Purpose: To develop biophysics-based method for estimating perfusion Q fromarterial spin labeling (ASL) images using deep learning. Methods: A 3D U-Net(QTMnet) was trained to estimate perfusion from 4D tracer propagation images.The network was trained and tested on simulated 4D tracer concentration databased on artificial vasculature structure generated by constrained constructiveoptimization (CCO) method. The trained network was further tested in asynthetic brain ASL image based on vasculature network extracted from magneticresonance (MR) angiography. The estimations from both trained network and aconventional kinetic model were compared in ASL images acquired from eighthealthy volunteers. Results: QTMnet accurately reconstructed perfusion Q fromconcentration data. Relative error of the synthetic brain ASL image was 7.04%for perfusion Q, lower than the error using single-delay ASL model: 25.15% forQ, and multi-delay ASL model: 12.62% for perfusion Q. Conclusion: QTMnetprovides accurate estimation on perfusion parameters and is a promisingapproach as a clinical ASL MRI image processing pipeline.