Exploring unsupervised multivariate time series representation learning for chronic disease diagnosis

Abstract

The application of various sensors in hospitals has enabled the widespread utilization of multivariate time series signals for chronic disease diagnosis in the data-driven world. The key challenge is how to model the complex temporal (linear and nonlinear) correlations among multiple longitudinal variables. Due to scarcity of labels in practice, unsupervised learning methods have already become indispensable. However, state-of-the-art approaches mainly focus on the extraction of linear correlation-induced feature connectivity network, e.g., Pearson correlation, partial correlation, etc. To this end, for chronic disease (e.g., Parkinson disease) diagnosis, an unsupervised representation learning method is first designed to obtain informative correlation-aware signals from multivariate time series data. At the core is a contrastive learning framework with a graph neural network (GNN) encoder to capture the inter-correlation and intra-correlation of multiple longitudinal variables. Then, the previously learned representations are sent to a simple fully connected neural network, which can be trained using fewer labels compared with end-to-end complex supervised learning models. Further, to assist the decision-making process in the high-stake chronic disease detection task, model uncertainty quantification is enabled according to evidential theory. The experimental results on two public Parkinson's disease data sets show the expressiveness of the learned embeddings, and the final lightweight classifier achieves the best performance. IntroductionWith the development of science and technology, people's living standards have been greatly improved. However, chronic diseases such as heart disease, cancer, and diabetes are still the leading causes of death and disability in many

Recent advances in domain-driven data mining

“…Another recent progress is the improved ability of the generative models that learn not to score or classify but to create rich outputs such as images, texts, or audio. We also continue seeing more expansion in the field of graph neural network, where models learn and reproduce attributes of a graph data structure [52].…”

“…Deep representation learning has attracted much attention in recent years. For chronic disease diagnosis, Zhang et al [52] designed an unsupervised representation learning method to obtain informative correlation-aware signals from multivariate time series data. The key idea was a contrastive learning framework with a Graph Neural Network (GNN) encoder to capture inter-and intra-correlation of multiple longitudinal variables.…”

“…Another contributor to this proliferation is the increase in the quantity of data collected, stored, and appropriately documented for mining since the benefits of leveraging this data has become more apparent. Some of the works in this special issue demonstrated how data mining techniques can be applied in agriculture [2], healthcare and medicine [22,52], and city planning [19].…”

“…These models are often trained with self-supervision on large data and contain millions or billions of learned parameters, such as models for language processing (e.g., BERT, GPT-3, and XLNet) and image classification (ResNet, EfficientNet, Inception). A fundamental property of many generalized models is their ability to encode the input data into a vectorized representation, as evidenced by Zhang et al [52].…”

Machine Learning Classifiers to Evaluate Data From Gait Analysis With Depth Cameras in Patients With Parkinson’s Disease

“…Machine learning (ML) techniques have been studied in several medical areas including PD (Sidey-Gibbons and Sidey-Gibbons, 2019 ) in order to classify healthy volunteers from patients using voice analysis (Ozkan, 2016 ), feet pressure systems (Abdulhay et al, 2018 ), RGB-D cameras (Buongiorno et al, 2019 ; Jaggy Castaño-Pino et al, 2019 ), optoelectronic motion analysis system (Varrecchia et al, 2021 ), wearable sensors such as accelerometers or inertial measurement units (IMU; Yoneyama et al, 2013 ; Caramia et al, 2018 ), walkway pressure analysis (Wahid et al, 2015 ), and variables associated with knee and trunk rotation (Varrecchia et al, 2021 ). Other studies have been using unsupervised learning to extract features in the initial stages of the disease (Singh and Samavedham, 2015 ), propose a method to obtain informative correlation-aware signals (Zhang et al, 2021 ), and evaluate clustering algorithms to support the prediction of the disease (Sherly Puspha Annabel et al, 2021 ). Most of the studies that aimed to classify healthy people from PD patients focused solely on leg variables or arm variables or axial trunk and knee rotation even though the disease involves all four limbs and the first affected are the arms (Ospina et al, 2018 ; Monje et al, 2021 ).…”