Development of Segmentation and Classification Techniques for Prostate and Breast Ultrasound Image Analysis


Student thesis: Doctoral Thesis

View graph of relations


Related Research Unit(s)


Awarding Institution
Award date20 Dec 2017


Breast and prostate cancer are the most common cancers in women and men, respectively, in the world. In Hong Kong, prostate cancer is the fifth leading cause of cancer deaths in male and there is an increasing trend in both number of deaths and death rate of this disease [Dep13]. As for female, breast cancer is the most common cancer in Hong Kong. And the number of cases diagnosed has tripled from 1993 to 2014 [20115]. Efforts to improve survival rate of these two cancers have been highly successful. This improvement is largely due to the development of diagnosis and treatment methods for these two types of cancers.

Prostate segmentation from transrectal ultrasound (TRUS) images plays an important role in the diagnosis and treatment planning of prostate cancer. In this study, a fully automatic slice-based segmentation method was developed to segment TRUS prostate images. The initial prostate contour was determined using a novel method based on the radial bas-relief (RBR) method and a false edge removal algorithm we proposed here. 2D slice-based propagation was used in which the contour on each image slice was deformed using a level-set evolution model driven by edge-based and region-based energy fields generated by dyadic wavelet transform. The optimized contour on an image slice propagated to the adjacent slice, and subsequently deformed using the level-set model. The propagation continued until all image slices were segmented. To determine the initial slice where the propagation began, the initial prostate contour was deformed on each transverse image. A method was developed to self-assess the accuracy of the deformed contour based on the average image intensity inside and outside of the contour. The transverse image on which highest accuracy was attained was chosen to be the initial slice for the propagation process. Evaluation was performed for 336 transverse images from 15 prostates that include images acquired at mid-gland, base and apex of the prostates. The average mean absolute difference (MAD) between algorithm and manual segmentations was 0.79 ± 0.26 mm, which is comparable to results produced by previously published semi-automatic segmentation methods. Statistical evaluation shows that accurate segmentation was not only obtained in mid-gland, but also the base and apex regions.

The novel method based on RBR method and a false edge removal algorithm that have been used for prostate contour initialization were extended to the study of breast cancer. Accurate detection and clinical decision of breast lesion can increase the chance of survival significantly. As a new developed technique, supersonic shear wave imaging has become a valuable tool for tumor classification. This study is to develop an automatic diagnosis system that can outline lesion boundary automatically and distinguish benign from malignant breast lesion based on a set of elastographic features, with biopsy result as the reference standard.

Conventional ultrasonography (US) and supersonic shear-wave elastography (SWE) elastography images of 187 breast tumors were obtained. 50 tumors were used as parameter tuning. After ultrasound elastography rebuilding, automatic lesion segmentation was applied on the rest 187 tumors (113 benign, 74 malignant) ultrasound images, the segmentation result was mapped into the elastography. Then a total 286 features of multiple banding peripheral tissue were extracted with compared by feature selection method. A support vector machine (SVM) classifier was used for optimum classification via combination of these features. Classification accuracy was evaluated by comparison with the results obtained in histopathologic examination of biopsy samples. Classification performance based on SWE was compared with manual rating of B-mode US images based the Breast Imaging-Reporting and Data System (BI-RADS).

Quantitative evaluation with elastic properties of multiple banding peripheral tissue exhibited good discriminating ability between malignant and benign lesions detected in ultrasound. The pipeline was evaluated on 137 lesions consisting of 49 malignant and 88 benign lesions. Feature selection was performed 20 times using different sets of 18 lesions (10% of the study population). Leave-one-out SVM classification was performed in each of the 20 experiments with a mean sensitivity, specificity and accuracy of 90.7%, 94.6% and 93.1% respectively.

The result verified that the proposed automatic CAD system allows efficient outlining the lesion area objectively. Feature was extracted to determine whether the lesion is malignant. The final classification result indicates that the system has the ability to determine lesion boundary from ultrasound image rapidly and elastographic features are shown to be useful in breast lesion classification.