Penalized Maximum-Likelihood PET Image Reconstruction for Lesion Detection

Detecting cancerous lesions is one major clinical application of positron emission tomography (PET). Statistical reconstruction methods based on the penalized maximum-likelihood (PML) principle have been developed to improve image quality such as spatial resolution, noise variance, contrast-to-noise ratio, etc. However, these technical metrics do not necessarily reflect the performance of a specific clinical task. In this dissertation, we focus on a task-specific metric to evaluate image quality and optimize PML reconstruction for lesion detection through penalty design, resolution modeling, and dynamic imaging.;To assess lesion detectability in 3D images, we develop a multiview channelized Hotelling observer (mvCHO) to mimic the condition where a human observer examines three orthogonal views of a 3D image for detecting lesions. We derive simplified theoretical expressions that allow fast prediction of lesion detectability based on the mvCHO in PML reconstruction. The theoretical results are then applied to guide the design of a shift-variant quadratic penalty function in PML reconstruction to maximize lesion detectability in fully 3D PET. The proposed method is evaluated using computer-based Monte Carlo simulations as well as clinical patient data with superimposed lesions.;The second focus of this dissertation is to improve lesion detection through resolution modeling (RM). Incorporating resolution modeling within reconstruction has been shown to improve spatial resolution and tumor contrast, by modeling resolution degrading phenomena such as positron range, inter-crystal scatter and photon non-collinearity. Here we extend our theoretical analysis of the PML reconstruction to quantitatively evaluate the effect of image-space RM on detection performance. We then propose a task-based resolution modeling in PML reconstruction and validate it using computer-based Monte Carlo simulations as well as real data.;Furthermore, we extend our theoretical analysis of static PET reconstruction to dynamic PET. We study both the conventional indirect reconstruction and direct reconstruction for Patlak parametric image estimation. In the indirect reconstruction, Patlak parametric images are generated by reconstructing a sequence of dynamic PET images first and then performing Patlak analysis on the time activity curves (TACs) pixel-by-pixel. In the direct reconstruction, Patlak parametric images are estimated directly from raw sinogram data by incorporating the Patlak model into the image reconstruction procedure. The PML reconstruction is used in both the indirect and direct reconstruction methods. Simplified analytical expressions for evaluating lesion detectability on Patlak parametric images have been derived and applied to optimize the regularization parameter value for lesion detection. The proposed method is validated using computer-based Monte Carlo simulations. Good agreements between the theoretical predictions and the Monte Carlo results are observed. Both theoretical predictions and Monte Carlo simulation results show the benefit of the indirect and direct methods under optimized regularization parameters in dynamic PET reconstruction for lesion detection, when compared with the conventional static PET reconstruction.