Categorical Descriptors Research Articles

CatEmbed: A Machine-Learned Representation Obtained via Categorical Entity Embedding for Predicting Adsorption and Reaction Energies on Bimetallic Alloy Surfaces.

Machine-learning models for predicting adsorption energies on metallic surfaces often rely on basic elemental properties and electronic and geometric descriptors. Here, we apply categorical entity embedding, a featurization method inspired by natural language processing techniques, to predict adsorption energies on bimetallic alloy surfaces using categorical descriptors. Using this method, we develop a machine-learned representation from categorical descriptors (e.g., surface composition, adsorbate type, and site type) of the slab/adsorbate complex. By combining this representation with numerical features (e.g., slab metal stoichiometric ratios), we create the CatEmbed representation. Remarkably, decision tree models trained using CatEmbed, which includes no explicit geometric information, achieve a Mean Absolute Error (MAE) of 0.12 eV. Additionally, we extend this technique to predict reaction energies on bimetallic surfaces, creating the CatEmbed-React representation, which achieves an MAE of 0.08 eV. These findings highlight the effectiveness of categorical entity embedding for predicting adsorption and reaction energies on bimetallic alloy surfaces.

Infants can use temporary or scant categorical information to individuate objects

In a standard individuation task, infants see two different objects emerge in alternation from behind a screen. If they can assign distinct categorical descriptors to the two objects, they expect to see both objects when the screen is lowered; if not, they have no expectation at all about what they will see (i.e., two objects, one object, or no object). Why is contrastive categorical information critical for success at this task? According to the kind account, infants must decide whether they are facing a single object with changing properties or two different objects with stable properties, and access to permanent, intrinsic, kind information for each object resolves this difficulty. According to the two-system account, however, contrastive categorical descriptors simply provide the object-file system with unique tags for individuating the two objects and for communicating about them with the physical-reasoning system. The two-system account thus predicts that any type of contrastive categorical information, however temporary or scant it may be, should induce success at the task. Two experiments examined this prediction. Experiment 1 tested 14-month-olds (N = 96) in a standard task using two objects that differed only in their featural properties. Infants succeeded at the task when the object-file system had access to contrastive temporary categorical descriptors derived from the objects’ distinct causal roles in preceding support events (e.g., formerly a support, formerly a supportee). Experiment 2 tested 9-month-olds (N = 96) in a standard task using two objects infants this age typically encode as merely featurally distinct. Infants succeeded when the object-file system had access to scant categorical descriptors derived from the objects’ prior inclusion in static arrays of similarly shaped objects (e.g., block-shaped objects, cylinder-shaped objects). These and control results support the two-system account’s claim that in a standard task, contrastive categorical descriptors serve to provide the object-file system with unique tags for the two objects.

Trait-mediated processes and per capita contributions to ecosystem functioning depend on conspecific density and climate conditions

The ecological consequences of environmental change are highly dependent on the functional contributions of the surviving community, but categorical descriptors commonly used to project ecosystem futures fail to capture context dependent response variability. Here, we show that intraspecific variability for a range of sediment-dwelling marine invertebrates is moderated by changes in the density of conspecifics and/or climatic conditions. Although these trait-mediated changes result in modifications to ecosystem properties, we find that the contributions of individuals to functioning are not necessarily additive but, instead, are a result of alterations to per capita performance. Our findings also indicate that trait variation within species can exert a greater influence on functioning than that of trait variation between species. Hence, projections of likely functional outcomes that scale from mean trait values are unlikely to be robust, highlighting a need to account for how and when intraspecific variability results in context-dependent community responses to change.

Dental Dynamics: A Fast New Tool for Quantifying Tooth and Jaw Biomechanics in 3D Slicer.

Teeth reveal how organisms interact with their environment. Biologists have long looked at the diverse form and function of teeth to study the evolution of feeding, fighting, and development. The exponential rise in the quantity and accessibility of computed tomography (CT) data has enabled morphologists to study teeth at finer resolutions and larger macroevolutionary scales. Measuring tooth function is no easy task, in fact, much of our mechanical understanding is derived from dental shape. Categorical descriptors of tooth shape such as morphological homodonty and heterodonty, overlook nuances in function by reducing tooth diversity for comparative analysis. The functional homodonty method quantitatively assesses the functional diversity of whole dentitions from tooth shape. This method uses tooth surface area and position to calculate the transmission of stress and estimates a threshold for functionally homodont teeth through bootstrapping and clustering techniques. However, some vertebrates have hundreds or thousands of teeth and measuring the shape and function of every individual tooth can be a painstaking task. Here, we present Dental Dynamics, a module for 3D Slicer that allows for the fast and precise quantification of dentitions and jaws. The tool automates the calculation of several tooth traits classically used to describe form and function (i.e., aspect ratio, mechanical advantage, force, etc.). To demonstrate the usefulness of our module we used Dental Dynamics to quantify 780 teeth across 20 salamanders that exhibit diverse ecologies. We coupled these data with the functional homodonty method to investigate the hypothesis that arboreal Aneides salamanders have novel tooth functions. Dental Dynamics provides a new and fast way to measure teeth and increases the accessibility of the functional homodonty method. We hope Dental Dynamics will encourage further theoretical and methodological development for quantifying and studying teeth.

