Trends & Opportunities in Visualization for Physiology: A Multiscale Overview

4.1. Spatio-Temporal Organization

To minimize semantic collisions or confusion, we classify literature into scales along a spatio-temporal axis that is roughly discretized according to biological complexity: molecule, cell, tissue, and organ. This discretization is inspired by the organization of physiology textbooks [HG11]. Fig. 5 (left) shows the distribution of non-survey papers we collected that are categorized according to this scale. Works that span multiple scales are counted once for each scale, e.g., a work that we classify as both molecule and cell scale is counted in both the molecule and cell groupings.

We bundle temporality into this scale discretization based on the fact that, as structures increase in physical size, they tend to be involved in more biologically-complex processes that take more time to complete. This relationship between increased structural size/complexity and time has been discussed elsewhere in different domain contexts [SPW*08; SS14; DMRM17; GM17]. We can observe this phenomenon in Fig. 5 (right), which represents collected works classified in a range according to the scale that the input data spatially and temporally resolve to and up to the spatial and temporal scale of the structure and process of interest, e.g., the whole brain. For example, while EEG measures neural activity, we cannot visualize individual neurons with this modality, and that is not the intent of conducting these types of studies. Visualizations of EEG data fall in the organ scale. The relationship between space and time is not perfectly linear, as reflected by the dark groupings in the upper center and right regions of the chart. The upper part corresponds to organ-scale processes that occur over the range of seconds, such as a heartbeat or a full breath cycle, while the right part corresponds to the time for the expression of a single gene [MJM*10].

Our classification system does not formally extend beyond the organ scale for a few reasons. First, restricting the scales we examine keeps the scope of this survey manageable. In addition, our preliminary searches found limited visualization research that exists purely at the system– or organism–level, and the tasks and visualization techniques implemented are similar to those observed at the organ scale. We briefly review examples of works beyond the organ scale, as well as selected works that use true multiscale approaches, where the visualization aims to facilitate a task at three or more scales, in Sec. 9.

Although an enormous range of physiological processes occur at all scales of the body, we focus our survey on a few categories of processes at each scale that are timely for the physiology domain. Processes occurring at the molecule scale include molecular dynamics, e.g., the motion of atoms and molecules, reactions between molecules where electrons and/or atoms are exchanged, and molecular pathways, which describe a chain of molecular reactions. Cell-scale functions that we highlight include cell dynamics and interactions, such as how the cell develops and communicates with its environment. We include the dynamics of the cell's organelles at this scale, such as mitochondrial activity. Tissue-scale functions consider the behavior of aggregates of cells of the same type and include tissue dynamics, such as growth, and tissue interactions, such as signal propagation in neural tissue. At the organ scale we consider processes related to blood flow as well as the functioning of the heart, brain, and lungs. The body of visualization literature at this scale is large in correspondence to the maturity of data acquisition techniques available and the ease at which these processes may be captured or simulated.

Non-human studies present an issue in this measurement-based classification system. For example, a visualization of the neural pathways in a fruit fly brain concerns micrometers, while a human brain measures in centimeters. Although the focus of this survey is on human physiology, there is immense value in considering model organism physiology. These experiments tend to be more innovative, with correspondingly greater likelihood of exciting visualization opportunities. In cases where we include model organism physiology visualizations, we map the organism's scale to the human scale. Following this logic, we classify, e.g., a visualization of fruit fly brain activity at the organ scale.

4.2. High-Level Visualization Task

A subsequent layer categorizes the literature according to three high-level visual tasks: exploration, analysis, and communication, as illustrated in Fig. 6. These tasks are drawn from Brehmer & Munzner's typology of abstract visualization tasks [BM13]. We chose high-level, rather than low-level, tasks to provide a clear picture of the broad needs and challenges users face in visualizing physiology and how this compares across scales. We first considered categorizing works according to visualization technique, e.g., direct visualization. However, since task ultimately drives the chosen visualization technique, we feel that this is a more meaningful classification mechanism that furthermore has been the basis of classification in other surveys.

Exploration tasks often arise when the user is unsure of what the data contain. In the context of the data visualization pipeline, the user typically wishes to minimally abstract the data and produce a visual mapping that is as close as possible to reality. They do this to explore what the data actually contain. This is often a preliminary step in a larger analytical process. Analysis tasks occur when the user may be more sure of the intrinsic characteristics of the data, but now want to extract meaning from these data. Analysis often relates closely with exploration, where a user may begin with an exploratory approach to generate a hypothesis, then perform low-level analytical tasks alongside statistical methods to follow up on their hypothesis. In the visualization pipeline, analysis involves production of new artefacts through data transformation, derivation, and abstraction [AMM*07]. Although important for any task, audience is a key part of a communication task, where a visualization is created to underline key concepts of the data for presentation, education, or enjoyment to a particular group, whether to peers or to a broader audience. Visualizations developed for this task are often further abstracted from the data than in analysis- or exploration-oriented tasks, and can incorporate cinematic or storytelling elements to convey the author's interpretation of the data. While nearly all publications include figures to communicate scientific results, for this survey, we identify uses of visualization for communication beyond what is achievable with standard, out-of-the-box tools.

Many visualizations cannot be defined through strictly one of these tasks and rather are often generated to meet a combination of tasks. A work created to explore the data may also specify a visual analysis task. For example, ZigCell3D [dHKMK13] visually explores simulations of cellular functions while also providing tools for the visual analysis of the underlying simulation. The same data may be visualized for a communication task if the data are more visually abstracted or if annotations or glyph overlays are added to tell a story about the underlying information. This may also summarize key findings from, e.g., a visual analysis session for a broader audience. We apply weighted categorizations to each work, excluding surveys that cover many works. This produces a vector of three values between 0 and 1 for exploration, analysis, and communication, respectively. We then use this vector to position a work in a barycentric coordinate space, which allows us to compare and contrast between similar works within and between different spatio-temporal scales.

We arrange works graphically within a triangle where each of the three points corresponds to a high-level task, as shown in the left of Fig. 6. Exploration resides at the top of the triangle to reflect that, when exploring data, we are in a position of knowing the least about what we are looking for and/or the data are in their least abstracted form. Moving clockwise to the right corner is analysis, where we usually know something about the data and what we are looking for. Communication resides at the left corner, where data are highly abstracted and summarized in order to present, communicate, or serve data for enjoyment. The set of four triangular glyphs in Fig. 6 summarizes our visual taxonomy. Each triangle represents a single scale space, where the three triangle points represent the three respective visualization tasks. Circles indicate the position of each work as encoded by its balance of exploration, analysis, and communication tasks. Circle darkness and size dually encode the number of works with a given task categorization that we collected in our survey.

