Thermal theory based feature extraction method for high noise PET images

In this paper, we propose an image feature extraction method based on the thermal theory for high noise PET images. The PET imaging, which records physiological activities of tissues, is broadly used to provide diagnostic information for investigating disorders in clinical. To extract desired regions of interest (ROIs) from noisy PET images for clinical applications is an important issue. The proposed method hypothesizes an image as a pseudo-object, and each pixel with different intensity in the image is defined as different pseudo-substance and has its specific heat capacity. Observing physical thermodynamic phenomenon, the pseudo-substances those have similar specific heat capacity characteristics will be fuse by heating and cooling the pseudo-object over and over. That is, image pixels with similar intensity will converge to closed level, and the desired image features can be extracted. To evaluate the performance of the proposed method, a set of normal FDOPA-PET images and three different abnormal cases include AVM, NPC and brain tumor PET images are used in this study. As results, the difference between the extraction regions obtained from presented method and the ROIs drawn manually by clinical physician is less than 1% in average. Furthermore, the method also features automatic extraction procedure and shorts processing time.