Enhancing Inference on Physiological and Kinematic Periodic Signals via Phase-Based Interpretability and Multi-Task Learning (original) (raw)
Related papers
Generative Models for Functional Data using Phase and Amplitude Separation
2012
Constructing generative models for functional observations is an important task in statistical functional analysis. In general, functional data contains both phase (or x or horizontal) and amplitude (or y or vertical) variability. Traditional methods often ignore the phase variability and focus solely on the amplitude variation, using cross-sectional techniques such as fPCA for dimensional reduction and data modeling. Ignoring phase variability leads to a loss of structure in the data and inefficiency in data models. This paper presents an approach that relies on separating the phase (x-axis) and amplitude (y-axis), then modeling these components using joint distributions. This separation, in turn, is performed using a technique called elastic shape analysis of curves that involves a new mathematical representation of functional data. Then, using individual fPCAs, one each for phase and amplitude components, while respecting the nonlinear geometry of the phase representation space; impose joint probability models on principal coefficients of these components. These ideas are demonstrated using random sampling, for models estimated from simulated and real datasets, and show their superiority over models that ignore phase-amplitude separation. Furthermore, the generative models are applied to classification of functional data and achieve high performance in applications involving SONAR signals of underwater objects, handwritten signatures, and periodic body movements recorded by smart phones.
Combining Data-Driven and Knowledge-Guided Methods to Induce Interpretable Physiological Models
2011
In this paper, we review the paradigm of inductive process modeling and examine its application to human physiology. This framework represents models as a set of interacting processes, each with associated differential or algebraic equations that express causal relations among variables. Simulating such a quantitative process model produces trajectories for variables over time that one can compare to observations. Background knowledge about candidate processes enables search through the space of model structures and their associated parameters, and thus identify quantitative models that explain time-series data. We present an initial process model for aspects of human physiology, consider its uses for health monitoring, and discuss the induction of such models. In closing, we consider related efforts on physiological modeling and our plans for collecting data to evaluate our framework in this domain.
Real time reconstruction of quasiperiodic multi parameter physiological signals
EURASIP Journal on Advances in Signal Processing, 2012
A modern intensive care unit (ICU) has automated analysis systems that depend on continuous uninterrupted real time monitoring of physiological signals such as electrocardiogram (ECG), arterial blood pressure (ABP), and photo-plethysmogram (PPG). These signals are often corrupted by noise, artifacts, and missing data. We present an automated learning framework for real time reconstruction of corrupted multi-parameter nonstationary quasiperiodic physiological signals. The key idea is to learn a patient-specific model of the relationships between signals, and then reconstruct corrupted segments using the information available in correlated signals. We evaluated our method on MIT-BIH arrhythmia data, a two-channel ECG dataset with many clinically significant arrhythmias, and on the CinC challenge 2010 data, a multi-parameter dataset containing ECG, ABP, and PPG. For each, we evaluated both the residual distance between the original signals and the reconstructed signals, and the performance of a heartbeat classifier on a reconstructed ECG signal. At an SNR of 0 dB, the average residual distance on the CinC data was roughly 3% of the energy in the signal, and on the arrhythmia database it was roughly 16%. The difference is attributable to the large amount of diversity in the arrhythmia database. Remarkably, despite the relatively high residual difference, the classification accuracy on the arrhythmia database was still 98%, indicating that our method restored the physiologically important aspects of the signal.
CLOPS: Continual Learning of Physiological Signals
ArXiv, 2020
Deep learning algorithms are known to experience destructive interference when instances violate the assumption of being independent and identically distributed (i.i.d). This violation, however, is ubiquitous in clinical settings where data are streamed temporally and from a multitude of physiological sensors. To overcome this obstacle, we propose CLOPS, a healthcare-specific replay-based continual learning strategy. In three continual learning scenarios based on three publically-available datasets, we show that CLOPS can outperform its multi-task learning counterpart. Moreover, we propose end-to-end trainable parameters, which we term task-instance parameters, that can be used to quantify task difficulty and similarity. This quantification yields insights into both network interpretability and clinical applications, where task difficulty is poorly quantified.
SimPer: Simple Self-Supervised Learning of Periodic Targets
Cornell University - arXiv, 2022
From human physiology to environmental evolution, important processes in nature often exhibit meaningful and strong periodic or quasi-periodic changes. Due to their inherent label scarcity, learning useful representations for periodic tasks with limited or no supervision is of great benefit. Yet, existing self-supervised learning (SSL) methods overlook the intrinsic periodicity in data, and fail to learn representations that capture periodic or frequency attributes. In this paper, we present SimPer, a simple contrastive SSL regime for learning periodic information in data. To exploit the periodic inductive bias, SimPer introduces customized augmentations, feature similarity measures, and a generalized contrastive loss for learning efficient and robust periodic representations. Extensive experiments on common real-world tasks in human behavior analysis, environmental sensing, and healthcare domains verify the superior performance of SimPer compared to state-of-the-art SSL methods, highlighting its intriguing properties including better data efficiency, robustness to spurious correlations, and generalization to distribution shifts.
