[Diagnosis of prostate cancer using SVM-based ultrasound images].

This paper presents a computer-aided diagnosis method for prostate cancer detection using Trans-rectal ultrasound(TRUS) images. Firstly, statistical texture analysis is implemented in every ROI in segmented prostate images. From each ROI, grey level difference vector features, edge-frequency features and texture features in frequency domain are constructed. Then, the number of features is reduced using ANOVA statistics to select the optimal feature subset. Finally, SVM is applied to the selected subset for detecting the cancer regions. Experimental results show that the proposed algorithm can recognize and detect the cancer images effectively so as to supply essential information for a diagnosis.

[Breast tumor classification based on serial grayscale ultrasound images].

This paper proposes an algorithm of evaluating the compression depth, and then to extract four normalized mammary elasticity characteristic parameters with respect to the compression depth. The classification experiments show that these elasticity parameters have a good capability in determining whether the tumor is benign or malignant, and if combined with morphological parameters, the accuracy, sensitivity and specificity can be improved and increased to 95.19%, 98.82% and 92.16%, respectively.

[Segmentation of the prostate ultrasound image based on an improved geodesic active contour model].

In this paper, we propose that, the need of the costly re-initialization procedure can be completely eliminated by using the variation formulation, thus increasing the speed of computing operations. The edge detecting function in the geodesic active contour model is improved by incorporating a prior knowledge. The accuracy of the segmentation algorithm can be enhanced by using the minimal variance. Experimental results show that the proposed algorithm can segment the prostate ultrasound images effectively and avoid the problem of contour leakage.