Junyu Wang 王君宇

I am a Postdoctoral Scholar at Stanford University with solid expertise on signal processing, image processing, computer vision and machine learning centralized on cardiac magnetic resonance imaging (CMR). Specifically, my PhD dissertation is about advanced image acquisition and reconstruction techniques for first-pass CMR perfusion imaging.

CMR is a medical imaging technology for non-invasive assessment of the function and structure of the cardiovascular system. The development of CMR is an active field of research and continues to see a rapid expansion of new and emerging techniques.

High-resolution Spiral First-pass Myocardial Perfusion Imaging

DEep learning-based rapid Spiral Image REconstruction (DESIRE) for High-resolution Spiral First-pass Myocardial Perfusion Imaging

DEep learning-based rapid Spiral Image REconstruction (DESIRE) for High-resolution Spiral Real-time Cine Imaging

High-resolution Cartesian Perfusion Imaging with Deep learning-based Rapid Image Reconstruction

High-resolution Spiral First-pass Myocardial Perfusion Imaging

Coronary artery disease (CAD) is a major public health concern. According to a report from the American Heart Association in 2021, an estimated 18.2 million adult Americans have CAD. CAD is responsible for 1 in every 7 deaths in the United States. Cardiac magnetic resonance (CMR) quantitative myocardial first-pass perfusion imaging is a non-invasive and non-ionizing technique for diagnosing CAD which provides an accurate assessment of myocardial ischemia and a comprehensive evaluation of myocardial function and infarction.

Despite multiple potential advantages of CMR perfusion imaging, current clinically available techniques have limited in-plane spatial resolution (~2-3 mm) and incomplete heart coverage, which impede the assessment of transmural perfusion differences and underestimate the extent of ischemia. With higher spatial resolution, there is an increased ability to detect transmural perfusion differences between the epicardium and the endocardium, which could improve the ability to detecting obstructive CAD.

We aim to develop fast and high-resolution quantitative perfusion imaging techniques with whole-heart coverage. Specifically, we have demonstrated excellent performances utilizing variable density spiral trajectories, parallel imaging, compressed sensing, outer volume suppression (OVS) and simultaneous multi-slice (SMS) techniques.

Example cases for 2D (a) and SMS MB=2 w/o OVS acquisition (b) with 1.25 mm in-plane resolution from clinical patients with whole-heart coverage at 3T SIEMENS scanners.

High-resolution spiral first-pass perfusion video with 1.5 mm in-plane resolution from a patient using SMS acquisition (MB=3) and SMS-Slice-L1-SPIRiT reconstruction at 1.5 T SIEMENS Aera scanner.

Related publications:

Wang, J, Yang, Y, Weller, DS, et al. High spatial resolution spiral first-pass myocardial perfusion imaging with whole-heart coverage at 3 T. Magn Reson Med. 2021; 86: 648– 662. https://doi.org/10.1002/mrm.28701
Wang J, Yang Y, Zhou R, Sun C, Jacob M, Weller DS, Epstein F, Salerno M. High resolution spiral simultaneous multi-slice first-pass perfusion imaging with whole-heart coverage at 1.5 T and 3 T. In Proceedings of the ISMRM 28th Annual Meeting and Exhibition, 2020, p. 1313 (oral presentation, Summa cum Laude Awards).

DEep learning-based rapid Spiral Image REconstruction (DESIRE) for High-resolution Spiral First-pass Myocardial Perfusion Imaging

As shown above, spiral perfusion imaging techniques, using a motion-compensated compressed sensing-based L1-SPIRiT based reconstruction, are capable of whole-heart high-resolution perfusion imaging at both 1.5 T and 3 T. However, this reconstruction is performed off-line and takes ~1 hour per slice.

To address this limitation, we sought to develop a DEep learning-based rapid Spiral Image REconstruction technique (DESIRE) for high-resolution spiral first-pass myocardial perfusion imaging for both 2D and SMS acquisitions with whole-heart coverage. Utilizing the proposed 3D U-Net based denoising architecture, following pre-processing steps of coil selection, motion correction, and NUFFT, the image reconstruction time could be shortened from ~30 minutes per dynamic series to under 3 minutes with a network inference time of ~100 ms while still maintaining high image quality as compared with a current state‐of‐the‐art compressed sensing algorithm (L1-SPIRiT), making online reconstruction feasible.

The proposed technique demonstrated excellent performance on high-resolution spiral perfusion imaigng at both 1.5 T and 3 T.

A patient underwent clinical stress spiral perfusion imaging with 1.25 mm in-plane resolution using the 2D acquisition at 3 T. Images were reconstructed using the proposed DESIRE technique and L1-SPIRiT which served as the reference. The perfusion defect showed in DESIRE had good agreement with the reference. The cardiac catherization showed that the left anterior descending artery had the complete occlusion. The signal plot demonstrated that the temporal fidelity using the DESIRE had good agreement with the reference and the inputs with the preserved temporal fidelity at myocardium circled by the yellow line.

High-resolution spiral SMS perfusion imaging at 1.5 T reconstructed with SMS-Slice-L1-SPIRiT and the proposed DESIRE technique. (A) shows an example case of 9 slices of 1.5 mm resolution perfusion images at middle time frame with an SMS factor of 3 from a patient. (B) shows an example case of 8 slices with an SMS factor of 4 from a patient.