Literature Overview. The scale and task categorizations for each literature item collected for our survey can be browsed in the References section. Scale is labeled with a grayscale color tile, and related surveys are labeled with a black tile. Individual categorized works include a miniature bar graph that indicates the task(s) addressed, i.e., exploration (yellow), analysis (magenta), and communication (blue), from a range of 0 to 1. An interactive overview of the complete literature collection is available at https://lauragarrison87.github.io/star.web/vis_tool.

5.1. Molecular Dynamics

Molecules are flexible and dynamic structures that frequently transition between conformational states. These structural dynamics are due to interactions between a molecule's atoms, with nearby atoms from their environment, and environmental conditions like temperature and pressure [Rin18]. Molecular dynamics are characterized by the time scale of their conformation fluctuations (kinetics) and the amplitude and directionality of the fluctuations (structure). These fluctuations form a multidimensional energy landscape. Local fluctuations typically occur over nanoseconds, while global fluctuations can span microseconds to seconds. These global fluctuations are big conformational changes that signify protein-protein interactions, or reactions that initiate a molecular pathway, e.g., signal transduction [HK07]. Domain researchers are particularly interested in this energy landscape as it applies to understanding mechanisms of disease and for drug design.

Visualization approaches that target the flexibility of molecular structures often use nonphotorealistic visualization techniques that show molecular surfaces at atomic resolution. Ball-and-stick and ribbon visualization representations are also commonly used [JH14; KKF*17]. Color is often assigned to highlight differently-flexible regions. In addition, many visual analysis methods incorporate simple graphical elements, such as glyphs or path-line visualization techniques.

Approaches that are mainly exploratory in nature are intended to allow researchers to browse and familiarize themselves with the results of a molecular dynamics simulation. These approaches tend to use minimal abstraction and encodings that are familiar to the domain [GBC*14; LBPH10]. This also includes tools like VMD [HDS96] or PyMol [Sch15], which are widely adopted in the application domain. Visual elements may be used to draw out features within the data for exploration, such as pathlines to indicate atomic paths that drive changes in overall molecular shape [DRSR15].

As researchers become more familiar with the data, they may switch from exploration in an overview to analysis of a particular region of a molecule or molecular complex. Visual abstraction methods that exploit the hierarchical structure of molecules are useful to facilitate toggling between exploration and more focused identification and comparison tasks [LVRH07].

Analytical approaches tend to incorporate interactive techniques and/or statistical methods. These allow researchers to identify and compare specific information about how parts of the molecule are moving in relation to one another. These approaches are of particular interest for researchers in drug development and protein engineering. A structural visualization of the molecule is usually important alongside 2D plots showing, e.g., trajectories [GBB*19]. For example, Fioravante et al. [FSTR13] use principle component analysis to cluster molecules that have correlated motions, while Schmidt et al. [SBH02] derive mean shape conformations from the data to allow researchers to identify and compare metastable conformations. Other methods incorporate additional visualizations, such as time curve plots and heatmaps, to help researchers identify particular shape changes or constraints of interest [DCS12; TTL*15]. Interactive filtering techniques also help researchers identify particular movements of interest [HV00].

Conformation changes of a molecule affect not only its outer shape but also the shape of cavities or tunnels in the molecule. The shape of these tunnels affects the ability of a ligand, i.e., a signaling molecule, to travel to its binding site within a molecular cavity or tunnel [HG11]. These approaches usually include a mix of direct 3D visualization methods alongside heavily abstracted methods to accommodate a specific goal, e.g., to understand how the shape of a tunnel changes over time. Visual analysis methods include aggregating a molecular dynamics time sequence to a single contour plot [BLG*15]. Heatmaps to show variation in tunnel properties, such as tunnel centerline length, amino acid composition, and bottleneck size, can also be paired with direct visualization of a molecule [BJG*15; GHX*20], as shown in Fig. 7. More extensive visual abstraction from the original molecule shape can be used to understand dynamic structural changes and energy landscapes without occlusion [KBP*16; KFS*16; LAQS20].

Limited visualization research is dedicated to the communication of molecular dynamics, as much of this work comes from collaborations with domain experts with specific exploratory or analytical goals. Communication-oriented works use graphical elements, such as arrow glyphs, to illustrate molecular flexibility [BJG12]. Tools geared towards medical illustrators, such as Molecular Maya (mMaya) [Cla22], allow artists to animate molecular motions.

Summary. Most of the works visualizing molecular dynamics are targeted at domain experts for a combination of exploratory and analytical tasks. The time scale over which molecules change shape ranges over at least nine orders of magnitude. This presents a yet–unsolved visualization challenge to provide exploratory and analytical tools to experts to review and identify movements of interest in a vast temporal space.

5.2. Molecular Interactions

Molecular interactions can lead to an (ir)reversible reaction between two molecules [Ede13]. This can change the properties of the input molecule(s), synthesize a new molecule, e.g., polymerization, or destroy a molecule. Enzymes speed up the rate of a specific chemical reaction within a cell. Ligands form a complex with another molecule, often a protein, at a binding site [HG11]. This binding initiates a series of reactions. The time scale of molecular interactions is large, ranging from nano- to seconds, which presents a similar visualization challenge as we discussed in Sec. 5.1.

Similar to molecular dynamics, visual approaches to molecular interactions are strongly spatial and tend to focus on exploratory and analytical tasks for domain experts. Experts are often interested in exploring a simulation of interactions between molecules and in analyzing those interactions, e.g., protein-ligand interactions, that could lead to binding events that trigger a molecular reaction. Multi-view visualization approaches are ubiquitous, where at least one view typically uses surface models and nonphotorealistic rendering techniques to visualize the molecule(s) of interest at atomic resolution. Coloring of the molecule(s) is often according to biochemical properties or measures of uncertainty. Standard information visualization techniques, e.g., line, bar, and scatter plots accompany the spatial view to describe interaction energies and other important simulation parameters.

Key research questions relate to positional relationships between the protein and ligand, which influence the likelihood of binding. In some instances, the researcher wishes to observe such interactions in living cells, as Kerppola's work demonstrates [Ker06]. Detailed position and interaction information from structural and simulation data can be shown on a per-atom basis through direct visualization of structural and simulation data. Glyphs and color-coding on molecular isosurfaces often enrich the visualization with additional information [TWK*11; TMJ*15].