Qualitative validation of the modified Medical Research Council (mMRC) dyspnoea scale as a patient-reported measure of breathlessness severity

IntroductionThe modified Medical Research Council (mMRC) dyspnoea scale is a measure of breathlessness severity recommended by guidelines and utilised as an inclusion criterion or endpoint for clinical trials. No studies have been conducted to validate the categorical descriptors against the dyspnoea severity grade. MethodsThis study utilised cognitive interviews (Think Aloud method) to assess the content validity of the mMRC scale among 16 participants (13 with cardiac/respiratory disease). Participants were recruited to achieve representation across a variety of demographic factors. Interviews were conducted remotely via video conferencing and participants were presented with all 5 mMRC descriptors on screen in random order then asked to rank the statements “in order from the best breathing to the worst breathing”. ResultsMean age of participants was 57 years (range 22–84 years). Eleven had multimorbidity (≥2 comorbidities) including COPD, asthma, lung cancer, lung infection, interstitial lung disease, heart failure, depression, and anxiety. Length of time with breathlessness ranged between 2 weeks and >25 years. The median rank of the mMRC grade descriptors was concordant for mMRC grades 0, 1 and 4 but not grades 2 and 3. Even so, substantial heterogeneity was found in the distribution of responses for mMRC grade 0. ConclusionOur study found substantial heterogeneity in participant grading of the mMRC descriptors, particularly for grades 0, 2 and 3, indicating that mMRC might not be a good discriminator of difference or change in dyspnoea severity. This study demonstrates the importance of content validation even for long-established PROs like mMRC.

A Computational Probe into the Behavioral and Neural Markers of Atypical Facial Emotion Processing in Autism.

Despite ample behavioral evidence of atypical facial emotion processing in individuals with autism spectrum disorder (ASD), the neural underpinnings of such behavioral heterogeneities remain unclear. Here, I have used brain-tissue mapped artificial neural network (ANN) models of primate vision to probe candidate neural and behavior markers of atypical facial emotion recognition in ASD at an image-by-image level. Interestingly, the image-level behavioral patterns of the ANNs better matched the neurotypical subjects 'behavior than those measured in ASD. This behavioral mismatch was most remarkable when the ANN behavior was decoded from units that correspond to the primate inferior temporal (IT) cortex. ANN-IT responses also explained a significant fraction of the image-level behavioral predictivity associated with neural activity in the human amygdala (from epileptic patients without ASD), strongly suggesting that the previously reported facial emotion intensity encodes in the human amygdala could be primarily driven by projections from the IT cortex. In sum, these results identify primate IT activity as a candidate neural marker and demonstrate how ANN models of vision can be used to generate neural circuit-level hypotheses and guide future human and nonhuman primate studies in autism.SIGNIFICANCE STATEMENT Moving beyond standard parametric approaches that predict behavior with high-level categorical descriptors of a stimulus (e.g., level of happiness/fear in a face image), in this study, I demonstrate how an image-level probe, using current deep-learning-based ANN models, allows identification of more diagnostic stimuli for autism spectrum disorder enabling the design of more powerful experiments. This study predicts that IT cortex activity is a key candidate neural marker of atypical facial emotion processing in people with ASD. Importantly, the results strongly suggest that ASD-related atypical facial emotion intensity encodes in the human amygdala could be primarily driven by projections from the IT cortex.

A machine learning approach for predicting the empirical polarity of organic solvents

A neural network architecture was found to efficiently predict the empirical polarity parameter ET(30) using simple to compute and interpretable six quantum mechanical, topological and categorical descriptors.

