Segmentation and analysis of colour Doppler images of tumour vasculature.

A technique has been developed to segment (separate), from a digitized colour Doppler video image, the colour and greyscale information and then to estimate from the colour information the original mean Doppler frequency shift data from which the image was created. The remapped velocity image is then analysed to extract numerical features of the tumour vasculature. The present version of the software is set-up to work for an Acuson 128 colour Doppler system using the V4 colour scale, although it should work well with any system which modulates only two colours for each flow direction and displays a colour calibration scale at the side of the image. Accuracy of classification of greyscale, colour and flow direction was estimated as being in the region of 95% for typical breast tumour images. The degree of agreement between the remapped colour velocity values and those stated by the scanner at the same image locations was evaluated in terms of the linearity of the relationship (> 99%), precision (better than +/- 5%) and accuracy (better than 7.6%). We investigated the value, for diagnosis and assessment of response of, a variety of characteristics of the displayed vascularity. At present, the software calculates the following vascular image features within any region of interest defined by the operator: mean displayed velocity, maximum displayed velocity, standard deviation of displayed velocity, total area occupied by colour signal, percentage area occupied by colour signal, area integral of displayed velocity and the total displayed velocity per unit area.