Engineering Uncertain Time for its Practical Integration in Ontologies
Knowledge-Based Systems Journal (link)

Abstract: Ontologies are commonly used as a strategy for knowledge representation. However, they are still presenting limitations to model domains that require broad forms of temporal reasoning. This study is part of the Onto-mQoL project and was motivated by the real need to extend static ontologies with diverse time concepts, relations and properties, which go beyond the commonly used Allen´s Interval Algebra. Therefore, we use the n-ary relations as the basis for temporal structures, which minimally modify the original ontology, and extend these structures with a generic set of time concepts (moments and intervals), time concept properties (precise and uncertain), time relations (interval-interval, interval-moment, and moment-moment), and time relation properties (qualitative and quantitative). We divided the scientific contribution of this study into three parts. Firstly, we present the ontological temporal model (classes and properties) and how it is integrated into static ontologies. Secondly, we discuss the creation of axioms that give the semantics for precise temporal elements. Finally, as our main contribution, these ideas are extended with axioms for uncertain time. All these elements follow the Ontology Web Language (OWL) standards, so this proposal is still compatible with the main ontology editors and reasoners currently available. A case example demonstrates the use of this approach in the nutrition assessment domain.

Online resource material.
Behaviour recommendations with a deep learning model and genetic algorithm for health debt characterisation
Journal of Biomedical Informatics (link)
Human behaviour is a dense longitudinal multi-featured measure that directly impacts the health of individuals in the short and long terms. Therefore, issues usually emerge from the insistence on performing risky behaviours, such as smoking or eating fast foods, which continuously increase the gap between current and beneficial health states. This paper introduces the term “health debt” as an economic metaphor to represent the quantification of this gap in domains such as sleep, contributing to physical and mental health states. Then, we present a theoretical framework that relies on behaviour change recommendations to quantify this debt. The practical instantiation of this framework relies on passively assessed sleep related data via personal wearable devices, and uses of an attention-based predictive model as the fitness function of a genetic algorithm that acts as a recommender. We evaluate this proposal by means of a case example aimed at improving the sleep duration of individuals. Results show, for example, that the use of individual rather than generic datasets produces more accurate models. At the same time, the use of constraints on the variability of behaviours features generates more feasible recommendations. These foundations open new research opportunities to support the adoption of preventive medicine based on longitudinal wearable passive data analysis.
Designing ontologies for behaviours based on temporal passive data
Applied Ontology (link)

The use of ontologies to model human behaviours that affect health is challenging since this process involves data from multiple inter-related domains that unfold and evolve over time. However, while current ontology development methodologies are generic enough to model any domain of interest, they do not provide design guidelines for modelling time-related aspects. This paper proposes a methodology for ontology development that entails the requirements for behaviours modelling based on passive temporal data. Its main focus is on temporal representations of classes and their holistic relations since no other methodology approaches ontology design from its temporal perspective. We exemplify these ideas by modelling the sleep behaviour domain, its relations to other behavioural aspects, and its effects on health.

