Brain single-photon emission computed tomography (SPECT) imaging using I-123 DaTSCAN is an effective tool for the diagnosis and follow-up of Parkinson disease. Reducing acquisition time decreases the likelihood of patient motion, improves patient comfort, and increases scanner throughput. However, shorter acquisition times often result in degraded image quality with reduced diagnostic value. This study aimed to use deep learning (DL) approaches to enhance the image quality of accelerated SPECT DaTSCAN acquisitions. A total of 85 patients were scanned on a CZT-based GE StarGuide scanner with a standard acquisition time of 15 min. To simulate fast acquisition protocols, raw list-mode data were retrospectively undersampled to represent 20%, 25%, and 50% of the original acquisition time and then reconstructed using the clinical routine protocol. A modified U-Net DL architecture incorporating attention and transformer mechanisms was trained to convert the low-count images into full-time equivalents. The output images were compared to full-time reference images using subjective image quality assessment by nuclear medicine physicians as well as region-wise and voxel-wise quantitative metrics on the test set. Both voxel-wise and region-wise metrics showed significant improvement when using the DL models for all fast acquisition times. Subjective image quality evaluations confirmed that the DL-predicted images were noticeably superior to the original low-count reconstructions in terms of perceived image quality and diagnostic confidence. The mean absolute percentage error (MAPE) was reduced from ~ 30 to ~ 18% for fast 1/5 acquisition time images and from ~ 26 to ~ 16% for 1/4 acquisition time images. Our acquisition time-specific deep learning models effectively restore information loss in fast SPECT acquisitions and reduce image noise, without compromising diagnostic value. The models offer a promising approach for faster, patient-friendly brain SPECT imaging without sacrificing image quality or quantitative accuracy.