In iterative image reconstruction, different structures may reconstruct or evolve at different rates with iterations. From the difference between the images at different iterations, we extract feature images based on the evolution characteristics of different structures. We then use the feature images for advanced image processing. A TOF list-mode OSEM algorithm was used for iterative image reconstruction. A methodology was established to calculate feature images which distinguish regions of fast evolution (presumably associated with large and uniform structures) and slow evolution (associated with small structures such as lesions and cold regions). The feature images were then used to guide the in-reconstruction noise-suppression approaches (such as regularized reconstruction) and post-reconstruction noise-suppression approaches. Phantom and patient studies were used to demonstrate the proposed technique and its advantages. Results showed that with the feature images even very simple image domain techniques could achieve superior performance for the intended applications.