OpDis-PS
A diffusion posterior sampling method for electrocardiographic imaging (ECGI) that explicitly models signal-dependent operator mismatch through a learnable module and corrects the forward operator, improving reconstruction accuracy for ill-posed inverse problems.
Timeline (last 90 days)
OpDis-PS is a diffusion posterior sampler that explicitly models the dependence between operator mismatch and the signal.
OpDis-PS captures the mapping between mismatch and the cardiac signal through a learned module and performs posterior sampling using the corrected operator.
OpDis-PS uses a fixed prior under all operator conditions, avoiding retraining the prior for different conditions.
Under two mismatch conditions—lung conductivity variation and an additionally added subcutaneous fat layer—OpDis-PS achieves better reconstruction accuracy than classical regularization methods and learned predictors.
OpDis-PS achieves the largest performance improvement on held-out cardiac pathology data.
Current OpDis-PS research is still limited to simulated data, and validation on real clinical data is the key next step.
OpDis-PS's strategy of "fixed prior, corrected operator" can lower the computational and data barriers to clinical deployment.
The OpDis-PS framework can be extended to other imaging modalities that require posterior sampling, such as EEG source localization and electrical impedance tomography, and other inverse problems with model errors.
All Facts (8)
OpDis-PS is a diffusion posterior sampler that explicitly models the dependence between operator mismatch and the signal.
50%UnverifiedOpDis-PS captures the mapping between mismatch and the cardiac signal through a learned module and performs posterior sampling using the corrected operator.
50%UnverifiedOpDis-PS uses a fixed prior under all operator conditions, avoiding retraining the prior for different conditions.
50%UnverifiedUnder two mismatch conditions—lung conductivity variation and an additionally added subcutaneous fat layer—OpDis-PS achieves better reconstruction accuracy than classical regularization methods and learned predictors.
50%UnverifiedOpDis-PS achieves the largest performance improvement on held-out cardiac pathology data.
50%UnverifiedCurrent OpDis-PS research is still limited to simulated data, and validation on real clinical data is the key next step.
50%UnverifiedOpDis-PS's strategy of "fixed prior, corrected operator" can lower the computational and data barriers to clinical deployment.
50%UnverifiedThe OpDis-PS framework can be extended to other imaging modalities that require posterior sampling, such as EEG source localization and electrical impedance tomography, and other inverse problems with model errors.
50%