eduzhai > Applied Sciences > Engineering >

Towards Cardiac Intervention Assistance Hardware-aware Neural Architecture Exploration for Real-Time 3D Cardiac Cine MRI Segmentation

  • king
  • (0) Download
  • 20210506
  • Save

... pages left unread,continue reading

Document pages: 8 pages

Abstract: Real-time cardiac magnetic resonance imaging (MRI) plays an increasinglyimportant role in guiding various cardiac interventions. In order to providebetter visual assistance, the cine MRI frames need to be segmented on-the-flyto avoid noticeable visual lag. In addition, considering reliability andpatient data privacy, the computation is preferably done on local hardware.State-of-the-art MRI segmentation methods mostly focus on accuracy only, andcan hardly be adopted for real-time application or on local hardware. In thiswork, we present the first hardware-aware multi-scale neural architecturesearch (NAS) framework for real-time 3D cardiac cine MRI segmentation. Theproposed framework incorporates a latency regularization term into the lossfunction to handle real-time constraints, with the consideration of underlyinghardware. In addition, the formulation is fully differentiable with respect tothe architecture parameters, so that stochastic gradient descent (SGD) can beused for optimization to reduce the computation cost while maintainingoptimization quality. Experimental results on ACDC MICCAI 2017 datasetdemonstrate that our hardware-aware multi-scale NAS framework can reduce thelatency by up to 3.5 times and satisfy the real-time constraints, while stillachieving competitive segmentation accuracy, compared with the state-of-the-artNAS segmentation framework.

Please select stars to rate!

         

0 comments Sign in to leave a comment.

    Data loading, please wait...
×