Heartbeat Classification Model Using Attention Masking for Arrhythmia Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Automatic classification of electrocardiograms (ECGs) is challenging due to varying morphological characteristics and class imbalanced data, leading to biased classification algorithms that often misclassify minority class samples, which are crucial in scenarios like arrhythmia detection.
Innovation Solution
A heartbeat classification method using a computing device that generates an attention mask based on an assistant feature from the electrocardiogram signal, allowing for improved feature extraction and classification by masking feature maps, and a training method that resamples majority class samples to balance data distribution, enhancing the classification of minority class samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning-based heartbeat classification methods are used to achieve better generalization function, then classification accuracy for normal heartbeats is improved, but the model becomes biased towards majority class and fails to correctly classify minority class samples
Solution Approach 1:
The patent applies data augmentation techniques to generate synthetic minority class samples before training the deep learning model. This preliminary action balances the class distribution in the training data, preventing the model from becoming biased towards majority class and improving minority class detection reliability from the outset
Solution Approach 2:
The patent modifies the training process by implementing focal loss function and adjusting sampling strategies to change the weight parameters assigned to different classes. This parameter change allows the model to pay more attention to minority class samples during training, improving their classification accuracy without sacrificing overall performance
2Productivity
If classification algorithms are trained with imbalanced data, then training efficiency is improved due to sufficient representation of majority class, but classification accuracy for minority class samples deteriorates
Solution Approach 1:
The patent performs data augmentation and resampling of minority class samples before the training process begins. This preliminary action ensures that the training dataset has balanced class distribution, allowing the model to learn from sufficient examples of both majority and minority classes without compromising training efficiency
Solution Approach 2:
The patent introduces an intermediary data processing layer that includes techniques such as SMOTE (Synthetic Minority Over-sampling Technique) and adaptive sampling. This intermediary layer transforms the imbalanced dataset into a balanced training set, mediating between the original data distribution and the model's learning requirements
Data Source
AI summary
A computing device inputs a sample generated from an electrocardiogram signal to a heartbeat classification model, generates a feature map from the sample through multiple first layers of the heartbeat classification model, generates an attention mask based on an assistant feature generated from the feature map and the sample, generates a masked feature map by masking the feature map with the attention mask; and performs classification of the sample from the masked feature map through a second layer of the heartbeat classification model.