Generative Model Compressive Sensing for Biomedical Signal Recovery
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Continuous cardiac monitoring in free living conditions is challenging due to intermittent connections between wearable sensors and smartphones, leading to monitoring interruptions, especially when the smartphone is not in close proximity, and existing compression techniques either require complex recovery methods or do not provide sufficient sensing reduction.
Innovation Solution
The implementation of generative model-based compressive sensing (GenCS) that sparsifies biological signals by removing morphology parameters, allowing for efficient sampling and transmission of temporal parameters, enabling diagnostically equivalent signal reconstruction using a homotopy recovery algorithm and generative modeling.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If compressive sensing is used to reduce sampling rate, then sensing reduction is achieved, but recovery method complexity increases
Solution Approach 1:
The patent applies a transformation to sparsify the biological signal before sampling, removing morphology parameters and leaving only temporal parameters. This preliminary sparsification step enables compressive sensing to work more effectively with fewer samples while simplifying the recovery process, as the signal structure is optimized beforehand.
Solution Approach 2:
The patent extracts and removes morphology parameters from the biological signal, keeping only temporal parameters for compression and transmission. This extraction separates the signal into essential temporal information that can be efficiently compressed while discarding redundant morphological details.
2Loss of substance
If GeMREM is used to reduce communication, then communication reduction is achieved, but sensing reduction is not provided
Solution Approach 1:
The patent merges the advantages of both CS and GeMREM by combining compressive sensing with generative modeling. The system uses a transformation to sparsify the signal (like GeMREM) and then applies compressive sensing to further reduce sampling requirements, achieving both sensing reduction and communication reduction simultaneously.
3Reliability
If smartwatch is used as intermediate storage, then monitoring continuity is improved, but computation and storage constraints increase
Solution Approach 1:
The patent extracts only the essential temporal parameters from the biological signal after sparsification, rather than storing or processing complete signals. This extraction of minimal necessary information reduces computation and storage requirements on the smartwatch while maintaining monitoring continuity.
4Measurement precision
If Nyquist rate sampling is used, then signal accuracy is maintained, but data processing time increases
Solution Approach 1:
The patent applies a transformation to sparsify the signal before sampling, which creates a compressed representation that retains essential information. This preliminary sparsification allows accurate recovery with fewer samples, reducing data processing time while maintaining signal accuracy.
Data Source
AI summary
Methods and systems are described for sensing and recovery of a biological signal using generative-model-based compressive sensing. A transformation is applied to sparsify the quasi-periodic signal removing morphology parameters and leaving temporal parameters. The sparsified signal is sampled and the sampled signal data is transmitted to a base station. A homotopy recovery algorithm is applied to the received sampled signal data by the base station to recover the temporal parameters of the biological signal. Generative modelling is applied using previously captured morphology parameters to generate a reconstructed signal. Finally, the reconstructed signal is adjusted and scaled based on the recovered temporal parameters to provide a reconstructed signal that is diagnostically equivalent to the original biological signal.