Developing a computational representation of human physical activity and exercise using open ontology-based approach: a Tai Chi use case

Abstract

Many studies have examined the impact of exercise and other physical activities in influencing the health outcomes of individuals. These physical activities entail an intricate sequence and series of physical anatomy, physiological movement, movement of the anatomy, etc. To better understand how these components interact with one another and their downstream impact on health outcomes, there needs to be an information model that conceptualizes all entities involved. In this study, we introduced our early development of an ontology model to computationally describe human physical activities and the various entities that compose each activity. We developed an open-sourced biomedical ontology called the Kinetic Human Movement Ontology that reused OBO Foundry terminologies and encoded in OWL2. We applied this ontology in modeling and linking a specific Tai Chi movement. The contribution of this work could enable modeling of information relating to human physical activity, like exercise, and lead towards information standardization of human movement for analysis. Future work will include expanding our ontology to include more expressive information and completely modeling entire sets of movement from human physical activity.

I. Introduction

Physiology is the study of the normal functioning of a living organism, and its components and has been formalized as the study of vital functions of the human body [1]. The field has a close association with anatomy to understand functions about the parts of an organism [1]. Physiology provides insight within medicine as it examines the function of the human body to determine when an individual is not in a normal condition [1]. A subfield of physiology is exercise physiology, which focuses on the physiological response of an individual during physical activity, exercise, sport, or athletic competition [2]. Exercise physiology within medicine can improve functional capacity and metabolic health to either delay or prevent health burdens associated with metabolic disorders (i.e. obesity, diabetes, etc.) [2].

Movement is an essential aspect of normal functioning for a living organism and can be described through physiology. Physiological movements for exercise physiology emphasize the body’s changing movements through unsupported positions [3]. Modeling movement incorporates the usage of anatomical terminology using physiological parameters [3]. Defining movement with a physiological context is beneficial because it provides a reference for healthy mobility. It can be used to determine the difference between sedentary physiology and exercise physiology [4]. A person’s physiology dictates their ability to move and should therefore be understood in the context of the field.

Recent advancements in biotechnology which is associated with the increase of computation resources, have enabled the biomedical science revolution [5]. Computation physiology has arisen to predict and uncover characteristic features of physiological and pathological states in humans through models [5], [6]. These models have good diagnostic capabilities and are used in cardiovascular physiology, but they are limited due to simplifications made creating the system [5], [6]. The main aim of computational physiology is to couple the models together to create an integrated model that is capable of analyzing physiology through different spatial scales (i.e. organism, organs, cellular, etc.) and physical processes (i.e. sedentary, exercise, etc.) [5]. Therefore, there is a gap within computational physiology due to a lack of models for different scales and physical processes to create a holistic integrated model.

Tai Chi Chuan (TCC) is a popular form of moderate-intensity exercise that originates from China during the late Ming (1368–1644) and early Qing (1644–1911) dynasties and has not been modeled through computational physiology [7], [8]. Tai Chi prioritizes deep regulated breathing and a relaxed mindset throughout the exercise to incorporate strength and balance [7], [9]. Regular practice of Tai Chi is associated with increased strength in skeletal muscles, the metabolic response, cardiorespiratory function, and the digestive system [7], [9], [10]. Historically, TCC has been used within Chinese medicine to regulate the body’s inability to adjust to irregularities within the body (i.e. illness and disease) to return to normal bodily function [11]. One type of sequence for TCC is the 24 posture TCC sequence which repeats movements for both the right and left side and has been expanded upon with an additional 48 posture TCC [11]. The 24 posture TCC sequence is an introductory sequence into TCC which prioritizes intentional breathing with smooth movements performed at an even pace [11]. The 48 posture TCC sequence incorporates more dynamic movements which are characteristic of martial arts by incorporating balancing poses and kicks [11].

Tai Chi is associated with health-promoting benefits for diseases such as high blood pressure, heart disease, arthritis, and diabetes. Additionally, Tai Chi has been shown to improve an individual’s quality of life (QOL; i.e. physical, emotional, social, and cognitive functions) due to its physical and psychological benefits [9], [11]. Tai Chi’s ability to improve QOL is essential to improve an individual’s health and has been utilized more within health care as complementary medicine. For example, cancer is the leading cause of global death and its symptoms along with side effects of medical treatment, negatively impact an individual’s QOL [12], [13]. Cancer patients are commonly within the below average category of QOL and are most affected by pain, sleep problems, fatigue, and depression [13]. Tai Chi can improve the QOL of cancer patients because it emphasizes relaxation with intentional breathing and strengthens the body physically through movement [14], [15]. There is also the opportunity to improve falls among the aging population with consistent Tai Chi exercises [16].

