Automatic detection of closed tumor contours in medical ultrasound images on the basis of level-dependent spatial summation

The development of a tumor may be monitored over a period of time on the basis of its contours in a series of B-mode ultrasound images. However, contour detection is hampered by specific noise and is therefore often performed manually. This paper proposes a new approach to the automatic estimation of closed contours of hypoechoic and hyperechoic tumors. First, local averaging of gray level information in the image is performed by Level-Dependent Spatial Summation. The filtered image exhibits reduced speckle contrast while the edges of macro structures are sufficiently well preserved to allow automatic contour detection. Second, an optimum threshold is automatically selected to segment the filtered image in “tumor” and “background”. Using the resulting binary image, the tumor contour is estimated as the collection of tumor pixels that touch the background. Encouraging results are presented of the automatic detection of the contour of a hypoechoic breast tumor