Works that target the identification of important interaction and binding events incorporate multiple data sources, e.g., simulation and mass spectrometry [MB19], often in interactive views with some level of guidance. Aggregation of trajectory data to aid the analysis process is common. In addition, navigational techniques help experts locate features of interest in these often large and highly complex datasets. Such techniques can allow users to reveal different levels of detail on-demand [AKCL19], incorporate focus+context techniques, as in the CLISD view of protein-ligand interactions by Schatz et al. [SFS*21], allow filtering of subsets of trajectories [JFB*19], and allow users to jump to different parts of the simulation timeline [DHR*18]. Visual abstraction of the hierarchical structure of molecules may also be exploited for the analysis of different configurations of protein complexes [FJK*19]. Many works visualize 3D molecular structures alongside interaction energies and other important molecular parameters. These parameters may be represented by glyphs as by Hermosilla et al. [HEG*17], or using scatter and box plots as by Furmanova et al. [FJB*16].

Works that emphasize the broad scale of space and time over which such interaction events can occur use adjustable aggregation measures to manage spatial and temporal complexity [BTM*19; PBBH19; VS21]. Other works eschew 3D structural information entirely in favor of abstracted graphics to visualize pairwise interactions of interest [VHG*18; ZLW*21; SC21].

Tools and techniques from medical illustration and animation can be used to explore and share possible hypotheses in modeling environments, such as AutoDesk Maya [NLHI20], or to communicate molecular interactions between experts or to other stakeholders. Approaches that give users tools to create molecular interactions through rule-based frameworks enable exploration and sharing of the resulting simulation data [GCK*18; NSK*21]. Guided, interactive exploration through a rule-based simulation to track interactions in a molecular environment allows users to see the direct output of the simulation results or understand the spatial context of reaction events between molecular structures. Some of these methods employ illustrative techniques, such as focus+context, that are approachable for education and outreach [MPSV14], and incorporate multiple temporal scales into the visualization [KPV*14]. Visual complexity of molecular interactions scenes is an ongoing challenge, but research has shown that oversimplifying the crowded environments in which such interactions take place can be counterproductive to learning [JM12].

Figure 8 The Compressed Ligand Interaction Sequence Diagram (CLISD) [SFS*21] provides an overview of protein-ligand interactions over the course of a simulation, with unimportant residues visually de-emphasized and important residues and related parameter values given greater visual emphasis through size and color cues. Reproduced under Creative Commons CC BY license.

Summary. Exploring and analyzing molecular interactions is valuable for experts to understand and identify features and behaviors that can be used for pharmacological research. The main challenge to visualization is to continue researching effective methods that allow experts to understand the massive simulation datasets that are generated. This can be achieved via interactive tools to enable the identification of reaction events that occur very briefly within a temporal space that spans several orders of magnitude.

5.3. Molecular Pathways

Reactions between molecules create small changes in their immediate environment that trigger other reactions. This chain of reactions describes a molecular pathway [HG11]. Metabolism, signal transmission, and gene regulation and expression pathways are essential to life. Metabolic pathways describe the sequence of chemical reactions that occur in our bodies, such as the process for a cell to break down food into energy, or a pathway that builds a new molecule. Signal transduction pathways move a signal from the exterior to the interior of a cell with the help of proteins embedded in the cell surface known as receptors. Gene regulatory pathways turn genes on or off. When a gene is turned on, this allows the process of gene expression to occur, which transcribes and translates DNA instructions to create, e.g., a specific protein [Ins20]. Pathways do not exist in isolation and interact together in larger networks.

Understanding the participants, sequence, and timing of molecular pathways is key to understanding physiology at larger scales. Given the complexity of the input data, visual methods to address these goals tend to target expert user exploration and analysis tasks. These incorporate varying degrees of abstraction and interactivity. 2D information visualization techniques dominate, with networks being the most common technique, to show a sequence of steps in a pathway. Heatmaps, line plots, chord diagrams, and histograms are common for the visualization of gene expression.

The most straightforward visual methods allow experts to explore and identify the sequence of actors that participate in a given pathway(s) use node-link diagrams. Perhaps one of the most well-known pathway exploration tools, Cytoscape [SMO*03], uses node-link diagrams to visualize complex pathways and networks for users to explore and query. Such diagrams show entities in highly abstracted glyphs, often indicate reaction direction, and can indicate the location where the pathway takes place [LHM*09; KS20]. Brushing and linking [GVS*20], filtering [LPK*13], comparison [Sch03], and focus+context [JKL*10] techniques for detailed analysis are often supported. Numerous works have explored different layout algorithms to reduce crossover and clutter of these complex and often crowded visualizations [BALJ06]. Many implement graphical representations using, e.g., a subway map metaphor [LYKB08], that are approachable to broader audiences [CBF*15; KDE*15]. Visualizations of pathway simulations can be abstracted in 2D as line charts or heatmaps [SMW*21] to help experts to better understand the timing of pathways. An entirely different pathway simulation approach by Le Muzic et al. [MWPV15] employs an agent-based approach with 3D molecular structures to tell a multi-temporal scale story that provides insights to both experts and broader audiences alike.

Identifying and comparing levels of gene expression can provide valuable information to researchers on the activity of a given pathway, while studying gene co-expression can provide understanding of patterns and similarity of certain expression pathways. Gene expression and co-expression data are most commonly displayed in heatmaps, parallel coordinates, and chord diagrams, as shown in Fig. 9 [LWLG21]. Tools like Caleydo [LSKS10] enable the exploration and analysis of large-scale pathway data alongside gene expression data, using node-link diagrams and heatmaps in a 2.5D layout. OmicsTide [HFKN21] uses clustering with profile plots in a Sankey diagram to compare trends from gene expression and proteomic data. Some tools capture the multiscale nature of gene expression in visualizations that span the scale of individual nucelotides to entire chromosomes [MMP09]. Gene expression is a dynamic and fluctuating process. Other tools allow for exploration and analysis of temporal patterns of these fluctuations [MWS*10], and in some cases use clustering methods to facilitate pattern identification [CAM18].

Researchers are similarly interested in identifying and comparing concentrations of metabolites in specific locations of the body. This information provides another perspective on the activity of certain pathways. Tools for visual exploration and analysis of metabolite concentrations are useful to understand metabolic profiles of diseases at a molecular level [NLK*14]. Such approaches can use basic statistical methods alongside heatmaps [GVC*20], violin [JEG*19], or star charts [JML19].

Strong communication-oriented approaches to visualizing molecular pathways often draw inspiration from medical illustration and use cinematic elements to convey pathway information, such as Berry's animations showing the process of DNA transcription in real-time [Ber18]. In this way, the molecular dynamics and reactions between molecules at key steps in the pathway can be visualized in a larger context. Large charts showing pathway elements, when mainly used for communication, usually rely on abstraction of visual elements to create a scene that balances accuracy with readability [BVR*17; GMF*21].