A. Ontologies for Biomedical Knowledge

Biomedical ontologies, a representational software artifact that models domain knowledge, have been adopted by the biomedical community to classify real world entities and express consensus-based scientific knowledge. Essentially, ontologies leverage controlled vocabularies to symbolically represent domain knowledge in the form of a network graph of vocabularies, analogous to a human concept of a mental/mind map. These ontologies are created using machine-readable syntax like RDF [17], and OWL2 [18]. This allows for machines to understand and “know” domain information from the ontology’s knowledge base. With OWL2, the machine syntax provides semantics to further enhance the meaning of the information in the network graph of vocabularies. Currently OBO (Open Biological and Biomedical Ontology) Foundry [19], NCBO’s (National Center for Biomedical Ontology) Bioportal [20] and the OntoFox servers [21] are major repositories supporting ontologies of various medical and biological knowledge.

B. Research Objective

The purpose of this study is to create a model for computational physiology which depicts kinetic human movement through an ontology. This study models the physiological movements of a simplified 24 posture and 48 posture TCC sequence commonly used for beginners. We detail our initial work in developing an ontology that models human physical movements involving anatomical parts and formal physiological movements. We call this ontology the Kinetic Human Movement Ontology (KHMO) that leverages controlled vocabularies from open sourced biomedical ontologies found in the OBO Foundry. We discuss future direction and use cases on how an ontology that represents physical human movements could model physiological data and express meaning from the representation of physiological movements.

II. Method

A narrative review was conducted for the background of physiology and its subfields (i.e. exercise physiology, sedentary physiology, and computation physiology) to determine the relationship between the field and medicine. A narrative review was conducted for the background of TCC (i.e. its origins, popularity, and the improvement of QOL) and the physiological movements of the 24 posture and 48 posture TCC sequences were identified. The 24 posture and 48 posture TCC sequences were selected for the study because they are commonly used for TCC practice and can be performed by beginners [11].

Within an Excel spreadsheet, each stance was numbered and named in order of the 24 posture and 48 posture TCC sequence of movements. Each stance was differentiated between the start (A) and end (B) to represent the dynamics within a stance and between stances. Movement for start (A) was described as the difference in the body’s position from the previous stance’s end (B) based on the sequential order. Movement for end (B) was described as the difference in the body’s position based on the start (A) of the same stance. Movement was described using anatomical terminology in reference to simplified body parts (i.e. palm, forearm, hip, knee, waist, foot, etc.) and differentiating between the right and left side of the body (Table 1). Usage of anatomical terminology within a movement was recorded individually for each stance’s start (A) and end (B). Visual depictions of the 24 posture and 48 posture TCC sequence were created in a tablet drawing application using a stylus from one perspective and uses color code to differentiate the position of the body (towards=open outline and rearward=opaque body), identifies each side of the body (right=red and left=blue), differences in hand/feet position (purple=plantarflex/protonate and green= dorsiflex/supinate), and arrows to show the direction of motion based on the previous movement (orange=movement) (Figure 1).

TABLE I.

Anatomical terminology defining movements used in the 24 posture and 48 posture TCC sequences. The terminology defines movement throughout the sequences as differences in posture based on the previous stance. The highlighted terminology indicates its usage through color coding within the drawings created for the model.

Terminology	Definition
Flex	decreasing the angle of a joint (bending)
Extend	increasing the angle of a joint (straightening)
Dorsiflex	decreasing the angle of the ankle joint
Plantarflex	increasing the angle of the ankle joint
Elevate	moving a body part in a superior direction (up)
Depress	moving a body part in an inferior direction (down)
Evert	rotating the ankle away from the other ankle
Abduct	moving a limb away from the medial line of the body
Adduct	moving a limb towards the medial line of the body
Lateral Rotation	rotating a limb away from the medial line of the body
Medial Rotation	rotating a limb towards the medial line of the body
Supinate	rotating the forearms so the palm faces up if the forearm is flexed
Pronate	rotating the forearm so the palms face down if the forearm is flexed
Retract	posterior movement (towards the back of body) of the arm at the shoulder
Protract	anterior movement (towards the front) of the arm at the shoulder
Inferior	below
Superior	above
Anterior	front
Posterior	back

Fig. 1.

