Massively parallel image reconstruction for the BNL breast scanner PET tomograph using CUDA

We describe the implementation of an image reconstruction algorithm using the parallel processing capabilities of graphics processors. We are designing a new breast scanner which will allow simultaneous acquisition of PET and MRI images. The breast scanner is based on the technology of the much smaller RatCAP PET detector. The image reconstruction of the breast scanner poses a significant computing challenge, which we hope to alleviate with the processing power of modern GPUs. We describe the status of the reconstruction and discuss goals and future possible improvements.