Summary. Similar to molecular dynamics and interactions, the majority of visualization research works focus on expert-centered exploration and analysis tasks. The extraordinary complexity and volume of these data often necessitate guidance in interactive methods, and many approaches use statistical methods to reduce the analysis space alongside minimalist graphical elements. Further research into methods that facilitate a greater degree of exploration for hypothesis generation of these data, while managing the volume of information present, is an ongoing visualization challenge and opportunity for all molecular processes. Visual communication research for pathways is also important to develop further. Giving the public better tools to understand how diseases work, such as in COVID-19, can improve adherence and trust in public health protocols. Understanding physiology at this scale is essential, as molecular dynamics, interactions, and pathways work in concert to trigger behavioral and physical responses that form the foundation of cell physiology.

6.1. Cellular Dynamics

The dynamics of a cell are dictated by molecular pathways and by behaviors of its organelles, which are themselves modulated by molecular pathways. These pathways drive the dynamics of a cell's organelles, the ability of a cell to move in its environment, the suite of internal mechanisms that dictate a cell's growth, and that lead to cell division and death, to name a few processes. We also discuss visualizations for whole cell models.

Organelles participate in and facilitate the network of pathways that drive the overall behavior of a cell. Visualization tasks related to organelles are often exploratory in nature, e.g., to observe the effects of an experimental condition under microscopy. Visualization methods from the domain often show time-lapse imaging data unmodified or surface renderings. This can help users to understand the shape changes a cell nucleus undergoes in response to experimental conditions [SKZ*15], the movement of cellular vesicles [RVT*20], or the compaction of chromatin in the nucleus over different phases of the cell cycle [OPD*17]. Analysis-oriented approaches often color-code regions of interest on volume-rendered segmented data or raw data slices to identify and compare features that indicate the functioning of underlying pathways [JLC*17; Gla20]. Glyphs are used to annotate features of interest on imaging data as a process occurs [WAM*20], and heatmaps can quantify interactions between organelles over the course of an experiment [VCL*17].

Simulations using 3D surface models can help answer questions about how organelle structures move and behave. For example, Waltemate et al. [WSB14] visualize membrane dynamics at molecular resolution in small “patches” of the cell membrane. More recent works visualize microtubule dynamics [KVGM19] or dynamics of mitochondria and cell transport vesicles [PKWM20], the latter of which is shown in Fig. 10. Such visualizations are adaptable for use in education environments, with systems like LifeBrush designed to explore mesoscale environments, e.g., the mitochondrial membrane, at molecular resolution in VR [DSJ19]. Even further toward communication are hand-crafted animations, such as the ground-breaking Inner Life of the Cell [Bol06], which shows the interplay between various organelles and molecules within the cell using cinematic techniques and visual abstraction to focus the narrative.

The individual dynamics and interactions of organelles influence and facilitate the cell's response to input from its environment and internal mechanisms that push the cell through its life cycle of growth, division, and death. Visualization of cell movements in 2D or 3D to discover and understand behaviors under experimental conditions is common in the domain, e.g., to understand cell cycle progression [SKM*08; HRP*12].

Researchers may be interested in localizing subcellular structures and interactions through direct observation of imaging data [ASTM07] or may supplement imaging output with histograms, time plots, and similar aggregate visualizations to quantify features of interest [XMM*20]. Approaches may also use such aggregate visualizations to visualize results of classifications based on live microscopy alongside gene and protein expression data [BHK*21]. Dimensionality reduction methods, such as tSNE and HSNE, are useful methods to characterize and compare different cell behaviors and types from high-volume experimental data. Visualizations can map the results of these methods to a scatter plot, with color coding the cell type according to an expression marker [HPvU*16], and include guidance methods to more easily explore and analyze the data [HPvU*17]. Others deal with these high-volume data by identifying exemplar cells, either automatically or through user specification, to track specific derived attributes, such as growth, in response to drug treatments for cancer research in a multi-view visualization [LPJ*22]. Simulations allow researchers to manipulate cellular parameters for exploration and analysis. Simulations of parts of the cell cycle can be queried and explored for hypothesis generation in tools like CellCycle-Browser [BYG*18], which uses an interactive multi-view visualization containing heatmaps, line, and scatter plots to show the results of parameter changes to the system.

Whole-cell physiology is naturally multiscale, with limited works addressing visualization and specific user tasks at both molecular and cellular scales. These works enable experts to better understand intracellular functional and structural relationships. Highly abstracted approaches, similar to node-link network diagrams to represent molecular pathways, can visualize the multiscale interactions that occur within the cell [QHW*21] for querying or exploration. Visualizations of whole-cell simulations in 3D are useful to put molecular pathways into context, such as the effects of signal transduction on the cell's function [FKRE09; FKRE10] or the conditions and events that lead to cell death [FDSE11; FKE13; SBD*14]. WholeCellViz [LKC13] and ZigCell3D [dHKMK13], the former which is shown in Fig. 11, are whole-cell modeling frameworks. These frameworks allow researchers to explore and analyze cellular simulations in a biological context, from the molecular to the entire cell scale. They include pathway information as maps, as well as animation. ZigCell3D also incorporates imaging data and 3D models. A recent structural model of a whole Mycoplasma cell [MAK*22] provides an unprecedented means for researchers and the general public to explore and understand the structural and functional relationships of entities within the cell.

Summary. The visualization of cellular dynamics puts molecular pathway information into a cellular context and enables understanding of overall cell behavior. Experts are often interested in exploring and quantifying these data directly from imaging methods, with analysis of key features in aggregated plots. Further research into more interactive methods to facilitate analysis that allow experts to move away from simple rendering of microscopy data is a possible direction to explore. Very few works, especially from the visualization community, support expert study of organelle dynamics and behavior. This is an open space for visualization research. Multiscale visualization becomes truly meaningful at this scale to connect molecules with cellular behaviors. While numerous methods allow for exploration of whole-cell physiology, analysis of such models remains relatively limited, and this is another future research opportunity. Finally, in some contexts, communication-oriented approaches can serve both experts and a broader audience equally, as cells are less conceptually-abstract entities than molecules. Research into such approaches, particularly with regards to public health and in facilitating conversations on the mechanisms of disease, is an exciting challenge.

6.2. Cellular Interactions

In reality, cells do not exist in isolation. Their physiology is strongly influenced by interactions with their environment and neighboring cells. In this section, we discuss works that focus on the behavior and fate of individual cells, where understanding the environment and neighboring cell interactions are key to the user task.

As in previous topics, many visualization works are born out of collaborations with domain experts and mainly address exploratory and analytical tasks. These methods allow researchers to browse experimental, imaging, or simulation data to understand cell communication, lineages, and migratory patterns.

