Super-resolution of PET image based on dictionary learning and random forests

Abstract Positron emission tomography (PET) is an imaging technique for nuclear medicine and clinical diagnosis that is widely used in oncology and clinical medicine. However, PET has limitations related to its lower resolution than other medical imaging modalities, such as X-ray computed tomography (CT) and magnetic resonance imaging (MRI). In this paper, we propose an improved super-resolution (SR) method based on dictionary learning and random forests for the PET system to improve the resolution of PET images. First, we process the acquired high-resolution (HR) PET images ourselves to obtain multiple types of low-resolution (LR) PET images. Next, we directly train the mapping from LR to HR PET patches using random forests. Experimental results based on both clinical and medical images show that the proposed method is effective in improving PET image quality in terms of numerical criteria and visual results. The proposed method can minimize noise and artifacts without blurring the edges of the PET image, which can preserve important structural details, such as those indicating lesions. Therefore, the proposed method has excellent potential for applications in actual clinical and medical systems.