Drawing of stance 2A, The Wild Horse Parts Its Mane (Ye Ma Fen Zong). Movement (orange) for this stance occurs within the right arm (red) where it elevates to chest level with the palm pronated (purple). The left arm (blue) remains stationary at waist level with the palm supinated (green). There is no movement within the legs; the right leg (red) maintains the body weight while the left leg (blue) is slightly elevated with the foot dorsiflexed (purple).

A. KHMO Development

From the spreadsheet, we devised a representational model and terminologies that can be sourced from external ontologies. Appendix A shows our proposed model describing the human movement using controlled terminologies. Essentially, each movement includes a beginning stance (i.e., posture) and ending stance that precedes one another. Abstractly a movement is a chain of stances that precedes one another that includes a variety of physiological occurrents and anatomical entities.

We defined specific body parts that were engaged in the stance, and we defined physiological movements. Both of these are mentioned in the aforementioned centralized spreadsheet.

The model’s terminology was based on ontologies we reviewed. We performed a review using the National Center for Biomedical Ontology’s Bioportal and OBO Foundry to find ontologies and keywords for reuse. We reused terminologies primarily from the Neuro Behavioral Ontology (NBO) [22], Foundation Model Anatomy (FMA) ontology [23], and the Ontology of Biomedical Investigations [24]. These ontologies provided indirect support from other OBO ontologies such as the Basic Formal Ontology, Chemical Entities of Biological Interest ontology, NCBITaxon, and the Uberon ontology. For FMA, it provided a collection of anatomy involved in the stances, and the NBO provided supporting terminologies for movements. We pieced together these terminologies to create our core model for KHMO. We used OBO ROBOT [25] to extract and merge together the selected seed terms (Figure 2).

Fig. 2.

Part of development process involving extraction and merging existing terms from ontologies.

We used a combination of STAR and MIEROT [26] extraction methods to capture the essential terms associated with our seed terms’ hierarchy and class definition. This was to ensure that we have a complete model for KHMO. The commands and targeted seed terms are found in our GitHub repository.

We developed software code using OWL API [18], Apache POI (a Java library to manage and interface with spreadsheet data), Google Guava, and Apache Commons for utility functions. The software code was developed to extract information from our spreadsheet and generate an ontology instantiated for our Tai Chi use case. Our code is available on Github [27].

III. Results

The Kinetic Human Movement Ontology (KHMO) is hosted on our GitHub repository (https://github.com/ProfTuan/KHMO). The ontology was encoded in OWL2 using Protégé [28] and is available in both as a canonical version and a merged version (Figure 3). KHMO has 109 classes and 4235 axioms, with 2 data properties and 115 object properties. We also provided our example of the Tai Chi example as an OWL2 file, to demonstrate how we linked the KHMO to Tai Chi movements and terminologies. In addition, we used the FaCT++ reasoner [29] provided through Protégé authoring environment to validate its logical consistency, in which it passed the validation check.

Fig. 3.

Screenshot of the Protégé environment with KHMO

Using the software management code that leverages OWL API, we imported information about Tai Chi movement and converted the information into instance data. The outcome of this effort resulted in an ontology that models a sequence of movement of Tai Chi. Figure 4 shows a visualization of how the instance data of Tai Chi are structured by the KHMO. Our effort also linked the images associated with the stances to the ontology using schema.org’s image annotation. Overall, with the software and the table data, we can potentially reuse our tools to model additional Tai Chi movements as well as other forms of human physical activity.

Fig. 4.

Illustration showing the instance-level abstraction for Yan Shou Liao Quan

In reference to Figure 4, we show an example of KHMO_0000233 (Cover Hand and Strike with Fist (Yan Shou Liao Quan) beginning) that precedes the stance of KHMO_0000234 (Cover Hand and Strike with Fist (Yan Shou Liao Quan) ending). Both instances are assertions of the “human stance” class of KHMO. With KHMO_0000233, like all of the data that was imported were represented as instances. The instances were given an auto-generated identifier (KHMO_[7 digits], to conform with OBO Foundry standards), and a definition of the instance. Also, we imported the image (developed by co-author EN) to link a descriptive visualization of the movement.

