Premium
SU‐FF‐J‐128: An Automatic Method to Delineate 18F‐FDG PET Tumor Volumes for Radiation Treatment Planning
Author(s) -
Teo BK,
Seo Y,
Xia P,
Bacharach S L,
Franc B L,
Hasegawa B H
Publication year - 2007
Publication title -
medical physics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.473
H-Index - 180
eISSN - 2473-4209
pISSN - 0094-2405
DOI - 10.1118/1.2760633
Subject(s) - nuclear medicine , positron emission tomography , partial volume , radiation treatment planning , medical imaging , image resolution , radiation therapy , computer science , medicine , artificial intelligence , radiology
Purpose: To develop and evaluate an automatic method of segmenting biological tumor volumes in 18 F‐FDG‐PET images for radiation treatment planning. Method and Materials: We developed a method of correcting errors in biologic tumor volume due to the finite spatial resolution in 18 F‐FDG PET images. The PET image is acquired, then used to generate a “spillover image” S , defined as the difference between an iteratively deconvolved image and the original PET image. The biologic tumor boundary was defined along the contour where the spillover image S changes sign. This algorithm was validated with tumors simulated with various sizes (11–28 mm) and background levels, and also with PET/CT images from 10 patients having FDG‐avid head/neck tumors having anatomical boundaries clearly defined with CT. PET‐derived biologic tumor volumes were compared against the CT‐derived anatomic tumor volumes and against the volumes extracted by applying a simple threshold to the PET image. Results: The biologic tumor volumes derived with the proposed method matched the CT‐derived anatomic tumor volumes for all tumor sizes having background activities from 0 to 33% of the tumor activity in the simulated data, and agreed to within 90% for all tumors evaluated in the patient data. For all tumor sizes, the PET volumes derived from this algorithm matched more closely with the CT‐derived volumes than those obtained with a fixed threshold applied to the PET images. Conclusion: The proposed algorithm incorporates the spatial resolution of the imaging system to automatically segment biologic tumor volumes. The algorithm automatically compensates for differences in tumor size and tumor:background contrast, and does not require a fixed threshold to be applied to the PET image. Such a scheme will be useful for defining biologic tumor volumes with 18 F‐FDG PET for radiation treatment planning. Research sponsored by Siemens Oncology Care.