Heartbeat Classification Using Batch-Weighted Loss CNN
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional methods for classifying heartbeats using single-lead portable ECG devices face challenges due to dataset imbalance and computational expense, leading to low classification accuracy, as they heavily rely on data preprocessing and integration of synthetic data which can result in biased outcomes and increased training time.
Innovation Solution
A method and apparatus that address dataset imbalance by determining a loss function based on batch-level loss weights, training a learning model using this approach, and employing a Convolutional Neural Network (CNN) with batch normalization and dropout layers to classify heartbeats without additional data preprocessing, enabling high classification performance with single-lead ECG data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional methods use data preprocessing and synthetic data integration to handle dataset imbalance, then classification accuracy may improve, but training time increases and computational cost rises
Solution Approach 1:
The patent changes the parameter of loss function weighting by introducing batch-level dynamic loss weights that adapt to the actual distribution of heartbeat classes in each batch. This replaces the need for synthetic data generation and complex preprocessing with a parameter adjustment in the loss function, achieving balanced classification without increasing training time
Solution Approach 2:
The patent extracts and removes the unnecessary data preprocessing steps (noise removal, feature extraction) and synthetic data integration from the conventional pipeline. By using raw ECG waveforms directly as input to the CNN model, it eliminates the time-consuming preprocessing stage while maintaining classification accuracy through the batch-weighted loss function
2Measurement precision
If conventional methods rely on data preprocessing such as noise removal and feature extraction, then classification performance may improve, but computational expense increases
Solution Approach 1:
The patent extracts and removes the computationally expensive data preprocessing steps (noise removal, feature extraction) from the classification pipeline. By feeding raw ECG waveforms directly into the CNN model, it eliminates unnecessary computational operations while maintaining high classification accuracy through the batch-weighted loss function that handles class imbalance at the training objective level
Solution Approach 2:
The patent replaces the mechanical preprocessing pipeline (filtering, feature extraction) with a data-driven approach where the CNN model learns directly from raw waveforms. The batch-weighted loss function substitutes for the need for manual feature engineering, allowing the model to automatically learn discriminative features while handling class imbalance
3Measurement precision
If synthetic data is generated to balance dataset classes, then classification accuracy for minority classes may improve, but training time and computational resources increase
Solution Approach 1:
Instead of changing the data distribution through synthetic data generation, the patent changes the loss function parameters by introducing dynamic batch weights. This parameter adjustment in the training objective directly addresses class imbalance without requiring additional data generation, thus avoiding increased training time and computational resources
Solution Approach 2:
The patent inverts the conventional approach by not trying to balance the data distribution, but instead balancing the loss contribution of each class through weight adjustment. Rather than generating more minority class samples, it amplifies the learning signal from minority classes through the batch-weighted loss function, achieving the same effect more efficiently
Data Source
AI summary
A method for classifying types of heartbeats includes obtaining a dataset including multiple heartbeat waveform data, generating, from the dataset, input data regarding a heartbeat waveform for training and generating a learning model to which the generated input data is input and from which a heartbeat type of the heartbeat waveform for training is output, training the learning model by determining a loss weight of each batch sampled from the dataset and determining a loss function based on the loss weight of each batch, and inputting a heartbeat waveform for test to the learning model and classifying a heartbeat type of the heartbeat waveform for test.