Label noise and self-learning label correction in cardiac abnormalities classification

Abstract

Objective. Learning to classify cardiac abnormalities requires large and high-quality labeled datasets, which is a challenge in medical applications. Small datasets from various sources are often aggregated to meet this requirement, resulting in a final dataset prone to label noise owing to inter- and intra-observer variability, and different expertise. It is well known that label noise can affect the performance and generalizability of the trained models. In this work, we explore the impact of label noise and self-learning label correction on the classification of cardiac abnormalities on large heterogeneous datasets of electrocardiogram (ECG) signals. Approach. A state-of-the-art self-learning multi-class label correction method for image classification is adapted to learn a multi-label classifier for electrocardiogram signals. We evaluated our performance using 5-fold cross-validation on the publicly available PhysioNet/Computing in Cardiology (CinC) 2021 Challenge data, with full and reduced sets of leads. Due to the unknown label noise in the testing set, we tested our approach on the MNIST dataset. We investigated the performance under different levels of structured label noise for both datasets. Main results. Under high levels of noise, the cross-validation results of self-learning label correction showed an improvement of approximately 3% in the Challenge score for the PhysioNet/CinC 2021 Challenge dataset and, an improvement in accuracy of 5$\%$ and reduction of the expected calibration error of 0.03 for the MNIST dataset. We demonstrate that self-learning label correction can be used to effectively deal with the presence of unknown label noise, also when using a reduced number of ECG leads.

A comprehensive review on efficient artificial intelligence models for classification of abnormal cardiac rhythms using electrocardiograms

“…A few shot learning approach using siamese convolutional networks (SCNN) [148] was recently proposed; this method achieved an accuracy of up to 95% on the publically available INCART 12-lead Arrhythmia dataset [149] . Vázquez et al adapted a self-learning multi-class label correction method to learn a multi-label classifier for ECG signals and evaluated the model using 5-fold cross-validation [150] . They successfully demonstrated that self-learning label correction can effectively address unknown label noise and improve classification accuracy even with the reduced number of ECG leads.…”

“… [148] (2023) Siamese-CNN INCART 12-lead Arrhythmia dataset A few shot learning approach using siamese convolutional networks for Arrhythmia classification. Contrastive Loss was used for training the model which gave best performance, as the loss function [150] (2022) 1D-CNN PhysioNet/CinC 2021 A self-learning multi-class label correction method to learn a multi-label classifier in presence of noisy labels. …”

1D-Convolutional Neural Networks can Quantify Therapy Content of Children and Adolescents Walking in a Robot-Assisted Gait Trainer

“…A 1D-CNN similar to [16] was used for classification. The model takes the sEMG pattern of a single stride, consisting of the M. semitendinosus and the M. rectus femoris, as input and the predicted instruction as output.…”

“…This could have lowered the accuracy. To eliminate the effect of this label noise on the classifier, label correction during training could be applied [16].…”