Impact of Label Noise on the Learning Based Models for a Binary Classification of Physiological Signal

Abstract

Label noise is omnipresent in the annotations process and has an impact on supervised learning algorithms. This work focuses on the impact of label noise on the performance of learning models by examining the effect of random and class-dependent label noise on a binary classification task: quality assessment for photoplethysmography (PPG). PPG signal is used to detect physiological changes and its quality can have a significant impact on the subsequent tasks, which makes PPG quality assessment a particularly good target for examining the impact of label noise in the field of biomedicine. Random and class-dependent label noise was introduced separately into the training set to emulate the errors associated with fatigue and bias in labeling data samples. We also tested different representations of the PPG, including features defined by domain experts, 1D raw signal and 2D image. Three different classifiers are tested on the noisy training data, including support vector machine (SVM), XGBoost, 1D Resnet and 2D Resnet, which handle three representations, respectively. The results showed that the two deep learning models were more robust than the two traditional machine learning models for both the random and class-dependent label noise. From the representation perspective, the 2D image shows better robustness compared to the 1D raw signal. The logits from three classifiers are also analyzed, the predicted probabilities intend to be more dispersed when more label noise is introduced. From this work, we investigated various factors related to label noise, including representations, label noise type, and data imbalance, which can be a good guidebook for designing more robust methods for label noise in future work.

Cluster consistency: Simple yet effect robust learning algorithm on large-scale photoplethysmography for atrial fibrillation detection in the presence of real-world label noise

“…Although false alarms exist, we can still observe accurate predictions on external test sets even with conventional CE loss. This phenomenon shows deep learning is robust to label errors to certain level, which is same as we concluded in the previous work [10]. The accurate predictions do show great potential of using similar way to generate large-scale dataset with no human expert involved.…”

“…The inaccurate alarm information would cause label errors when it is used to auto-annotate the PPG signals. Although our previous work [10] concluded that deep learning is more robust to label noise than conventional models, the need for robust learning from noisy labels is still important.…”

Machine Learning Approaches of cotact PPG and its Potential Research Challanges

“…One of the main advantages of using Machine Learning in plethysmography is its ability to handle large and complex datasets [3]. By analyzing a large amount of data, Machine Learning algorithms can improve the accuracy and reliability of diagnostic tests based on plethysmography.…”

Plethysmography Signal Assessment using Deep Learning Algorithms

“…One of the main advantages of using deep learning in plethysmography is its ability to handle large and complex datasets [3]. By analyzing a large amount of data, deep learning algorithms can improve the accuracy and reliability of diagnostic tests based on plethysmography.…”

“…Plethysmography is commonly used in research and clinical settings to diagnose and monitor various medical conditions, including respiratory, cardiovascular, and vascular disorders [3]. It is a non-invasive and relatively simple technique that can provide valuable information about the function of different body parts [4].…”