ECG Model Pretraining With Masked Temporal Patches
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current self-supervised machine learning processes for electrocardiogram data require extensive image transformation and processing, leading to inefficiencies due to high memory footprint and processing power demands, and often necessitate converting ECG signals into images, further reducing efficiency.
Innovation Solution
A system and method for training machine learning models using unlabeled electrocardiogram signals by creating overlapping temporal patches from ECG data, masking these patches, calculating reconstruction errors, and adjusting model parameters to minimize errors, without converting ECG signals into images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If ECG signals are converted into images for self-supervised machine learning processing, then the model can utilize image transformation techniques, but the memory footprint and processing power requirements increase significantly
Solution Approach 1:
The patent replaces the mechanical/image-based processing system with a direct signal-processing system. Instead of converting ECG signals to images and applying image transformation techniques, the system processes ECG signals directly in their native temporal format, substituting image-based mechanical operations with signal-based computational operations that are more efficient for this data type
Solution Approach 2:
The patent changes the fundamental parameter representation from spatial/image-based to temporal/signal-based. By working directly with temporal patches of ECG signals rather than converting them to image format, the system alters the processing parameters to match the inherent nature of ECG data, reducing unnecessary transformation overhead
2Adaptability or versatility
If image transformation techniques are used for self-supervised machine learning on ECG data, then the processing can leverage existing image-based models, but the memory footprint increases
Solution Approach 1:
The patent substitutes image-based processing mechanisms with direct signal-processing mechanisms. By eliminating the image conversion step and working directly with ECG signal patches, the system reduces the memory footprint associated with storing and processing image representations while maintaining model versatility through adapter layers
3Adaptability or versatility
If ECG signals are converted to images and processed through intensive image transformation, then self-supervised learning can be applied, but the overall processing efficiency decreases
Solution Approach 1:
The patent replaces the inefficient image-based processing system with a direct signal-processing system. By substituting image conversion and transformation operations with direct temporal patch processing, the system eliminates unnecessary computational steps while preserving self-supervised learning capabilities through contrastive learning on signal patches
Solution Approach 2:
The patent maintains continuous useful action by processing ECG signals in their native temporal format without interruption for format conversion. The system continuously processes temporal patches directly, maintaining the natural temporal continuity of ECG signals throughout the processing pipeline, which improves processing efficiency
Data Source
AI summary
A system for training machine learning models with unlabeled electrocardiogram signals, the system including a memory containing instructions configurating a processor to receive a plurality of electrocardiogram (ECG) data in a textual format, create one or more overlapping temporal patches from the plurality of ECG data, mask at least one temporal patch from the one or more overlapping temporal patches, pretrain an ECG machine learning model to predict the at least one masked temporal patch from the one or more overlapping temporal patches, adjust one or more parameter values of the ECG machine learning model as a function of the at least one predicted masked temporal patch and the at least one masked temporal patch and train the ECG machine learning model as a function of the one or more parameter values and a labeled set of ECG training data.