2021 43rd Annual International Conference of the IEEE Engineering in Medicine & Biology Society (EMBC), 2021
Respiratory rate (RR) is a clinical sign representing ventilation. An abnormal change in RR is often the first sign of health deterioration as the body attempts to maintain oxygen delivery to its tissues. There has been a growing interest in remotely monitoring of RR in everyday settings which has made photoplethysmography (PPG) monitoring wearable devices an attractive choice. PPG signals are useful sources for RR extraction due to the presence of respiration-induced modulations in them. The existing PPG-based RR estimation methods mainly rely on hand-crafted rules and manual parameters tuning. An end-to-end deep learning approach was recently proposed, however, despite its automatic nature, the performance of this method is not ideal using the real world data. In this paper, we present an end-to-end and accurate pipeline for RR estimation using Cycle Generative Adversarial Networks (CycleGAN) to reconstruct respiratory signals from raw PPG signals. Our results demonstrate a higher RR estimation accuracy of up to 2× (mean absolute error of 1.9±0.3 using five fold cross validation) compared to the state-of-th-art using a identical publicly available dataset. Our results suggest that CycleGAN can be a valuable method for RR estimation from raw PPG signals.
From Frequency Content to Signal Dynamics Using DNNs
IEEE ACCESS, 2022
This study developed a novel method for analyzing and decomposing a signal into its main dynamics for small and large timescales. Our proposal is based on a decoupled hybrid system of convolutional and recurrent neural networks that uses as inputs the power spectrum and spectrogram of a given signal, giving as output the dynamic behavior. We define the dynamic classification predicted of the signal using previously known dynamics characterized through training signals: periodic, quasi-periodic, aperiodic, chaotic, and randomness. We created a synthetic dataset comprising more than 50 training signals from different categories. For the real-world dataset, we used photoplethysmographic signals from 40 students obtained from a Spanish medical study. We tested the developed system's performance in real biological and synthetical signals, obtaining noteworthy results. All the results are evaluated qualitatively and quantitatively. Still, given the novelty and the lack of similar works, we cannot compare reliably and rigorously our results with other works, at least quantitatively. We can retrieve from the exposed results in this work three key ideas: the DNN-based solutions are capable of learning and generalizing the dynamics behavior of signals; the proposal learned correctly to distinguish between the reference dynamics provided and find some unidirectional similarities in the aperiodicity cases; and the results obtained using real-world PPG signals reveal that biological signals seem to exhibit a multi-dynamic behavior that changes depending on the used timescale, being quasi-periodically dominant in the short-term and aperiodically dominant in the long-term.
Inference of Irregular Cardiac Activity Using Neural Temporal Perturbation Fields
Khulna University Studies
Early diagnosis of irregular cardiac activity through existing tools such as Electrocardiogram and greater understanding of the underlying processes is critical for saving lives. Cardiac activity originates from a deterministic dynamical system of heart with trajectories following a linear map. Irregular cardiac activity observed in arrhythmia patients adds nonlinearities to the evolution function of the dynamical system underneath. Therefore, it is of great importance to quantitatively measure this non-linearity as a biomarker for impending cardiac diseases in patients. In this work, we formulated a novel mechanism named Neural Temporal Perturbation Field where perceived nonlinearities are modeled through deep neural network with perturbated inputs. Here, we examined the nonlinear state space by modeling the volatility of outputs for slightly adjusted inputs. We discovered that volatility characteristics clearly define a decision threshold that may be employed as a biomarker in cli...
Deep-Learning-Driven Techniques for Real-Time Multimodal Health and Physical Data Synthesis
Electronics
With the advent of Artificial Intelligence for healthcare, data synthesis methods present crucial benefits in facilitating the fast development of AI models while protecting data subjects and bypassing the need to engage with the complexity of data sharing and processing agreements. Existing technologies focus on synthesising real-time physiological and physical records based on regular time intervals. Real health data are, however, characterised by irregularities and multimodal variables that are still hard to reproduce, preserving the correlation across time and different dimensions. This paper presents two novel techniques for synthetic data generation of real-time multimodal electronic health and physical records, (a) the Temporally Correlated Multimodal Generative Adversarial Network and (b) the Document Sequence Generator. The paper illustrates the need and use of these techniques through a real use case, the H2020 GATEKEEPER project of AI for healthcare. Furthermore, the pape...
Towards Uncovering Feature Extraction From Temporal Signals in Deep CNN: the ECG Case Study
2020 International Joint Conference on Neural Networks (IJCNN), 2020
Despite all the progress made in biomedical field, the Electrocardiogram (ECG) is still one of the most commonly used signal in medical examinations. Over the years, the problem of ECG classification has been approached in many different ways, most of which rely on the extraction of features from the signal in the form of temporal or morphological characteristics. Although feature engineering can led to adequately good results, it mostly relies on human ability and experience in selecting the correct feature set. In the last decade, a growing class of techniques based on Convolutional Neural Network (CNN) has been proposed in opposition to feature engineering. The efficiency and accuracy of CNN-based approaches is indisputable, however their ability in extracting and using temporal features from raw signal is poorly understood. The main objective of this work was to uncover the differences and the relationships between CNN feature maps and human-curated temporal features, towards a ...