Related publications:

Wang, J, Weller, DS, Kramer, CM, Salerno, M. DEep learning-based rapid Spiral Image REconstruction (DESIRE) for high-resolution spiral first-pass myocardial perfusion imaging. NMR in Biomedicine. 2022;e4661. doi:10.1002/nbm.4661
Wang J, Weller D, Salerno M. DESIRE: DEep learning-based rapid Spiral Imaging Reconstruction for high-resolution spiral first-pass myocardial perfusion imaging with whole-heart coverage. In Proceedings of the SCMR 24th Annual Scientific Sessions, 2021 (oral presentation, ECA finalist).
Wang J, Rodriguez Lozano P, Salerno M. High-resolution Spiral First-pass Myocardial Perfusion Imaging at 1.5 T with DEep learning-based rapid Spiral Image REconstruction (DESIRE). In Proceedings of the ISMRM 30th Annual Scientific Sessions, London, England, UK, 2022 (digital poster).

DEep learning-based rapid Spiral Image REconstruction (DESIRE) for High-resolution Spiral Real-time Cine Imaging

Cardiac magnetic resonance (CMR) real-time cine imaging, which does not require breath-holding or ECG gating, is clinically useful particularly for patients with impaired breath-hold capacity and/or arrhythmias.

Spiral acquisitions, which provide high acquisition efficiency and insensitivity to motion artifacts, can require a long reconstruction time particularly for compressed-sensing or other iterative reconstruction techniques. As such they cannot provide immediate feedback to the imager.

In this work, we sought to develop high-resolution real-time cardiac cine imaging technique using rapid spiral acquisitions and deep learning-based rapid image reconstruction and quantification. We extended the DESIRE technique developed in spiral perfusion imaging to the spiral real-time imaging. The proposed reconstruction technique showed good performance on both bSSFP imaging at 1.5 T and GRE imaging at 1.5 T and 3 T.

The bSSFP real-time cine video at 1.5 T SIEMENS Aera scanner with 1.5 mm in-plane resolution from a patient reconstructed using the proposed DESIRE technique and the L1-SENSE.

The GRE real-time cine video at 1.5 T SIEMENS Aera scanner with 1.5 mm in-plane resolution from a patient reconstructed using the proposed DESIRE technique and the L1-SENSE.

Cardiac GRE cine imaging at 3 T from a healthy volunteer reconstructed using the proposed DESIRE technique and the spiral L+S served as the reference.

Related publications:

Wang, J, Weller, DS, Kramer, CM, Salerno, M. DEep learning-based rapid Spiral Image REconstruction (DESIRE) for high-resolution spiral first-pass myocardial perfusion imaging. NMR in Biomedicine. 2022;e4661. doi:10.1002/nbm.4661
Wang J, Zhou R, Wang X, Awad M, Salerno M. Free-breathing High-resolution Spiral Real-time Cardiac Cine Imaging at 1.5 T with DEep learning-based Spiral Image REconstruction (DESIRE). In Proceedings of the ISMRM 30th Annual Scientific Sessions, London, England, UK, 2022 (oral Power Pitch).
Wang J, Zhou R, Salerno M. Free-breathing High-resolution Spiral Real-time Cardiac Cine Imaging using DEep learning-based rapid Spiral Image REconstruction (DESIRE). In Proceedings of the ISMRM 29th Annual Meeting and Exhibition, 2021, p. 877 (digital poster).

High-resolution Cartesian Perfusion Imaging with Deep learning-based Rapid Image Reconstruction

As Cartesian perfusion techniques are most commonly deployed clinically, we also sought to evaluate the proposed compressed sensing (CS), SMS, and deep learning techniques for Cartesian high-resolution perfusion imaging for both 2D and SMS acquisitions with an MB factor of 2 using a 2D Poisson-Disc incoherent sampling pattern along temporal dimension.

Currently, clinical 2D Cartesian perfusion imaging using CS-based image reconstruction such as L1-SENSE provides fast and high-resolution imaging, but it cannot provide whole-heart coverage. CS-based Cartesian SMS perfusion imaging can provide whole-heart coverage but is limited by slow reconstruction, with both techniques taking ~30 minutes to reconstruct each dynamic series. Motion correction (MOCO) is essential for perfusion imaging such that better qualitative and quantitative analysis can be conducted. However, this is relatively time consuming.

These limitations impede the clinical on-line translation of high-resolution Cartesian perfusion imaging. To address this, we sought to develop a fully automatic deep learning-based (DL) respiratory motion-corrected image reconstruction technique for Cartesian 2D and SMS high-resolution perfusion imaging at 3 T with 1.6 mm in-plane resolution to provide rapid and high-quality reconstruction, advancing the clinical workflow.

Cartesian perfusion imaging from a healthy volunteer underwent 2D acquisition. The image is reconstructed using the inverse Fast Fourier transform (initial image), compressed sensing (reference) and the proposed deep learning method. Excellent image quality was demonstrated using the proposed image reconstruction network. B) shows the temporal fidelity of the signal intensities at myocardium in the middle slice.

SMS MB=2 Cartesian perfusion imaging from a healthy volunteer with 6 slices reconstructed using the proposed reconstruction network. Good image quality was demonstrated using the proposed image reconstruction network. (B) shows the temporal fidelity of the signal intensities at myocardium in one of the middle slices shown in (A).

Evaluation of motion correction (MOCO) for example cases with 2D (A) and SMS MB=2 (B) acquisitions. x-t profiles for the dashed line pointed in a middle frame demonstrate the excellent performance of the proposed technique.

Related publications:

Wang J, Salerno M. Deep learning-based Rapid Image Reconstruction for High-resolution Cartesian First-pass Myocardial Perfusion Imaging at 3 Tesla. In Proceedings of the SCMR 26th Annual Scientific Sessions, 2023 (rapid fire presentation).

Google Sites

Report abuse