An alternative approach to dimension reduction for pareto distributed data: a case study

Deep learning models are tools for data analysis suitable for approximating (non-linear) relationships among variables for the best prediction of an outcome. While these models can be used to answer many important questions, their utility is still harshly criticized, being extremely challenging to identify which data descriptors are the most adequate to represent a given specific phenomenon of interest. With a recent experience in the development of a deep learning model designed to detect failures in mechanical water meter devices, we have learnt that a sensible deterioration of the prediction accuracy can occur if one tries to train a deep learning model by adding specific device descriptors, based on categorical data. This can happen because of an excessive increase in the dimensions of the data, with a correspondent loss of statistical significance. After several unsuccessful experiments conducted with alternative methodologies that either permit to reduce the data space dimensionality or employ more traditional machine learning algorithms, we changed the training strategy, reconsidering that categorical data, in the light of a Pareto analysis. In essence, we used those categorical descriptors, not as an input on which to train our deep learning model, but as a tool to give a new shape to the dataset, based on the Pareto rule. With this data adjustment, we trained a more performative deep learning model able to detect defective water meter devices with a prediction accuracy in the range 87–90%, even in the presence of categorical descriptors.

Similarity Among Physical Phenomena Recognized on the Basis of the Classification of Existing Knowledge

This work presents the synthetical mathematical analysis of available knowledge regarding physical phenomena of research interest. The work is not focusing on providing the phenomena according to the physical laws but rather because of them, hence, it is grounded on the philosophically defined concept of "similarity", and progresses to the mathematical treatments of those factors and parameters that are involved into the similarity validation among physical phenomena.A critical validation regarding the effectiveness of such an approach was also performed, in order to conceptualize the relevance of the factors and parameters interactions as a potential control tool against engineering-based hypothesis. Such factors and parameters are generated through the description and delimitation of the system of interest. A "matrix" is used for the classification of the existing knowledge regarding this system. It is consisted of the categorical descriptors of the system in question and the levels of these descriptors.A mathematical analysis of this "matrix" supports that all the existent perceptions of a physical phenomenon constitute a four-dimensional vector space. Within this space, the concept of similarity allows for the definition on which of a specific non-linear mapping that might be applied to strictly classify the existing knowledge about the phenomenon in question.Similarity is used here to define the conditions and the constrains that this mapping must satisfy. In conclusion, the applicability of the suggested approach on an engineering approach regarding a physical problem, was also demonstrated in this study.

The effects of warming and soil chemistry on bacterial community structure in Arctic tundra soils

Soil microbial communities in the Arctic play a critical role in regulating the global carbon (C) cycle. Vast amounts of C are stored in northern high latitude soils, and rising temperatures in the Arctic threaten to thaw permafrost, making relatively inaccessible C sources more available for mineralization by soil microbes. Few studies have characterized how microbial community structure responds to thawing permafrost in the context of varying soil chemistries associated with contrasting tundra landscapes. We subjected active layer and permafrost soils from upland and lowland tundra sites on the North Slope of Alaska to a soil-warming incubation experiment and compared soil bacterial community profiles (obtained by 16S rRNA amplicon sequencing) before and after incubation. The influence of soil composition (characterized by mid-infrared [MIR] spectroscopy) on bacterial community structure and class abundance was analyzed using redundancy and correlation analyses. We found increased abundances of Alphaproteobacteria, Gammaproteobacteria, and Bacteroidetes [Sphingobacteriia] post incubation, particularly in permafrost soils. The categorical descriptors site and soil layer had the most explanatory power in our predictive models of bacterial community structure, highlighting the close relationship between soil bacteria and the soil environment. Specific soil chemical attributes characterizing the soil environments that were found to be the best predictors included MIR spectral bands associated with inorganic C, silicates, amide II (CN stretch), and carboxylics (C–O stretch), and MIR peak ratios representing C substrate quality. Overall, these results further characterize soil bacterial community shifts that may occur as frozen environments with limited access to C sources, as is found in undisturbed permafrost, transition to warmer and more C-available environments, as is predicted in thawing permafrost due to climate change.

Variability in in vivo studies: Defining the upper limit of performance for predictions of systemic effect levels

