Aug 3, 2021

UII America Research Results Accepted at Premier Medical Imaging Conferences in 2021

Congratulations to our US team for having their research accepted at the International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI) this year. MICCAI is the top conference in the field of medical image artificial intelligence, with an acceptance rate of about 30%. Congratulations to our coauthors.

In the work “Multi-scale Neural ODEs for 3D Medical Image Registration“, UII America team proposed a new deep learning framework to solve the image fusion (image registration) problem across different image modalities. The technology can be utilized in image-guided surgery, image-based diagnosis, radiation therapy, motion estimation, and more. This novel method will learn a registration optimizer via a multi-scale neural ODE model optimizing the registration inference in a much faster way compared to prior methods.

Figure 1. Overview of proposed image registration method.

In addition, three abstract proposals from UII America have been accepted by the International Society for Magnetic Resonance in Medicine (ISMRM), also this year. In “Cardiac Functional Analysis with Cine MRI via Deep Learning Reconstruction“, our scientists proposed a new deep learning reconstruction method for under-sampled 3D MULTIPLEX MRI data reconstruction, removing the artifacts and keeping fidelity with the true measurement. In “Fully Automated Myocardium Strain Analysis using Deep Learning“, a deep-learning-based fully automated myocardium strain assessment system that requires zero human intervention was proposed for accurate strain assessment, which can simplify cardiologists’ work and thus reduces their workload. In “Accelerating 3D MULTIPLEX MRI Reconstruction with Deep Learning“, we evaluated and compared the cardiac functional values (EDV, ESV, and EF for LV and RV, respectively) obtained from highly accelerated MRI acquisition using DL based reconstruction algorithm (DL-cine) with values from compressed sensing (CS-cine) and conventional retro-cine. The result shows that the utilization of the DL-Cine will significantly (e.g. 12x) reduces the data acquisition time compared to Retro-Cine, which can benefit, during MRI acquisition, patients who cannot hold their breath or have arrhythmia problems.