چكيده به لاتين
Nowadays, breast cancer is the leading cause of death for women all over the world. Since the reason of breast cancer is unknown, early detection of the disease plays an important role in cancer control, saving lives and reducing costs. Among different modalities, automated 3-D breast ultrasound (3-D ABUS) is a new and effective imaging modality which has attracted a lot of interest as an adjunct to mammography for women with dense breasts. In this thesis, two computer aided systems are proposed to assist radiologists: (1) a computerized system for finding the location of masses and 2) a computerized system for mass segmentation.
For mass detection, a multi stage algorithm has been designed. In the first step, a denoising method called Optimized Bayesian Non-Local Mean (OBNLM) filter is used to reduce the speckle noise. Consequently, preliminary suspicious candidate regions are extracted using a novel algorithm based on iso-contours. Afterwards, different domain specific filters such as area filter and circularity filter are utilized to exclude false positive regions. Then, a cascaded ensemble classifier consisting four Random Under-Sampling Boosting (RUSBOOST) classifiers is applied for further reduction of false positives. The performance of the proposed mass detection method was evaluated on 104 volumes from 74 patients, including 112 cancers. Based on Free Response Operating Characteristic (FROC) analysis, the CADe system achieved the region-based sensitivity and case-based sensitivity of 68% and 76% at one false positive per image.
In the presented mass segmentation method, a two-stage algorithm is used which considers the shape information of training masses to improve the segmentation accuracy. In the first step, an adaptive region growing algorithm is introduced to achieve a rough estimation of the mass boundary. Then, the resultant contour is fed to a novel edge based geometric deformable model as an initial contour. In a dataset of 50 masses including 38 malignant and 12 benign lesions, the proposed segmentation method achieved a mean Dice of 0.74±0.19 which outperformed the adaptive region growing with a mean Dice of 0.65±0.2 (p-value<0.02). Moreover, the resulting mean Dice was significantly (p-value<0.001) better than that of distance regularized level set evolution method (0.52±0.27). The supervised method presented in this paper achieved accurate mass segmentation results in terms of Dice measure. Moreover, for further comparison, a state-of-the-art deep learning based segmentation method called 3-D U-Net was used which achieved a mean Dice of 61%.
