A statistical clustering approach to visualizing the relationship between early and delayed images in whole-body FDG-PET

We propose a new method for diagnostic assistance in oncology, [fluorine-18]-2-fluoro-2-deoxy-d-glucose (FDG)-positron emission tomography (PET). Early and delayed scans were performed on 10 patients with lung cancer by use of an ECAT EXACT 47 PET scanner, and standardized-uptake-value (SUV) images were created. Three segmentation (S1, S2, and S3) maps were created from the early and delayed SUV images according to various thresholds (SUVthreshold = 2.0, 2.5, and 3.0) based on the early image and the percentage change defined as (SUVdelayed − SUVearly) × 100/SUVearly. Voxels that had larger voxel values in their early images than the SUVthreshold were clustered into three classes: S1 if the percentage change was larger than 10, S2 if the percentage change was between 0 and 10, and S3 if the percentage change was negative. The S1 segments showed malignant lesions clearly; however, the S2 segments showed an SUV that had decreased from the S1 areas due to the partial volume effect or misalignment between the early and delayed scans. The S3 areas showed benignity or physiologic accumulation. The segmented images, S1, S2, and S3, were useful for clinical diagnosis with dual-phase FDG-PET scans and offer an easy way of exploring the longitudinal alteration in the SUV.