Please use this identifier to cite or link to this item:
|Title:||Automated tumor detection and treatment response assessment with FDG-PET dynamic studies|
|Publisher:||Society of Nuclear Medicine|
|Source:||The Journal of nuclear medicine, 2010, v. 51, no. suppl. 2, 25 How to cite?|
|Journal:||The Journal of nuclear medicine|
|Abstract:||Objectives: This study aims to develop a novel accurate method for guided automated tumor detection initially and fully automated tumor detection after the therapy, to facilitate the assessment of the treatment response.|
Methods: In dynamic PET studies, the tissue time activity curves (TACs) can be modeled by linear subspaces with additive Gaussian noise, so that the unsupervised one class support vector cluster (OSVC) combined with the generalized likelihood ratio test (GLRT) subspace matched method can be introduced to detect tumors. The TACs from primary tumors and background ROIs, each manually roughly drawn in the pre-therapy PET image, are adopted to identify linear subspaces for GLRT. Then, the unsupervised OSVC method is applied to delineate outliers of each ROI for principle component analysis (PCA) analysis to get accurate linear subspaces. The PCA dominating components are constructed the subspaces for the GLRT subspace matched detector, which can generate the voxel by voxel GLRT statistic maps for the pre/post chemotherapy PET images. The resulting statistic maps are thresholded to locate the primary and metastatic lesions in the pre/post chemotherapy PET images.
Results: The method is validated by seven sets of patient data with lung cancer who were underwent 30 minutes dynamic PET/CT imaging before and one day after chemotherapy. The identified lesions were evaluated by experienced oncologists. In all cases, tumors and metastatic lesions are correctly identified automatically in the pre/post chemotherapy PET images, including those lesions with activity update reduced significantly after chemotherapy.
Conclusions: This novel method can be potentially used in clinical practice for automated tumor detection and treatment response assessment with minimum supervision in its initial diagnosis.
|Appears in Collections:||Journal/Magazine Article|
Show full item record
Checked on Jan 22, 2017
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.