Direct visualization of live cell imaging data provides an overview of cell division, adhesion, signaling, and movement patterns within different environments, e.g., tumor microenvironments [EEA*08]. Clustering methods can facilitate user exploration of intercellular communication networks from single-cell transcriptional data. COMUNET classifies and clusters cell types according to ligand-receptor pairs and visualizes these communication patterns in node-link diagrams [SS20]. Migration patterns of differentiating cells can be understood by visualizing, e.g., spatial transcriptomics data, as in sci-Space [SRB*21] (Fig. 12), where heatmaps of gene expression patterns are measured over pseudo-time to capture cell differentiation and migration in a tissue context. Clustering methods can classify cells according to their migratory and other behaviors from microscopy data, which is valuable for comparative analysis [FHWL12]. Simulations with simplified 3D spherical models to represent individual cells provide information and control on per-cell properties of division, adhesion, and other environmental variables in vitro [Pal08; GHF*18].

Cell lineages contain valuable information on patterns of cell division, growth, differentiation, and death over generations of cells. This is particularly important with stem cells, which have unique regenerative abilities that have massive implications in cancer and other areas of medical research. Statistical methods to make sense of these patterns, in combination with the branching tree structures of lineage diagrams, help researchers identify and compare factors that influence cellular genealogies [GLHR09; PKE15]. These approaches can include visualization of cells within a spatial context, with navigational tools to observe how cells divide and where they migrate, e.g., to neighboring blood vessels [WWB*14; SGAT21]. Uncertainty due to segmentation of microscopy data when tracking cell aggregation is a challenge. Tools like Uncertainty Footprint [WH17] attempt to visualize and quantify these uncertainties for domain experts.

A growing interest in public science education has led to the development of tools like Bioty [WSGR17], a real-time programming environment that visualizes cell interactions for non-experts. As for other topics, hand-crafted medical illustrations are used to educate audiences on cellular interaction processes, such as the communication between a neuron and muscle cell [Goo09].

Summary. Visualizing cellular interactions adds a degree of complexity to cellular function visualizations, as these leave the self-contained environment of the cell to include external parameters that increase the complexity of the system. Collaborations with experts provide a means to explore and analyze data acquired either experimentally or through simulation, where gaining an understanding of the data through exploration is equally important to more targeted analysis tasks. Developing methods to facilitate exploration and analysis of cellular microscopy and lineage information through visual abstraction while retaining expert trust is one research opportunity. We found few works visualizing cellular migration and adhesion, particularly from within the visualization community. Given the importance of these behaviors in normal development and disease, this is yet another research opportunity.

7.1. Tissue Dynamics

Tissue dynamics refers to the behavior of aggregates of cells in a given tissue type. Visualizing tissue dynamics simulations allows researchers to explore the general mechanisms of tissue growth and development under changing environmental conditions. Understanding spatial relationships at this scale is particularly important, and many visualizations represent simulation data as 3D surface models to capture the development of blood vessels (angio-genesis) [QMK*09], liver tissue regeneration [HBB*10], embryonic limb [CHC*05] and organ development [DKY98; CGN*10], and in a non-human case, wing development [BWB*12]. Exploring changes in skin tissue as a response to aging in 3D is also of interest [IAJG15], with the ability to change parameters to simulate the impact of disease or dehydration. In addition to visualizing normal processes, simulations are valuable sources to visualize pathologies related to tumor growth under changing environmental influences [TMS*11; TRM*12].

Although we have discussed gene expression data previously for the analysis of cellular interactions, this family of methods is useful for tissue-level visual exploration and to identify biomarkers in cell aggregates when performed in situ and paired with imaging data. These cell aggregates are often identified through clustering methods [SIL*20], and can then provide molecular and cellular resolution maps of the body, e.g., of embryonic tissue development in the first trimester [BGC*17]. Subsequent exploration of such tissue maps provides an opportunity to discover emergent properties at this scale. Multeesum [MMDP10] exemplifies this interplay between exploration and analysis, where comparing similar expression profiles of aggregates of cells allows researchers to form hypotheses about gene relationships and location. Numerous domain approaches also use visualization mainly to confirm hypotheses, e.g., the direct visualization of digital histology slide data to quantify the progression of liver tissue damage in fibrosis [LLJ*21].

Communication-oriented visualizations of tissue dynamics may take the form of adjustable simulations with easy-to-use interfaces and simple graphics that appeal to both researchers and a broader audience [WT21]. Animation of 3D models is also useful as an educational tool for showing the process of organ development, e.g., of the developing heart [SADC02]. Lastly, hand-crafted illustrations that describe models of tissue growth, as shown in Fig. 13, are invaluable to clearly and succinctly share models with peers [MMF20].

Summary. Visualization of tissue dynamics is often geared first towards exploration to familiarize oneself with the data, as data at this scale are typically complex and high-dimensional. Comparison tasks between groups are then common, where experts wish to identify parameters or biomarkers that define certain tissue behaviors or functions. Tissue dynamics are challenging to visualize in vivo, with approaches often using underlying processes such as gene expression or the presence of other biomarkers to characterize tissue functional properties. Simulations provide the means for visualizing truly dynamic growth processes in healthy and disease conditions. However, they often are abstracted from reality, with visualizations that expose only the final part of a multiscale story that is rooted in the molecular scale and with limited interactivity. Devising methods to enable fully interactive exploration and analysis of tissue dynamics, whether purely at the tissue scale or extending across scales, is a grand challenge and opportunity in visualization research.

7.2. Tissue Interactions

In this section, we focus on the interactions between different tissue types that allow for the passage and exchange of nutrients, as in tissue perfusion, or for the passage of electrical signals, as in signal propagation.

The function of blood flow on the microscopic scale is to supply, or perfuse, tissues in the body with oxygen, nutrients, and hormones and to transport waste products away into the appropriate “recycling” centers such as the lungs, kidneys, and liver. Different tissues have different perfusion rates, and visualization can be a powerful tool in profiling tissues based on these data. Perfusion data are particularly useful in identifying the extent, composition, and metabolic profiles of tumors.

Most use cases to visualize tissue perfusion are highly clinically-motivated with a particular set of analysis questions already in mind, although many methods incorporate a degree of exploration. These approaches incorporate structural visualizations of tumors and the surrounding tissue to provide context, and use derived multi-parametric imaging data to classify and visualize key physiological parameters. Approaches can use simple color overlays with parameters mapped to color channels [ABBM05] to quickly quantify values. Advanced visualization approaches allow for user interaction and exploration that incorporate time intensity curves [OPH*09; GPTP10; MNM*16], radar plots [MHBS20] as shown in Fig. 14, and scatter plots with glyphs encoding further information [MWH*20]. These representations allow experts to identify and compare features of tumor physiology.

In the human central nervous system, information is processed by signal transmission and propagation between neurons, where the extracellular space plays an important role in transmission and signal propagation.

