormal regions present in breast DCE-MR images. Such tools present to the physician both a qualitative and a quantitative description of the disease. The breast DCE-MR images used in this study was received from the Radiology Department of Kovai Medical Center and Hospital (KMCH). Coimbatore, Tamil Nadu, India. Images were acquired with patients prone to 1.5T scanner with use of a dedicated surface breast coil array. The imaging protocol included bilateral fat suppressed T1- weighted images in the sagittal plane of 1mm slice thickness and a slab interleaved 3D fat suppressed spoiled gradient echo after the injection of contrast.


The architecture of a CAE system proposed includes image pre-processing, segmentation of breast volume, edge enhancement, feature extraction, feature selection and the classification of suspicious tissue.


Pre-processing adapts the input image to a specific application. This phase could employ contrast enhancement techniques or methods for removing the noise. Pre-processing aims at improving the quality of each input image and reducing the computational burden for subsequent analysis steps. The application of pre-processing filter over the image aims at compensating the effects of random noise, while minimizing the loss of resolution.

Segmentation for Detection of ROI (Region of Interest)

The main goal of image segmentation plays a vital role in both qualitative and quantitative image analysis. In MRI, segmentation is the process of partitioning MR Images into regions, aiming to produce an image that is more meaningful and easier to analyse. Accurate segmentation of breast DCE-MR images is a key step in contouring during radiotherapy planning. The segmentation phase of CAE system presents two new intelligent methods for automatic segmentation and detection of suspicious lesions in DCE-MRI of the breast tissues. The methods are based on artificial neural networks and swarm intelligence.

The first method makes use of SOM (self-organising maps) based neural network approach for segmentation. In this method, classical K-means clustering is enhanced through the search of an optimized space in which to operate the clustering. It allows for the ability to make the clustering methods able to retain more information from the original image. This algorithm gives good results all around [6].

The second method applies swarm intelligence approach for segmentation; this method employs the artificial swarm bee colony algorithm to search for the set of cluster centers that minimizes a given clustering metric. This clustering algorithm converged to the maximum or minimum without becoming trapped at local optima. The ABC algorithm generally outperformed other techniques that were compared with it in terms of speed of optimization and accuracy of the results obtained. The results therefore confirms the usefulness of artificial bee colony clustering algorithm as an optimization tool and shows that it is very successful on optimization of medical image clustering [7,8].

The clustered output image obtained from the above discussed algorithms are then edge enhanced by unsharp filter [9], which is used to extract the edges of tumor very efficiently in MR images, followed by proper thresholding, the tumor or region of interest (ROI) is extracted from the edge enhanced image.

Feature Extraction and Selection

Once a lesion is detected in the ROI, characterization is necessary to estimate the pathological nature of the lesion, i.e., whether the lesion is benign or malignant. For the quantitative assessment, many features have been extracted from the masses of DCE-MR images extracted in segmentation phase of the computer-assisted evaluation system;

The implemented feature extraction procedure relies on the exploration of the textural characteristics of the extracted mass. The extraction methods of texture feature plays very important role in detecting abnormalities present in breast DCE-MRI. The statistical textures are found to be the best for image classification. The popular statistical feature extraction method is gray level co-occurrence matrix (GLCM). In this proposed work, a set of 18 features were extracted from the ROI, fourteen statistical measures for texture (GLCM) and four gray level histogram moments (GLHM) features. The features extracted are: Energy measure, correlation, inertia, entropy, difference moment, inverse difference moment, sum average, sum entropy, difference entropy, sum variance, difference variance, difference average, information measure of correlation, standard deviation, mean, variance, skewness and kurtosis.

Feature selection has been widely used to improve prediction accuracy of classifiers. Feature selection is defined as a series of actions to choose a subset of features that are relevant to correct classification based on specified evaluation and selection criteria.

The statistical hypothesis t-test is used to select the set of effective features for the classifier to obtain high accuracy. By providing the hypothesis t-test over all the extracted 18 features, the t-test results indicates that only 7 features can classify between the two clusters they are: Energy, entropy, mean, variance, skewness, standard deviation and kurtosis.

Mass Type Classification

It's difficult to achieve accurate findings and classifications of diseases, especially on cancers, which are very important in medical science. Accurate classification allows doctors to select suitable therapies and treatment for diseases.

The classification technique used in this research work is based on artificial neural network. It is trained using artificial bee colony (ABC) optimization algorithm. The basic idea behind the proposed classifier is to use ABC algorithm for searching the best combination of synaptic weights for neural network. The selected seven texture features were used to classify the mass with a three-layered neural network to predict the outcome of the biopsy. This work is the first experiment of artificial bee colony algorithm on classifying the ROI of breast DCE-MR imaging.

The performance of the proposed classifier structure is evaluated based on two validation methods: cross validation and leave-one -out cross validation, in terms of accuracy, specificity and sensitivity. The classification results showed an overall accuracy of 96.47%, sensitivity of 96.92%, specificity of 95% using leave-one -out cross validation and achieves an overall accuracy of 91.42 %, sensitivity of 92.30 %, specificity of 88.88 % using cross validation.

A quantitative measure of the accuracy of the classification technique is obtained by finding the area under the ROC curve (AUC) termed AZ. AZs were also calculated to evaluate the ability of the classifier using the cross validation and leave-one-out cross-validation. The experimental results suggest that suspicious lesions can be classified, with a high (AUC) AZ of 0.926 using cross-validation and (AUC) AZ of 0.967 using leave-one-out cross-validation, thus the overall results of both validation methods declare that the proposed classifier is actually a beneficial tool for the diagnosis of the breast cancer than several published studies [10].

Therefore, this research work demonstrates that improvements in accuracy, sensitivity and specificity are possible through automated image analysis. This will lead to a natural development of a CAE system capable of assisting health professionals in the painstaking task of tracing breast DCE-MRI in search of mass abnormalities and prognosis.


The authors would like to thank Dr.K.S.Murugan MD, DNB (Rad), fellow in MRI (Ger) of clarity imaging and Dr. R. Rupa, DMRD, DNB, Consultant Radiologist, Kovai Medical Center and Hospital (KMCH); Coimbatore, Tamil Nadu, India, for providing the breast DCE-MR images used for testing the algorithms proposed in the thesis and their valuable clinical advice.