Automated SPECT-Based Segmentation for Quality Assurance of CT-Delineated Tumor Volumes for 131I Tositumomab Therapy of Non-Hodgkins Lymphoma
R Thorwarth*, Y Dewaraja, S Wilderman, M Kaminski, A Avram, P Roberson, Univ Michigan, Ann Arbor, MISU-E-J-186 Sunday 3:00PM - 6:00PM Room: Exhibit Hall
Purpose: CT segmentation of tumor volumes suffers from many error sources which can impact the calculation of absorbed dose and 4D bio-effect modeling. Determination of gross tumor volumes from SPECT scans of anti-CD20 monoclonal antibody uptake was explored as a quality assurance step. Investigated were two count threshold methods (optimal and constant threshold) and one edge detection method.
Methods: Patient data were selected from a dataset of refractory non-Hodgkins lymphoma studies of ¹³¹I radioimmunotherapy. The calibration phantom geometry consisted of seven (1 mL to 95 mL) spheres in an elliptical tank. The background activity was 5 mCi with sphere activities adjusted for an approximate 6:1 concentration ratio. Phantom scans were repeated with no background activity. The patient and calibration sphere data were collected on a Siemens Symbia TruePoint SPECT/CT scanner and reconstructed with in-house OS-EM software. The optimal threshold technique used a calibration of threshold vs volume. The constant threshold method (40% and 60% thresholds explored) used a calibration of volume correction factor vs volume. The edge detection method used the Canny algorithm to map edges and determine an edge correction factor.
Results: Challenges included 1) resolving tumor volume from nearby significant uptake objects (organs, vessels or other tumors) and 2) identifying volume for large tumors with significantly non-uniform activity. A constant threshold of 60% with background and volume correction performed best. The determined volumes from patient data were 0.70-0.73 (>12mL) or 0.44-0.69 (<12mL) of the CT-delineated volumes. The lower than unity values were attributed to non-sharp activity boundaries. A priori volume estimates were required for object radii less than the scanner resolution (i.e. <12mL).
Conclusion: The constant threshold method can be used to check the validity of CT-delineated tumor volumes. An additional volume correction factor is required to account for non-sharp tumor edges.
Funding Support, Disclosures, and Conflict of Interest: Dr M Kaminski receives research support from GlaxoSmithKline and royalties from Bexxar.