Segmentation of PET volumes by iterative image thresholding.

UNLABELLED The segmentation of metastatic volumes in PET is usually performed by thresholding methods. In a clinical application, the optimum threshold obtained from the adaptive thresholding method requires a priori estimation of the lesion volume from anatomic images such as CT. We describe an iterative thresholding method (ITM) used to estimate the PET volumes without anatomic a priori knowledge and its application to clinical images. METHODS The ITM is based on threshold-volume curves at varying source-to-background (S/B) ratio acquired from a body phantom. The spheres and background were filled either with (18)F-FDG or Na(124)I ((124)I). These calibrated S/B-threshold-volume curves were used in estimating the volume by applying an iterative procedure. The ITM was validated with a PET phantom containing spheres and with 39 PET tumors that were discernable on CT by using whole-body (18)F-FDG (15 patients) and (124)I PET/CT (9 patients): The measured S/B ratios of the lesions were estimated from PET images, and their volumes were iteratively calculated using the calibrated S/B-threshold-volume curves. The resulting PET volumes were then compared with the known sphere inner volume and CT volumes of tumors that served as gold standards. RESULTS Phantom data analysis showed that the S/B-threshold-volume curves of (18)F-FDG and (124)I were similar. The average absolute deviation (expressed as a percentage of the expected volume) obtained in the PET validation phantom was 10% for volumes larger than 1.0 mL; sphere volumes of 0.5 mL showed a significantly larger deviation. For patients, the average absolute deviation for volumes between 0.8 and 7.5 mL was about 9% (31 lesions), whereas volumes larger than 7.5 mL showed an average volume mismatch of 15% (8 lesions). CONCLUSION The ITM sufficiently estimated the clinical volumes in the range of 0.8-7.5 mL; volumes larger than 7.5 mL showed greater deviations that were still acceptable. These findings are associated with the limitation of the ITM. The ITM is especially useful for lesions that are only visible on PET. As a consequence, the lesion dosimetry is feasible with sufficient accuracy using PET images only.