A general framework for automatic detection of matching lesions in follow-up CT

In follow-up CT examinations of cancer patients, therapy success is evaluated by estimating the change in tumor size from diameter or volume comparison between corresponding lesions. We present an algorithm that automatizes the detection of matching lesions, given a baseline segmentation mask. It is generally applicable and does not need an organ mask or CAD findings, only a coarse registration of the datasets is required. In the first step, lesion candidates are identified in a local area based on gray value filtering and detection of circular structures using a Hough transform. On all candidate voxels, a template matching is performed minimizing normalized cross-correlation. The method was evaluated on clinical follow-up data comprising 94 lung nodules, 107 liver metastases, and 137 lymph nodes. The ratio of correctly detected lesions was 96%, 84% and 85%, respectively, at an average computation time of 0.9 s per lesion on a standard PC.