Accurate quantification of cerebral blood flow (CBF) is essential for the diagnosis and assessment of a wide range of neurological diseases. Positron emission tomography (PET) with radiolabeled water (15O-water) is considered the gold-standard for the measurement of CBF in humans. PET imaging, however, is not widely available because of its prohibitive costs and use of short-lived radiopharmaceutical tracers that typically require onsite cyclotron production. Magnetic resonance imaging (MRI), in contrast, is more readily accessible and does not involve ionizing radiation. This study presents a convolutional encoder-decoder network with attention mechanisms to predict gold-standard 15O-water PET CBF from multi-sequence MRI scans, thereby eliminating the need for radioactive tracers. Inputs to the prediction model include several commonly used MRI sequences (T1-weighted, T2-FLAIR, and arterial spin labeling). The model was trained and validated using 5-fold cross-validation in a group of 126 subjects consisting of healthy controls and cerebrovascular disease patients, all of whom underwent simultaneous $15O-water PET/MRI. The results show that such a model can successfully synthesize high-quality PET CBF measurements (with an average SSIM of 0.924 and PSNR of 38.8 dB) and is more accurate compared to concurrent and previous PET synthesis methods. We also demonstrate the clinical significance of the proposed algorithm by evaluating the agreement for identifying the vascular territories with abnormally low CBF. Such methods may enable more widespread and accurate CBF evaluation in larger cohorts who cannot undergo PET imaging due to radiation concerns, lack of access, or logistic challenges.