In visualizing signal propagation, particularly in simulations, experts wish to understand the mechanism, path, and timing of these propagation events. Straightforward visualizations that plot spikes in signal propagation are relatively common in the domain literature, such as in Rhodes et al. [RPR*20]. Microscopy data often provide a structural foundation for visualizing simulations of signal propagation between neurons or in a multi-neuron network [LHH*12; BHdM*21]. We show an example of a multineuron simulation network from BioDynaMo in Fig. 15 that is realized through procedurally-generated surface models. Dimensionality reduction methods can facilitate exploratory visual analysis of signal data, and allow users to identify patterns that signify, e.g., key points of a behavioral task [BKSP11]. Abstracted 2D plots, such as L-plots proposed by Dunin-Barkowski et al. [DLO*10], allow experts to observe and compare neural signaling patterns. Signal propagation is influenced by several factors, e.g., the distribution and density of glycogen around a synapse, which is the space where two neurons meet. Abstractocyte [MAB*17] is mainly designed for the visual analysis of astrocyte structure and distribution around neurons, but its pipeline includes glycogen distribution analysis.

Simulations of signal propagation are not limited to neural tissue. Visualizing simulations of the electrical conduction system in cardiac muscle tissue is of interest for experts to understand the timing and rate of signal propagation in different phases of the cardiac cycle [HBC06; BCG*11].

Summary. Unlike perfusion of tissue, the majority of work we found for spatially visualizing signal propagation comes from simulation data. As hardware and software continue to advance and support more complex simulations, there will be a corresponding increasing need to provide visual methods to explore, analyze, and communicate these data to various stakeholders. In tissue perfusion, visualization research that further supports exploratory analysis to identify complex biomarkers is an ongoing challenge. Visualization approaches for both tissue perfusion and signal transduction tend to have strong analytical components, especially in the case of perfusion, where clinical diagnostic improvements are the driving need for these applications. Communication-oriented research works are limited at this scale. The clinical motivation for understanding physiology becomes even more apparent at the organ scale, which deals with the interplay between different tissue types on a larger scale.

8.1. Blood Flow

While we previously looked at blood flow from the lens of how it supplies nutrients to tissues (Sec. 7.2), researchers often are interested in the dynamics of blood itself as it travels through the heart and vessels of the body. Understanding patterns of blood flow can help researchers and clinicians make better decisions about patient health, such as when to operate on an aneurysm.

Related Surveys. This is a mature area with several surveys and state-of-the-art reports available. For further details on visualization techniques and challenges on this topic, we refer to reports by Markl et al. [MKE11], van Pelt et al. [vPV13], Vilanova et al. [VPP*14], and Stankovic et al. [SAG*14]. For further reading on visualization methods specific to PC-MRI blood flow, see Köhler et al. [KBvP*17]. Most recently, Oeltze-Jafra et al. [OMN*19] survey trends and challenges in visualizing medical flow data, where the primary focus is on blood flow data. These surveys highlight a mix of exploratory and analytical visualization tasks, where tasks are highly motivated by domain experts' needs to locate and identify flow features that impact patient health. General flow visualization techniques are commonly used, e.g., glyphs, textures, integral curves, line integral convolution (LIC), colored cut planes, extraction to surface models, e.g., streamlines. Contextualization of blood flow dynamics using image slice or volume rendering of the surrounding anatomy is key in nearly all blood flow visualization scenarios [OMN*19]. Advanced visualization techniques often incorporate multiple interactive views with facilities for validation, filtering of key parameters, and uncertainty analysis.

Exploratory tasks generally aim at obtaining an overview of flow patterns and often precede a quantitative workflow. These may visualize simulation data [RNNC15; NCW09], use techniques that combine different modalities to create the visualization [PH07; BFS*09; FFI*12], or use single imaging methods [BdHdK*16]. The containing structure of interest, e.g., a vessel or the heart, is useful to preserve for context. The paths and direction of flow data can be presented as pathlines [dHJEV16], streamlines [BFW*11], arrows [HPT*08], and pathlets [ASN*14], which can also be used for analysis. Newer approaches have experimented with effects like smoke or dye to better visualize time-varying flow patterns with indications of uncertainty [dHLJ*19]. Interactive elements, such as virtual probes [vPBB*11], aid in expert exploration of flow patterns prior to quantitative analysis.

Although illustrative techniques are often associated with communication-oriented tasks, most illustrative techniques employed for this topic are aimed at experts to facilitate exploration of flow data. These approaches can, e.g., reduce occlusion from vessel walls through adaptive surface visualizations [GNKP10; LGP14] or focus+context flow lens treatment [GNBP11]. Other illustrative approaches facilitate exploration of wall thickness relative to flow properties [LGV*16].

Many analytical approaches from the domain are limited in interactivity. These approaches again include a structural context, with similar visual representations for flow data as for general exploration. Visual representations also often use heatmap overlays and/or glyphs of the same styles mentioned for exploration, e.g., streamlines or arrows, to indicate flow velocities at particular points. This allows for feature quantification directly from imaging data [TSS*08; HPT*08]. Experts also often wish to quantify and compare flow rates in simulations relative to time-resolved imaging data [LGH*19].

Analytically-focused tools and methods developed from collaborations between domain experts and the visualization community often take more experimental or abstracted approaches to aid analysis tasks. For example, Angelelli et al. [AH11] flatten 3D tubular flow to 2D to compare flow patterns over time, as shown in Fig. 16. Semi-automatic classification and clustering methods are also common to aid expert identification and comparison of vortices and shapes in the data [OLK*14; ERH16; MBP*16; OCJP16]. Interactive linked views, particularly when including simulated and acquired data, can help experts to better evaluate hemodynamic patterns and model accuracy [LNHL20]. Multi-view visual analysis tools often rely on an interplay between exploration and analysis for users to get a sense of the data and browse interesting regions before identifying and comparing features to understand, e.g., aneurysm rupture risk [MWPL20; MVG*21].

Approaches from the research community to present blood flow in education for a more general audience are limited. These methods may hint at the original flow data through animation but instantiate red blood cells to indicate to a broader audience what the flow represents [GMF*21], or employ more fanciful metaphors to show the passage of blood in a cardiac cycle [CHV*14].

Summary. Despite the extensive work on hemodynamics, further challenges and opportunities remain. The ultimate aim of many of these works is to make visualization of hemodynamics available in a clinical setting to aid in rapid and accurate identification of life-threatening flow behaviors. Studies that assess the possibility of real adoption of these techniques in clinical routine are an interesting avenue to explore. Communicating these data then to patients, in a way that is both understandable and minimally-alarming, is essential and remains an open challenge in visualization.

