چكيده به لاتين
Breast cancer is the most frequent type of cancer and the second leading cause of cancer deaths among women worldwide. Magnetic Resonance Imaging has an important role for detecting breast cancer. Evaluation of the enormous amount of images for each patient is a time-consuming process, and it depends on a radiologist’s expertise and experience. Computer-Aided Detection (CAD) systems are introduced to help the radiologist for analyzing biomedical data. In this research, a CAD system is presented for detection and classification of cancerous lesions in breast MRI.
Segmentation of whole breast is the first step to perform an automatic analysis. Most of the proposed methods in literature for breast segmentation are based on the visible contrast between the breast region and surrounding chest wall. Due to similarity between gray-level values of fibroglandular tissue and pectoral muscle, these methods are not applicable for breasts with fibroglandular tissue connected to the pectoral muscle. This research proposes two atlas-based methods using chest region and chest wall templates which are applicable for both complex cases with fibroglandular tissue connected to the pectoral muscle and simple cases with high contrast boundaries. Moreover, a new similarity measure criterion based on the geometric features of the chest wall is defined to select the most similar atlas. Then, the cancerous lesions are detected by a region-growing algorithm incorporating with Fuzzy C-means (FCM) clustering and vesselness filter. This is the first time that FCM clustering and vesselness filter are incorporated in the seeded region-growing algorithm. In next step, three groups of features which are morphology, kinetics and texture, are extracted from the segmented lesions. In this research, a novel texture feature which is called Co-occurrence of Local Frequency Detector (CLFD), is introduced and the performance of it is compared with Grey-Level Co-occurrence Matrix (GLCM) and Local Binary Pattern (LBP). Since, the analysis of the non-mass-like enhancing lesions is the main challenge for segmenting and classifying the cancerous lesions of breast, these lesions are specifically investigated in this research. The extracted features are fed into the Support Vector Machine (SVM) classifier and the lesions are classified into benign and malignant classes.
At 5.29 false positives per case, the CAD system accurately detects 94% of the breast lesions. Also, the proposed system achieves AUC, accuracy, sensitivity and specificity values of 0.94, 0.88, 0.85 and 0.89, respectively to classify benign and malignant lesions. The results prove the effectiveness of the proposed CAD system for segmenting and classifying the mass-like and non-mass-like enhancing lesions.
