Automatic Lesion Detection for Measuring Response using Dynamic FDG-PET

Abstract

AbstractThe fluorodeoxyglucose positron emission tomography (FDG-PET) has demonstrated advantages in the assessment of tumor response to therapy. The high-quality quantitative assessment usually relied on the accurate lesion detection. Hence, we aim to develop an objective method to automatically detect lesions for measuring tumor response to therapy using dynamic FDG-PET images. In our proposed method, the time-activity curves of dynamic PET images were modeled by linear subspaces with additive Gaussian noise. The matched subspace detectors coupled with unsupervised one class support vector cluster were introduced to detect the malignant lesions both in baseline and followup images. The physiological parameter was then estimated for each identified lesion in order to measure the tumor response to therapy. From the results, the automatic lesion detection method showed its great potential in clinical practice for facilitating tumor diagnosis and therapy assessments.