Single-photon emission computed tomography coupled with computed tomography (SPECT/CT) is an essential nuclear medicine imaging technique that captures images from radiotracers administered to patients. However, standard SPECT imaging protocols require prolonged acquisition times. Although rapid scan protocols can significantly reduce acquisition time, they introduce challenges such as increased Poisson noise, artifacts, and reduced spatial resolution, potentially compromising diagnostic accuracy. To address these issues, we propose AGFuse-Net, a multi-modality deep learning framework that reconstructs high-quality SPECT images from scans acquired in just 1/8 of the standard protocol duration. Our method integrates anatomical features from CT images with functional information from rapid SPECT scans using anatomy-guided attention gates. Specifically, we introduce a feature fusion module that employs spatial attention mechanisms to preserve anatomical boundaries and improve lesion detectability. Experimental evaluations demonstrate that AGFuse-Net effectively suppresses noise, eliminates artifacts, and preserves structural details, achieving image quality comparable to standard-duration SPECT scans. Furthermore, a clinical reader study validates that our approach supports accurate clinical diagnoses while leveraging the advantages of reduced imaging time.