Computer Analysis of Antibody Scans

Abstract A variety of tumours can be detected with radiolabelled antibodies and a gamma camera. Unfortunately, Low count rates and poor contrast combine to produce images that are often difficult to interpret. Subtraction, using a second isotope, may induce further artifacts. We have used automatic techniques to avoid observer bias, estimate statistical significance, and enhance resolution. Three separate techniques have been used: Firstly, the variance in the subtraction image was calculated and two standard deviations removed for each pixel and only remaining counts were regarded as significant. Secondly, noise was removed by processing with a Wiener filter and the result expressed in contour plots, each stepped in units of standard deviation. Thirdly, a non-linear deconvolution technique (maximum entropy) was used to remove both noise and camera blurring. Subtraction can then be effected using a local normalising factor, and lesions assessed by their statistical significance and shape. These techniques allow objective assessment of scans as well as improving resolution and sensitivity.