Listing 1.

SPARQL query to retrieve physical anatomy for Yan Shou Liao Quan

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>

PREFIX owl: <http://www.w3.org/2002/07/owl#>

PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

PREFIX xsd: <http://www.w3.org/2001/XMLSchema#>

PREFIX obo: <http://purl.obolibrary.org/obo/>

PREFIX khmo: <http://utmb.edu/ontology/khmo.owl#>

SELECT DISTINCT ?part ?stance

WHERE

{

?part rdf: type owl: Named Individual.

?part obo: BFO_0000050 ?stance.

?stance rdfs: label ?label.

FILTER (?label = ‘‘Cover Hand and Strike with Fist

(Yan Shou Liao Quan) beginning’’@en)

}

part

–––––––––––––––––––––––

left leg (for KHMO_0000233)

set of arms (for KHMO_0000233)

stance

–––––––––––––––––––––––

Cover Hand and Strike with Fist (Yan Shou Liao Quan)

beginning

Cover Hand and Strike with Fist (Yan Shou Liao Quan)

beginning

Listing 2.

SPARQL query to retrieve physiological movements for Yan Shou Liao Quan

PREFIX rdf: <http://www.w3.org/1999/02/22-rdf-syntax-ns#>

PREFIX owl: <http://www.w3.org/2002/07/owl#>

PREFIX rdfs: <http://www.w3.org/2000/01/rdf-schema#>

PREFIX xsd: <http://www.w3.org/2001/XMLSchema#>

PREFIX obo: <http://purl.obolibrary.org/obo/>

PREFIX khmo: <http://utmb.edu/ontology/khmo.owl#>

SELECT DISTINCT ?movement ?stance

WHERE {

?movement rdf: type owl: Named Individual.

?movement obo: RO_0002331 ?stance.

?stance rdfs: label ?label.

FILTER (?label = ‘‘Cover Hand and Strike with Fist

(Yan Shou Liao Quan) beginning’’@en)

}

movement

–––––––––––––––––––––––

medial rotation (for KHMO_0000233)

flexion (for KHMO_0000233)

stance

–––––––––––––––––––––––

Cover Hand and S trike with Fist (Yan Shou Liao Quan)

beginning

Cover Hand and Strike with Fist (Yan Shou Liao Quan)

beginning

To show a brief use of our ontology, we can ask the following question: “Which physical anatomical parts are involved initially in Yan Shou Liao Quan stance?” and “What standard physiological movements are involved initially in Yan Shou Liao Quan stance?” In the code listing (See Listing 1 and 2), we have SPARQL queries showing machine-based translation of the aforementioned queries and the responses tested through Protégé. Aside from the prospect of using this ontology to curate a knowledge base of physiological human movements, we can utilize the KHMO to query data and information relating to human physical activity.

IV. Discussion

In this work, we undertook the development of an ontology to represent human physiological movement using open-sourced controlled vocabularies from the OBO Foundry. Our core KHMO abstraction sufficiently captures basic information pertinent to human health movements. Sample SPARQL queries were developed to demonstrate retrieval of basic information from KHMO’s knowledge base, regarding the participation of physiological movements and anatomical body parts in human physical activity. Lastly, with the diverse vocabularies re-used to symbolically represent human kinetic movement, Protégé’s implementation of the FaCT++ reasoner indicated no inconsistencies with our model. Overall, this is the first attempt to utilize an ontology – with open sourced controlled vocabularies – to symbolically model human kinetic movement. While this was our first attempt toward a vision to computationally describe human movement, we foresee several opportunities for expansion and maturity of this project.