New approach methodologies (NAMs) for chemical hazard assessment are often evaluated via comparison to animal studies; however, variability in animal study data limits NAM accuracy. The US EPA Toxicity Reference Database (ToxRefDB) enables consideration of variability in effect levels, including the lowest effect level (LEL) for a treatment-related effect and the lowest observable adverse effect level (LOAEL) defined by expert review, from subacute, subchronic, chronic, multi-generation reproductive, and developmental toxicity studies. The objectives of this work were to quantify the variance within systemic LEL and LOAEL values, defined as potency values for effects in adult or parental animals only, and to estimate the upper limit of NAM prediction accuracy. Multiple linear regression (MLR) and augmented cell means (ACM) models were used to quantify the total variance, and the fraction of variance in systemic LEL and LOAEL values explained by available study descriptors (e.g., administration route, study type). The MLR approach considered each study descriptor as an independent contributor to variance, whereas the ACM approach combined categorical descriptors into cells to define replicates. Using these approaches, total variance in systemic LEL and LOAEL values (in log10-mg/kg/day units) ranged from 0.74 to 0.92. Unexplained variance in LEL and LOAEL values, approximated by the residual mean square error (MSE), ranged from 0.20 to 0.39. Considering subchronic, chronic, or developmental study designs separately resulted in similar values. Based on the relationship between MSE and R-squared for goodness-of-fit, the maximal R-squared may approach 55 to 73% for a NAM-based predictive model of systemic toxicity using these data as reference. The root mean square error (RMSE) ranged from 0.47 to 0.63 log10-mg/kg/day, depending on dataset and regression approach, suggesting that a two-sided minimum prediction interval for systemic effect levels may have a width of 58 to 284-fold. These findings suggest quantitative considerations for building scientific confidence in NAM-based systemic toxicity predictions.

Crop Origins and Phylo Food: A database and a phylogenetic tree to stimulate comparative analyses on the origins of food crops

AbstractMotivationKnowledge on the origins of food crops in the wild is abundant but fragmentary and unconsolidated, which hinders macroecological approaches to long‐standing questions on the phylogeography and history of crop origins.Main types of variable contained(a) Comprehensive checklist of crop species cultivated for food; (b) data on diverse continuous and categorical descriptors of antiquity of cultivation (earliest written records of cultivation and of domestication), organ harvested for primary use, and growth form; (c) identities, geographic distribution and current climate at the origin of crops’ wild progenitors; (d) a time‐calibrated molecular phylogeny (Phylo Food v 1.0) of food crops.Spatial location and grainGlobal. Spatial accuracy of geographic records for wild progenitors varies among Global Biodiversity Information Facility (GBIF) source databases. WorldClim data were retrieved at 2.5 min (approximately 4.5 km at the equator) spatial resolution.Time period and grainTemporal accuracy of estimation of cultivation antiquity varies among sources, but unit of measurements is years ago. The scale of the phylogenetic tree is millions of years ago.Major taxa and level of measurementFlowering plants. Specifically, 866 species of food crops and 901 species of wild progenitors. Most information was retrieved at the species level. However, geographic occurrence of wild progenitors was obtained from GBIF records of observations of individuals of a given species in the field.Software formatCrop Origins v 1.0 and Phylo Food v 1.0 are provided as zipped files in the Supporting Information of this paper. The zipped files contain a. csv file with v 1.0 of the Crop Origins database, a. tre file with Newick format as Phylo Food v 1.0, metadata and literature sources files for both resources, and R code used to build Phylo Food v 1.0. The resources are also deposited, and will be regularly updated, at https://rubenmilla.github.io/Crop_Origins_Phylo.

A data analysis of the Chilean housing stock and the development of modelling archetypes

Chile is a South American country experiencing greater social and economic development than the majority of its neighbours and whose economy is transitioning from developing to developed status. It is committed to reduce its greenhouse gas (GHG) emissions by 35–45% by 2030, requiring national energy demand reduction. Chile's housing stock is responsible for 15% of its total final energy consumption and so its Government is regulating the construction of new dwellings, although it is difficult to know if this policy will succeed or lead to unintended consequences affecting energy demand, GHG emissions, and occupant health. Measuring the effects in situ is often time and cost prohibitive and so the simulation of archetypal buildings is a common method of investigation. A range of data sources, such as censuses and building permits, are used to develop archetypal Chilean houses with statistically significant representative values for design parameters related to energy demand and indoor air quality. It finds that 496 archetypes can represent 100% of the Chilean housing stock and only 90 can represent 95% of the stock. The archetypes can be used to predict and evaluate the impacts of policies on indoor air quality and the energy demand of a stock of houses, or to guide future data gathering exercises. The data analysis highlights knowledge gaps in categorical descriptors and occupant behaviours, and poor granularity of physical data. These gaps should be filled by augmenting national surveys and complimented by field work.

Classification of Infant Vocalizations by Untrained Listeners.