8.2. Heart Function

Heart function is well-characterized in physiology and visualization research. In this section, we focus on visualization related to (1) the mechanics of the heart as a pump and (2) the cardiac conduction system of the heart, which is an electrical network that controls heart rate and rhythm [HG11]. Diseases related to these aspects of heart function include (1) myocardial ischemia, where the heart tissue does not receive enough blood from its supplying arteries, (2) heart failure, where the heart is unable to pump blood effectively, and (3) atrial fibrillation, a dysfunction of the cardiac conduction system that leads to irregular heart rate and rhythm. These diseases provide strong clinical motivation and drive many of the visualization use cases in this topic.

Related Surveys. Nazir et al. [NKA*19] survey the visualization of various aspects of heart function from the medical domain, focusing mainly on analysis for use clinical routine. Walton et al. [WBT*14] provide a broad overview of the methods and challenges in visualizing cardiovascular magnetic resonance imagery for clinical research before presenting a prototype approach for visualizing this type of data. Generation of surface models and volume renderings of the heart, paired with time-lapse video to describe deformation, are key visualization techniques for this topic. Heatmap visualizations commonly indicate parameters of interest, and the bull's eye plot is ubiquitous for the visual analysis of perfusion data to understand heart function.

Exploration-centered visualizations provide an overview to experts of general features and parameters related to shape changes, e.g., for specific chambers, valves, or the entire heart, in a cardiac cycle. These works visualize phases of the cardiac cycle from simulations on surface models (which typically are abstracted from acquired data), such as the LFX Virtual Cardiac Model [JJDS04] or the constrained Multi-linear Shape Model [IGV*10]. Patient-specific approaches visualize models in combination with acquired data [WWZS10; ANL*16], or only acquired data. Some works build predictive models, and preserve links between the simulation and the original data to understand the mapping procedure [SLS09]. Although we discuss blood flow extensively in Sec. 8.1, for completeness, we note that several approaches visualize patterns of blood flow to explore questions related to heart function, e.g., Kulp et al. [KMQ*11].

Approaches geared towards a combination of exploration and analysis, or focused more purely on analysis, often favor colormaps applied to mesh or imaging data to quantify parameters of interest. Rainbow colormaps are common, especially in the medical domain. These works often use multiple interactive views to link simulation, imaging, and derived statistical information. The main purpose of the visualization is to evaluate particular parameters, e.g., strain rate [HSTS98; PP00]. Evaluating the movement of particular landmarks can be aided by additional plots, such as parallel coordinates and heatmaps, to compare different motion parameters [THS*17], as shown in Fig. 17.

Perfusion data, which we visited previously in Sec. 7.2, are often used as a basis to determine heart functionality. Most use cases are tied to experts with a need to identify and quantify particular tissue properties in the context of a pathology, and while exploration is a component, it is usually not the main task. Many approaches incorporate a bull's eye plot in cases related to myocardial ischemia, where heart tissue does not receive the blood and nutrients it needs to function. This visualization is familiar to clinicians to quantify the extent of the damage to the heart tissue, often alongside structural representations of the coronary arteries [OGHP06; TBB*07; TBB*08; RBPJ16]. The bull's eye plot can be further adapted for targeted visual analysis of the motion of the left ventricle of the heart over time [SCK*16]. Approaches are often interactive to provide an exploratory element for the user. Glyphs can be incorporated, as by Meyer-Spradow et al. [MSD*08], to quantify local tissue perfusion across the whole heart. Statistical, e.g., PCA, and aggregate measures may be used to reduce the complexity of the data [ODH*07] and include representations of uncertainty [RBPJ16].

Visualization is also used to evaluate the accuracy of simulations against acquired data, where heatmaps [MBK*10] or juxtaposed line plots [HOR07] indicate shape prediction accuracy. Other methods focus heavily on patient-centered care and outcomes [XSZ*16], such as simulations of surgical procedures, e.g., mitral valve clipping, with quantitative evaluation through heatmaps to help predict patient outcomes [MVM*11].

Some heart simulations have been developed, not only for expert exploration, but for use in education and surgical training. These include Dayan et al.‘s [DOS*04] 3D animation of the dynamics of a simulated mitral valve and the virtual reality (VR) simulation of radio frequency ablation by Pernod et al. [PSRD10]. Pernod et al.‘s approach uses a heatmap to show membrane propagation potential, a tissue-scale process, on the heart surface. VR has also been used for patients in a biofeedback scenario to manage stress [GWZE18].

Summary. While full cardiac models of heart function for detailed exploration and analysis are of high interest to the medical community, visualization research efforts to aid these tasks are relatively limited in comparison to the efforts dedicated to blood flow. In the analysis of perfusion data for whole-heart pathologies, a higher volume of works focus on solving tasks for clinicians. However, studies on their actual adoption and utility in a clinical setting are limited. Communication-focused works that visualize heart function, while numerous in the field of medical illustration, remain a comparatively limited topic in visualization. A number of recent works to visualize heart function rely on multiscale models. We discuss these works in Sec. 9.

8.3. Lung Function

The main function of breathing, i.e., respiration, is to provide oxygen to the tissues in the body and to remove carbon dioxide [HG11]. We focus here on the in- and outflow of air from the lungs and on limited cases where research includes other organs affected by lung movements.

Experts interested in learning about the features of lung deformation during breathing often rely on simulations, commonly from statistical modeling, to visualize this process [SG05; SID*08; KLO10; EWSH11; WWLS13]. These generally integrate with imaging data to provide spatial context, often via surface modeling or volume rendering techniques. More recent approaches have used neural networks to reconstruct lung deformations as surface meshes from 4D CT data [WZH19].

Experts are also interested in understanding patterns and features of dynamic airflow. This interest may be in visualizing airflow patterns within bronchial tubes [STM08] or on a larger scale. Kim et al. [KBV*15] present a coupled model of tissue deformation on the level of the whole lungs alongside network airflow, enabling predictions of various dynamic flow properties. The model is multiscale, but, as shown in Fig. 18, does not necessarily provide spatial visualization of cell-level air exchange and lacks visual interactivity. Wiechert et al. [WCRW11] couple tissue- and organ-scale processes to allow visual exploration of tissue regions locally and the whole lung and airway system globally.

In more analytically-focused cases, visualization enables the comparison of movements of an object of interest against an acquired signal [LCPS10]. Similar to heart function, visualization is also used to evaluate model accuracy against acquired data and to assess and compare the magnitude of lung deformation using heatmaps [NTC*19].

Respiratory visualization models, in addition to parameter exploration, may also be used to compare characteristics of air flow in healthy versus diseased patients [KBV*15]. In other instances, experts use lung function to indirectly provide information in the analysis of other organs. For example, breathing exerts force on the kidneys. The degree that the kidneys are compressed during the breath cycle can be used to evaluate kidney fibrosis [SHHB16].