Online resource material.
Behavioural Data Categorization for Transformers-based Models in Digital Health
IEEE-EMBS International Conference on Biomedical and Health Informatics (link)
Abstract: Transformers are recent deep learning (DL) models used to capture the dependence between parts of sequential data. While their potential was already demonstrated in the natural language processing (NLP) domain, emerging research shows transformers can also be an adequate modeling approach to relate longitudinal multi-featured continuous behavioral data to future health outcomes. As transformers-based predictions are based on a domain lexicon, the use of categories, commonly used in specialized areas to cluster values, is the likely way to compose lexica. However, the number of categories may influence the transformer prediction accuracy, mainly when the categorization process creates imbalanced datasets, or the search space is very restricted to generate optimal feasible solutions. This paper analyzes the relationship between models’ accuracy and the sparsity of behavioral data categories that compose the lexicon. This analysis relies on a case example that uses mQoL-Transformer to model the influence of physical activity behavior on sleep health. Results show that the number of categories shall be treated as a further transformer’s hyperparameter, which can balance the literature-based categorization and optimization aspects. Thus, DL processes could also obtain similar accuracies compared to traditional approaches, such as long short-term memory, when used to process short behavioral data sequences. 
Transformers in Health: A Systematic Review on Architectures for Longitudinal Data Analysis
Artificial Inteligence Reviews (submitted)
Abstract: Transformers are state-of-the-art technology to support diverse Natural Language Processing (NLP) tasks, such as language translation and word/sentence predictions. The main advantage of transformers is their ability to obtain high accuracies when processing long sequences since they avoid the vanishing gradient problem and use the attention mechanism to maintain the focus on the information that matters. These features are fostering the use of transformers in other domains beyond NLP. This paper employs a systematic protocol to identify and analyze studies that propose new transformers’ architectures for processing longitudinal health datasets, which are often dense, and specifically focused on physiological, symptoms, functioning, and other daily life data. Our analysis considered 21 of 456 initial papers, collecting evidence to characterize how recent studies modified or extended these architectures to handle longitudinal multifeatured health representations or provide better ways to generate outcomes. Our findings suggest, for example, that the main efforts are focused on methods to integrate multiple vocabularies, encode input data, and represent temporal notions among longitudinal dependencies. We comprehensively discuss these and other findings, addressing major issues that are still open to efficiently deploy transformers architectures for longitudinal multifeatured healthcare data analysis.
A Behavior Sequence Transformer for Longitudinal Quality of Life Data Analysis
Public Health Journal (submitted)
Abstract: Quality of life (QoL) data are associated with individuals’ physical, psychological, and social aspects that evolve and unfold over time. Since they directly influence the health of individuals in the long term, their continuous and pervasive analysis has the potential to facilitate avoiding or attenuating future health issues. This paper adapts the Behavior Sequence Transformer (BST) for longitudinal and multi-featured sequences of QoL data. BST relies on temporal human routines and patterns to generate inductive outcomes. Therefore, our BST-based model uses the behavioral history and profile of individuals to predict health outcomes, according to input recommendations for behavioral changes. The practical demonstration of this strategy employed a subgroup (2682 samples) of the English Longitudinal Study of Ageing (ELSA) dataset, which maintains longitudinal QoL data of participants. Results show that the multi-featured aspect of this approach improves the model accuracy, while also supports the analysis of several recommendations for behavioral changes.
A Correlation Analysis Between Passively Assessed Gait Data and Brain Tumours Progress
Artificial Intelligence in Medicine (submitted)

Abstract: Magnetic resonance imaging (MRI) is commonly used to diagnose and follow the progress of brain tumours. However, they are expensive, not easily accessible, and require frequent visits of patients to specialised health centres. This paper investigates the use of passively assessed gait data as a proxy for MRI. Three hypotheses were evaluated: (1) There exists a set of gait features more sensitive to identify brain tumours; (2) The evolution of gait is correlated with the progress of brain tumours; and (3) Locations of brain tumours are also correlated to gait features. Data from UK Biobank healthy individuals (N=240 x 7 days) were used to define a baseline for gait analysis. Then, we evaluated the sensitivity of 28 gait features, which were used to differentiate this healthy population from brain tumours patients (BrainWear project, N=49 x 9 months). We evaluated several learning strategies to create static models that could identify the existence and position of brain tumours, while the longitudinal use of these models was evaluated to correlate gait patterns and tumour progress. Statistical tests demonstrated that healthy individuals and cancer patients present significant differences (p<0.05) regarding subsets of gait features. Moreover, classification models presented a maximum accuracy of 77% to differentiate these two groups. However, the tumours’ location identification was inaccurate (33% - 65%), mainly for some regions (e.g., right temporal). We also found correlation evidence between gait patterns and tumours progress and treatments, showing that the use of passive gait analysis has the potential to corroborate with MRI diagnoses.

Online resource material (to be inluded).


  logo-qol.png   Horizon2020.jpg