Automated segmentation of tumour changes in temporal PET-CT data

Positron emission tomography and computed tomography (PET-CT) is widely accepted as the best imaging in the management of cancer. To assess response to treatment, multiple scans of a patient are usually acquired and tumour changes are analysed at different stages during the treatment. However, it is difficult to assimilate these changes, especially where a patient has in excess of e.g. four temporal scans. We aimed to advance current single-time PET tumour segmentation algorithms to multiple-time PET studies. We therefore proposed a new treatment response confidence connected region growing (TRCCRG) algorithm. Our algorithm is interactive and includes a combined function of Radial profile plot and Rodbard curve fitting as a measure of tumour likelihood for the morphologically expanded volume in successive scans. Evaluation with 10 simulated PET data and 6 clinical lymphoma PET-CT patients (each with 4 studies) demonstrated that the proposed method was successful in detecting and delineating tumours.