High quality low-dose SPECT reconstruction using CGAN-based transformer network with geometric tight framelet

Single-photon emission computed tomography (SPECT) is a widely used diagnostic tool, but radioactive radiation during imaging poses potential health risks to subjects. Accurate low-dose single-photon reconstruction crucial in clinical applications of tomography. However, it remains challenging problem due the high noise and low spatial resolution reconstructed images. The aim study develop deep learning based framework for quality reconstruction. In proposed framework, conditional generative adversarial network (CGAN) was as backbone structure Residual Attention CSwin Transformer (RACT) block introduced basic building generator network. residual attention transformer has dual-branch structure, which integrates local modeling capability CNN global dependency improve More importantly, novel loss term on geometric tight framelet (GTF) designed better suppress image while preserving details greatest extent. Monte Carlo simulation software SIMIND produce images dataset evaluate performance method. results showed that method can reduce more preserve various situations compared several recent methods. To further validate method, we also verified generalization ability adaptable different level scenarios than other Our indicated tracer dose required without compromising