Evaluation of an Automated Whole-body Lesion Detection and Quantification Approach using PERCIST Criteria for 18F-FDG PET/CT in Oncology

1437 Purpose: To develop and evaluate a 3-D automated approach for whole-body lesion detection, segmentation and quantification using PERCIST criteria for 18F-FDG PET/CT in oncology. Methods: 18F-FDG PET/CT data of 30 oncology patients (33±14 BMI; 13±1 mCi FDG; 75±7 min p.i; TOF+PSF+OSEM (3i21s) 4x4x4 mm3 PET; mCT scanner) were collected for the assessment. PERCIST criteria based on liver spherical VOI statistics was applied to a whole-body contour to form initial individual lesions, followed by morphological processing (lesion split, cleanness, Boolean operations, edge adjustment, etc) to finalize the contours for each of the lesions. Bone segmentation was performed to separate lesions in bone from soft tissues. SUV/SUL peak was measured by moving a defined 1-mL spherical VOI over each of the tumor foci. Required procedures (36 steps) were specifically developed and built together into an all-in-one workflow in the MIMSoftwareInc platform to enable automated whole-body lesion detection and quantification (SUV/SUL peak, tumor volume and total lesion glycolysis). Lesion detection accuracy was evaluated by nuclear medicine physician. Results: The developed approach streamlined overall individual procedures from PET/CT misalignment correction and fusion, to PERCIST criteria application, lesion detection, VOI generation and contour optimization, to bone segmentation, quantitative measurements, display and database output, automatically. Depending on the number and size of lesions and the computational environment, a typical time of processing one study with 100 lesions was ~30 sec. It revealed an accuracy of 88±10% in lesion detection across patients. Lesions < 1ml cost additional burden in both processing time and measurement accuracy. More details and challenges as well as future development will be demonstrated in presentation. Conclusions: Innovative and efficient approaches to facilitate the process of reviewing and quantifying large number of lesions and patients in clinical PET/CT routines and research are expected. The study developed and evaluated a 3-D approach in such effort by utilizing PERCIST criteria and morphological processing, and it demonstrated a promising feasibility in automated whole-body lesion detection and quantification in oncologic 18F-FDG PET/CT.