Factor analysis: delineation of organ structures and automatic generation of in- and output functions in PET studies of prostate cancer

Factor analysis (FA) is used for extracting the properties of dynamic sets. Objective: FA was applied to dynamic PET studies to create factor images, from which ROIs were derived and input and output functions generated. These functions were subsequently used for kinetic modeling. This non-invasive, automated, and image based analysis should permit routine application of quantitative PET in cancer patients. Methods: in men with prostate cancer, dynamic PET studies were acquired on an ECAT HR+ system. After administration of 250-300 MBq of C-11 labeled acetate, data were acquired during 20 min. The framing rate was 12/spl times/10, 9/spl times/20, 5/spl times/60, 2/spl times/300 sec with a total of 28 frames. The images were reconstructed with iterative algorithms, a MAP for transmission, and OS-MLEM for emission scans. The body contour was determined with a 40% threshold on the transmission images. This threshold assured exclusion of the bed. All voxels included in the body contour were used for further processing. FA then extracted the shape of the pure time activity curves (TACs) of vascular input and tumor output functions. The factors were used to create functional images, from which ROIs could be generated with thresholding techniques. The ROIs were used to create image based TACs. Results: The automated procedure generated reliable curves in all patients. Since the magnitude of the factors is normalized, TACs have to be adjusted using a scale factor. Two methods were utilized: (1) reversed normalization, (2) image based parameters. In principle, the factors generated by FA have no spillover and are pure vascular curves. The method is operator independent and reproducible. Processing time was 7 min/patient on an UltraSPARC5. Conclusion: FA can noninvasively generate input and output functions from dynamic PET data. The automated procedure generated curves corresponding to vessels and tumors, and had a success rate of about 80%. This processing tool facilitates PET as a reproducible quantification method in routine oncological applications.