Improving the Detection of The Prostrate in Ultrasound Images Using Machine Learning Based Image Processing
This work aims to develop a method for accurate prostate segmentation in transrectal ultrasound (TRUS) images. However, accurate prostate segmentation remains a challenging task for many reasons, such as the missing/ambiguous boundary between the prostate and surrounding organs, the presence of shadow artifacts, and intra-prostate intensity heterogeneity. This work proposes a three-cascaded prostate segmentation framework, using only a few manually delineated points as a prior, including (1) an improved principal curve-based model is used to obtain the data sequences consisting of data points and projection indexes; (2) an improved differential evolution-based artificial neural network is used for training to decrease the model error; and (3) the artificial neural network's parameters are used to explain the smooth mathematical description of the prostate contour. Experimental results show that our proposed method achieves superior segmentation performance in prostate TRUS images than state-of-the-art methods.
