A fast GPU code for full Monte Carlo based SPECT reconstruction

• To improve image quality in SPECT/CT reconstructions, various approximate recovery resolution techniques have been developed and implemented in clinical practice. However, optimal image reconstruction requires accounting for all physical interactions of the emitted photons in the individual patient. The objectives for this study were to develop a novel Monte Carlo (MC) code for fast simulation of individual image projections, and to implement these projections in ordered subset expectation maximum (OSEM) reconstructions of SPECT/CTimages.Method: The MC code was written in Compute Unified Device Architecture language for a computer with four graphic processing units (GeForce GTX Titan X, Nvidia, USA). This enables simulations of parallel photon emission from the voxels matrix (1283 or 2563). Each CT number was converted to attenuation coefficients for photo absorption, coherent scattering and incoherent scattering. The type of interaction was determined by the ratio of attenuation coefficients in the CT voxels. For photon scattering the deflection angle was determined by the differential scattering cross sections. The accepted angle for photon interaction with the crystal was determined from the diameter and height of the collimator hole. Predefined energy and spatial resolution kernels for the crystal were used. TheMCcode was implemented intoOSEMreconstruction of 177Lu, 111In and 99mTc SPECT/CT images. The National Electrical Manufacturers Association (NEMA) image quality phantom was used to evaluate the performance of the MC reconstruction in comparison with clinical standard OSEM reconstructions and clinical state-of-the-art OSEM reconstructions with recovery resolution corrections. Results and conclusion: The performance of the MC code was 500 millions photons/s. The required number of photons emitted per voxel for obtaining low noise in the simulated image was 400 for a 1283 voxel matrix. With this number of emitted photons/voxel the MC-based OSEM reconstruction with 10 subsets was performed within 60 s/iteration. The images converged after 2-4 iterations, depending on the sphere sizes in the NEMAphantom. Thereby, the reconstruction time was <4 minutes. The contrast-to-noise level was slightly improved with increased number of emitted photons/voxel, and the reconstruction time was linearly depending on the number of emitted photons/voxel. The signal-to-background for the spheres in the NEMA phantom was clearly improved with MC-based OSEM reconstruction: e.g. for 177Lu the improvement was 37% compared to standard OSEM and 20 % compared to state-of-the-art OSEM. Furthermore, visual inspection of clinical investigations revealed clearly improved resolution and contrast with MC-based reconstruction.

Ämnesord

TEKNIK OCH TEKNOLOGIER  -- Medicinteknik (hsv//swe)

ENGINEERING AND TECHNOLOGY  -- Medical Engineering (hsv//eng)

Publikations- och innehållstyp

vet (ämneskategori)

kon (ämneskategori)