Summary. The bulk of literature we found from visualization and domain research focuses heavily on expert exploration of airflow and lung deformation over breath cycles, with limited tools for analysis and even more limited work in visualizing lung function for communication. These works primarily come from outside the visualization community and represent an open opportunity to develop visual methods to support experts in better understanding and analyzing lung function. These tasks are particularly important with the advent of COVID-19, as experts work to understand the long-term effects of this disease on lung function.

8.4. Brain Function

The brain is part of the central nervous system that contains more than 100 billion neurons. It is the primary seat of control for any process occurring within our bodies. Understanding brain function provides a key to understanding human behavior as well as neurological diseases and disorders. While our discussion of signal propagation in Sec. 7 focused on the propagation of action potentials between cells, we now discuss signal propagation and functional neural connections over the entire brain. This is known as functional connectomics, where the brain is modeled as a network [STK05].

Related Surveys. Margulies et al. [MBWG13] and Pfister et al. [PKB*14] provide an overview of ways that the human connectome, both structural and functional, can be visualized for different exploratory, analytical, and communication tasks. Node-link diagrams, scatter plots, dendrograms, and heatmaps are common techniques in the application domain to visualize synchronous activity between brain regions. Structural models often provide spatial context for functional connectivity and can be depicted as image slices, surface models, volume renderings, or, in the case of DTI data, through advanced techniques that include ellipsoid or brushstroke glyphs [LAK*98] and superquadric glyphs [Kin04].

Experts are interested in learning how different regions of the brain functionally connect and in exploring patterns of brain activation in response to the presence, or absence, of certain stimuli. This represents a broader-scope view of the questions experts have when studying signal propagation. Heatmaps superimposed onto imaging data or derived surface models to show activation regions is a common approach that allows experts to explore and evaluate functional imaging data [FC00; WVE*03; RTF*06]. Interactive exploratory methods include dynamic querying for structural and functional connectivity using DTI and fMRI data [SAM*05].

Visual analysis methods help experts identify functional parameters and connections of interest. Standard methods often use a correlation matrix to identify functional connectivity [PKB*14]. Interactive analysis approaches often incorporate linked views that incorporate structural data alongside plots containing additional functional information, e.g., time plots [JNM*09; Lun10] or radar plots [MMH*13]. Color-coded isosurfaces from structural imaging, with network and scatter plots to show functional connectivity and correlations [DVF*10], are another example of interactive visual analysis interfaces. These interactive, multi-view approaches may facilitate hypothesis generation in addition to confirmative analysis and integrate several data types and plots [JBF*19]. Clustering methods can group structural fibers from DTI data into functionally-meaningful bundles. These can then be color-coded to aid identification and comparison of functional groups [GGZ*13], or to identify and compare resting state networks that are again color-coded [VMH08]. Other classification approaches to aid analysis use contrast subgraphs, shown in Fig. 19, as a primary means to compare between groups [LBG20].

While the bulk of visualization research for brain function is targeted at experts, some works that use illustrative techniques can be used for communication to a broader audience. The work by Jainek et al. [JBB*08] exemplifies such an instance for their use of glows and a soft, harmonious color palette to show brain activity.

Summary. Visualizing brain function is challenging. This is owed to the high computational power needed to simulate activity over an entire brain, and the fact that many visualizations of brain activity are driven by imaging data that only indirectly indicate brain activity and function. An additional challenge is that functional networks and structural connections do not always overlap. Visualization research to depict these uncertainties can aid neuroscience researchers. Furthermore, tools tend to focus on tasks related to exploration and analysis for domain experts. The development of visual methods targeted toward clinical rather than research use to identify aberrant patterns of brain function is an ongoing challenge. Finally, as previously discussed, further research into communication-oriented approaches that facilitate doctor-patient communication and patient understanding are essential for raising the bar of health literacy and public health. For example, data-driven approaches can communicate public safety stories, such as the impact of alcohol on brain function when driving a vehicle.

8.5. Other Organ Function

Although we focus our survey mainly on blood flow and functions related to the heart, lungs, and brain, we briefly highlight other organ functions that are representative of further opportunities in visualization research for physiology.

Skeletal muscle plays a crucial role in body movement and defining anatomical shape. Lee et al. [LGK*12] review visual approaches, which often use 3D surface models, for modeling muscle deformation and simulation of skeletal muscle functions at this scale. Experts are chiefly interested in understanding the shape deformation that skeletal muscle undergoes during contraction and relaxation [BHGG12]. Visualizations represent muscles at different degrees of abstraction to serve different objectives in a simulation, e.g., a single action line to show the axis of movement [NT98] or reconstructing and simulating a subset of muscle fibers that capture the overall shape of the muscle as it deforms [KK14; KČ20; RMS20]. Visual analysis approaches are often interested in quantifying muscle properties during contraction, such as muscle stiffness [SSG*10] or muscle speed in contraction [ANB*03]. These data can be captured in multi-view visualizations that combine structural models with line plots that describe displacement, velocity, and acceleration over the course of the simulation [YGV*13].

The stomach and liver are part of the gastrointestinal system whose main functions are (1) to take in food and liquids and break them down into a usable form and (2) to remove waste from the body. The stomach is a highly elastic organ that serves as a temporary holding place for food and is responsible for its initial breakdown. To understand the stomach's changing dimensions over time, Gilja et al. [GHØB96] present a multi-view system that includes a planar map of a region of the stomach for experts to better understand these temporal changes. The liver is the centerpiece for many essential bodily functions, including blood and nutrient storage, with a complex physiology that could greatly benefit from visualization. Lin et al. [LJH04] visualize fluid transport through the liver and its vasculature at the organ and tissue scales, using simulations in combination with imaging and animation techniques. Their aim is to understand how the liver absorbs and metabolizes substances.

Metabolic activity, as discussed in the context of molecular pathways in Sec. 5.3, can be visualized on the organ level to assess for normal organ function. Approaches can be analytically-focused, as in Nguyen et al.'s [NEO*10] method to visualize uncertainty from PET kinetic modeling. Ropinski et al. [RVB*09] exemplify a more exploratory approach to visualizing organ-level metabolic activity. Their approach is also designed to facilitate doctor-doctor communication through interactive closeups, whereby users can adjust view layout and composition to best fit their communication agenda.

Summary. Organs function as the result of a chain of processes that begin at the molecular level and extend through the cell and tissue scales. While blood flow, heart, and brain function are especially well-covered in visualization research, the lungs and other organs that we briefly highlighted in this section have received less attention, although experts clearly benefit from visualization tools to aid their exploratory and analytical questions. As in other scales, we observe a comparative lack of visualization research oriented to communication tasks. This represents an open opportunity for future work. Furthermore, while visualizations of organ-scale processes are often highly clinically-motivated, relatively limited research investigates visual methods to aid practicing clinicians that are usable in a time-crunched environment.