Selected work

Representative Papers

Physics-Informed autoencoder for DSC-MRI Perfusion post-processing: application to glioma grading

Oct 13, 2025

Noise and motion artifacts in DSC-MRI perfusion imaging corrupt the deconvolution process, leading to biased cerebral blood flow (CBF) parameter estimation; moreover, existing deep learning methods rely on biased third-party deconvolution results as supervision. To address this, we propose a physics-guided autoencoder that embeds an analytical perfusion model—specifically, the singular value decomposition (SVD)-based residue function—into the decoder, enabling end-to-end, fully self-supervised training without external annotations. Our method directly reconstructs physiologically consistent perfusion parameters from raw time-signal curves. In glioma grading, it achieves performance comparable to state-of-the-art deconvolution algorithms (AUC ≥ 0.89), reduces computational cost by over 3×, and exhibits markedly improved robustness under high noise. The key innovation lies in the first deep integration of a differentiable biophysical model into an autoencoding architecture, thereby eliminating dependence on conventional deconvolution priors.

0 citationsRead paper

Improving Virtual Contrast Enhancement using Longitudinal Data

Sep 30, 2025

To address the risk of gadolinium deposition in the brain associated with repeated administration of gadolinium-based contrast agents (GBCAs), this study proposes a deep learning–based virtual contrast enhancement method that leverages longitudinal temporal information to reconstruct full-dose contrast appearance from low-dose GBCA scans. Methodologically, it introduces, for the first time, the use of a patient’s prior full-dose T1-weighted MRI as a longitudinal prior, integrated with the current low-dose acquisition to formulate a multi-stage, cross-temporal reconstruction model. Compared to single-session models, our approach achieves significant improvements in quantitative metrics—including PSNR and SSIM—and demonstrates strong robustness under simulated dose reductions of 1/4 to 1/2. The key contribution lies in longitudinal prior–driven spatiotemporal consistency modeling, which markedly enhances reconstruction stability and lesion conspicuity. This framework establishes a safer, more reliable low-dose MRI diagnostic paradigm for chronic neurological conditions requiring long-term surveillance, such as multiple sclerosis and glioma.

0 citationsRead paper

Recent publications

Latest Papers

Physics-Informed autoencoder for DSC-MRI Perfusion post-processing: application to glioma grading

Oct 13, 2025

Noise and motion artifacts in DSC-MRI perfusion imaging corrupt the deconvolution process, leading to biased cerebral blood flow (CBF) parameter estimation; moreover, existing deep learning methods rely on biased third-party deconvolution results as supervision. To address this, we propose a physics-guided autoencoder that embeds an analytical perfusion model—specifically, the singular value decomposition (SVD)-based residue function—into the decoder, enabling end-to-end, fully self-supervised training without external annotations. Our method directly reconstructs physiologically consistent perfusion parameters from raw time-signal curves. In glioma grading, it achieves performance comparable to state-of-the-art deconvolution algorithms (AUC ≥ 0.89), reduces computational cost by over 3×, and exhibits markedly improved robustness under high noise. The key innovation lies in the first deep integration of a differentiable biophysical model into an autoencoding architecture, thereby eliminating dependence on conventional deconvolution priors.

0 citationsRead paper

Improving Virtual Contrast Enhancement using Longitudinal Data

Sep 30, 2025

To address the risk of gadolinium deposition in the brain associated with repeated administration of gadolinium-based contrast agents (GBCAs), this study proposes a deep learning–based virtual contrast enhancement method that leverages longitudinal temporal information to reconstruct full-dose contrast appearance from low-dose GBCA scans. Methodologically, it introduces, for the first time, the use of a patient’s prior full-dose T1-weighted MRI as a longitudinal prior, integrated with the current low-dose acquisition to formulate a multi-stage, cross-temporal reconstruction model. Compared to single-session models, our approach achieves significant improvements in quantitative metrics—including PSNR and SSIM—and demonstrates strong robustness under simulated dose reductions of 1/4 to 1/2. The key contribution lies in longitudinal prior–driven spatiotemporal consistency modeling, which markedly enhances reconstruction stability and lesion conspicuity. This framework establishes a safer, more reliable low-dose MRI diagnostic paradigm for chronic neurological conditions requiring long-term surveillance, such as multiple sclerosis and glioma.

0 citationsRead paper