Last but not least, to further explain our vision, in Figure 5 we describe how data captured from hardware can be linked and aggregated with the KHMO. The classic theoretical DIKW (Data, Information, Knowledge, and Wisdom) pyramid offers a framework to understand the contribution of this work for health informatics and health data analytics [30], [31]. In our hypothesized example (Figure 5), human body motion is captured through sensors to provide data and signals. Then pattern recognition and machine learning algorithm classifies and segments the data to information. Presumably, ontologies (like KHMO) could provide the semantics to elucidate meaning from the information (i.e., knowledge), while the researcher would interpret and apply the knowledge (i.e., wisdom). With our ontology which reuses vocabulary from existing knowledge bases, there is the added benefit of enriching the knowledge with external heterogeneous resources – furthering the researchers interpretation and application of scientific knowledge to improve and understand human health and kinetic movements. Furthermore, there is the opportunity to leverage the ontology for visualization tooling, specifically to annotate and label human movement. This could not only benefit researchers for analytical and informational endeavors.

Fig. 5.

Proposed application of KHMO using the theoretical Data, Information, Knowledge and Wisdom model

A. Use Cases and Future Direction

This early work has a few potential directions. One limitation is that we focused on one Tai Chi sequence, but we envision furthering our work in comprehensively modeling additional Tai Chi movements. The potential outcome of that effort would allow researchers to analyze the impact of healthy human physical activity toward probable health outcomes. This would also yield the probability of standardizing and organizing data for physiological movement.

While we focused on Tai Chi for the aforementioned reasons, there is the potential in utilizing this ontology for other health-related kinetic activities that are important to health care research – human falls and physical human exercise. In addition, we can re-purpose and apply the tools of this work with other types of physical activity to represent human physiological movement.

Initial progress of KHMO will focus on exercises that can be performed by a wide audience, particularly the aging population as a means to mitigate lean muscle mass loss associated with aging [32]. Future expansion of KHMO would incorporate swimming into the model because it is a low-impact exercise commonly promoted and prescribed as a form of alternative medicine [33]. Swimming is ideal for the model because it is a dynamic exercise suitable for many people (i.e., overweight individuals, pregnant women, older adults, etc.) due to the buoyancy of water and decreased joint force compression [33]. Furthermore, it is the second most popular exercise within the US and most industrialized countries, making it a viable expansion for our ontology [33].

Currently, there is a gap in swimming research due to difficulties in making physiological measurements within the water [33]. Furthermore, swimming depends on experience to obtain the skill and technique to achieve sufficient exercise intensity [33]. Integration of swimming into KHMO would provide a structure to research the physiology of swimming and potentially determine more concrete health benefits of swimming. Future work on KHMO in swimming would begin with the competitive swimming strokes (Figure 6) [34]. The movements of each swimming stroke would initially be differentiated by start and end using physiological movements and physical anatomy and further expansion of KHMO would use the same method for intermediate movements within the stroke. The parts of the swimming stroke movements would then be integrated into the ontology and expand the field of computational physiology.

Fig. 6.

Competitive swimming strokes; Front Crawl (A), Backstroke (B), Breaststroke (C), and Butterfly stroke (D).

Finally, our core model for KHMO represents a subset of human physical anatomy (set of arms, left leg, etc.) and the formal name of physiological movement (abduction, retract, etc.) for each individual stance that compose a human movement. One aspect we could model is linking the physical anatomy with the physiological movement. This way, we would be able to query the knowledge base for questions like, “What anatomical part performs an abduction movement?” Overall, we intend to add information that would link the physiological movement with the anatomical part to our existing model. Lastly, there is an opportunity to include more granular physical anatomy involved in human movement, like parts of the skeleton and muscular anatomy.

V. Conclusion

In this paper, we introduced our preliminary release of the Kinetic Human Movement Ontology (KHMO), a biomedical ontology that models entities and their relationship information for human movements and postures pertaining to physical activities. The ontology reuses terminologies from the Neuro Behavior Ontology, Formal Model Anatomy Ontology, Ontology of Biomedical Investigations, and the Basic Formal Ontology. We developed a core model that links together our own developed terms with existing terms. Our early release is available on GitHub [27]. Our future direction will include expanding the expression and definition of human movement and experiment with our ontology model on other forms of human activity including expanding our representation of Tai Chi. Overall the KHMO has the potential to standardize our understanding of human movement and the physical entities and their movement that are in coordination.

Appendix A

APPENDIX: Kinetic Human Movement Ontology (KHMO) Model

Fig. 7.

The core representational model that structures terminologies for human movement. The core model reuses terminologies from Neuro Behavior Ontology, Formal Model Anatomy Ontology, Ontology of Biomedical Investigations, and the Basic Formal Ontology