Deep Learning Enhanced Virtual Photon-Counting Micro-CT Imaging Platform for Preclinical Cancer Studies
The simultaneous multi-energy capability of photon-counting detectors (PCDs) enables CT imaging with multiple contrast agents, which is particularly valuable for preclinical cancer studies such as photon-counting CT (PCCT) imaging of mouse models of head and neck squamous cell carcinoma (HNSCC). However, optimization of PCCT imaging parameters in vivo is constrained by radiation dose limits and the lack of ground truth measures of contrast uptake. Simulation-based approaches can address these limitations by enabling parameter optimization without radiation exposure to animals. Here, we describe our progress towards a simulation pipeline for preclinical PCCT cancer imaging. Using real PCCT scans of mice with HNSCC, we trained a denoising diffusion probabilistic model (DDPM) to generate tumor material maps and integrated these with a vascularized MOBY phantom. Matched real and simulated 3D-printed phantoms were used to calibrate simulated PCCT reconstructions via a two-step adjustment consisting of matrix multiplication rescaling and a multi-layer perceptron. Material decomposition from PCCT imaging simulation with two-step adjustment of our MOBY with HNSCC was compared to the ground truth using tumor metrics. This work demonstrates a workflow for truth-based simulation of preclinical PCCT cancer imaging that can help optimize parameters for future studies.
