Fusion of image reconstruction and lesion detection using a bayesian framework for PET/SPECT

We propose a new concept that fuses image reconstruction and lesion detection in PET/SPECT, and develop a MAP reconstruction method that produces separately a normal uptake image and an abnormal lesion image. In this method, a radiotracer image is modeled by a sum of a smooth background image and a sparse spot image, and each image is regularized by the different smoothness and/or sparseness penalties in the reconstruction cost function. To minimize the cost function containing the two image variables, an iterative alternating method is developed. Through computer simulation studies, we show that the proposed method achieves the separate reconstruction of the background image and the spot image well, and outperforms the conventional ML and MAP reconstruction methods in terms of visual image quality and contrast-noise performance. Finally, we show a preliminary reconstructed image of a real PET data.