FM-Recon
An unrolled reconstruction network framework that uses frozen features from a pretrained foundation model to enhance accelerated cardiac cine MRI reconstruction, injecting foundation model features into every layer of the network through feature extraction, refinement and fusion modules.
Timeline (last 90 days)
The paper "FM-Recon: Exploring Foundation Model Features for Accelerated Cardiac Cine MRI Reconstruction" systematically explores integrating pretrained foundation model features into accelerated cardiac cine MRI reconstruction.
The FM-Recon framework uses a frozen foundation model image encoder to extract multi-scale features, which after adjustment by a trainable refinement network are injected into each layer of the unrolled reconstruction network.
In FM-Recon the foundation model parameters remain frozen, and only the lightweight feature refinement and fusion modules are trained, to avoid catastrophic forgetting and reduce computational overhead.
The study conducted extensive experiments on the CMRxRecon2023 dataset, systematically evaluating the impact of foundation model features under different reconstruction settings.
At moderate acceleration factors, FM-Recon achieved better reconstruction quality than baseline methods without foundation model features; at extremely high acceleration factors, the improvement was limited.
Injecting features into every layer of the unrolled network is more effective than using features only at the input or output, indicating that continuous guidance during the iterative process is crucial.
FM-Recon's limitations include: experiments were conducted only on cardiac cine MRI, with other anatomical sites or modalities not yet validated; the choice of foundation model is relatively limited; and the feature fusion approach can be further optimized.
All Facts (7)
The paper "FM-Recon: Exploring Foundation Model Features for Accelerated Cardiac Cine MRI Reconstruction" systematically explores integrating pretrained foundation model features into accelerated cardiac cine MRI reconstruction.
50%UnverifiedThe FM-Recon framework uses a frozen foundation model image encoder to extract multi-scale features, which after adjustment by a trainable refinement network are injected into each layer of the unrolled reconstruction network.
50%UnverifiedIn FM-Recon the foundation model parameters remain frozen, and only the lightweight feature refinement and fusion modules are trained, to avoid catastrophic forgetting and reduce computational overhead.
50%UnverifiedThe study conducted extensive experiments on the CMRxRecon2023 dataset, systematically evaluating the impact of foundation model features under different reconstruction settings.
50%UnverifiedAt moderate acceleration factors, FM-Recon achieved better reconstruction quality than baseline methods without foundation model features; at extremely high acceleration factors, the improvement was limited.
50%UnverifiedInjecting features into every layer of the unrolled network is more effective than using features only at the input or output, indicating that continuous guidance during the iterative process is crucial.
50%UnverifiedFM-Recon's limitations include: experiments were conducted only on cardiac cine MRI, with other anatomical sites or modalities not yet validated; the choice of foundation model is relatively limited; and the feature fusion approach can be further optimized.
50%