Purpose To better enable communication among researchers, clinicians, and caregivers, we aimed to assess how untrained listeners classify early infant vocalization types in comparison to terms currently used by researchers and clinicians. Method Listeners were caregivers with no prior formal education in speech and language development. A 1st group of listeners reported on clinician/researcher-classified vowel, squeal, growl, raspberry, whisper, laugh, and cry vocalizations obtained from archived video/audio recordings of 10 infants from 4 through 12 months of age. A list of commonly used terms was generated based on listener responses and the standard research terminology. A 2nd group of listeners was presented with the same vocalizations and asked to select terms from the list that they thought best described the sounds. Results Classifications of the vocalizations by listeners largely overlapped with published categorical descriptors and yielded additional insight into alternate terms commonly used. The biggest discrepancies were found for the vowel category. Conclusion Prior research has shown that caregivers are accurate in identifying canonical babbling, a major prelinguistic vocalization milestone occurring at about 6-7 months of age. This indicates that caregivers are also well attuned to even earlier emerging vocalization types. This supports the value of continuing basic and clinical research on the vocal types infants produce in the 1st months of life and on their potential diagnostic utility, and may also help improve communication between speech-language pathologists and families.

Multi-context Knowledge Base using Calculated Descriptors from Xper3: the Archaeocyaths Knowledge Base example

Natural sciences need to make assertions about characteristics of taxa. Traits and qualities or descriptors and states become increasingly crucial as a resource for identification adapted to both scientists and the public. Specialists, non-specialists, and the general public need different strategies for accessing the information. Creating a knowledge base is time-consuming and adapting this base to several needs seems to increase the required time substantially. Specialists think that an identification tool requires a complete overhaul when they want to change the target audience or the language... so they do not wish to get involved. How to minimize creation and update time? How to aggregate data only once in order to create a single knowledge base with different access levels? Strategies for integrating different patterns of descriptors in a single knowledge base are useful to modulate descriptions without loss of information and with the certainty that everything is up to date in each context. This multi-context knowledge base, derived from a single trait dataset, can generate descriptions adapted to different contexts and users. To address this issue, we propose to use calculated descriptors in order to have on the same knowledge base, different versions of descriptors that are updated automatically when the reference trait is modified. Calculated descriptors are a distinctive feature of Xper3. These descriptors are automatically computed from other descriptors by using logical operators (boolean operators). Xper3 (http://www.xper3.com) is a web platform that manages descriptive data and provides interactive identification keys. Xper3 and its previous version, Xper2, have been used already for various taxonomic groups. We will focus on fossils in order to reveal how calculated descriptors in Xper3 knowledge bases can solve the multi-context problem. The main source of content is the archaeocyaths knowledge base (http://archaeocyatha.infosyslab.fr). Archaeocyaths are the first animals to build reefs during the Cambrian. They are calcified sponges without spicules. The archaeocyaths knowledge base is an efficient resource for scientific studies and a useful tool for non-specialists, especially with the support of calculated descriptors. Correspondence between archaeocyath and sponge morphologies is not ready yet, but everything will be included into PORO (The Porifera Ontology, http://purl.obolibrary.org/obo/poro/releases/2014-03-06/) in the short term (an anatomy ontology about sponges). In this knowledge base, calculated descriptors are used to: create a consistent multilingual interactive identification key (French and English are available and Russian is in draft), generate descriptors adapted to different level of expertise and reword morphological descriptors (adapted for identification) into homologous characters (adapted for phylogeny). create a consistent multilingual interactive identification key (French and English are available and Russian is in draft), generate descriptors adapted to different level of expertise and reword morphological descriptors (adapted for identification) into homologous characters (adapted for phylogeny). Xper2 and Xper3 are compatible with TDWG’s Structured Descriptive Data (SDD) format. Calculated descriptors do not exist in SDD format and so they are exported from Xper3 as categorical descriptors, therefore losing the origin of values. Calculated descriptors are powerful and we are interested in discussing these with SDD and Xper3 users in order to improve the user interface and develop new tools for the analysis of such descriptors.

A cross-cultural survey of umami familiarity in European countries

Although scientists have known about the umami taste for decades, it is only recently that it has gradually gained wider public recognition as the fifth primary taste. Umami is elicited by l-glutamate, some amino acids and purine nucleotides. Glutamate is found in a wide range of foods but despite that, this taste is not generally included in taste evaluation tests in European countries because it has been found to be hard to conceptualize by the population. To the best of our knowledge, a comprehensive survey on umami familiarity in European countries has not been conducted. Therefore, we chose three countries representative of northern (Finland), central (Germany) and southern Europe (Italy) for our study. Each group contained 300, 271 and 252 participants, respectively. We collected the categorical descriptors spontaneously expressed by volunteers after tasting an umami solution alone, next to the hedonic value perceived in comparison with a watery salt solution and pure water. A significant difference in the hedonic response by country was found. The Finnish sample group used the correct word “umami” at a higher rate (15%) than Italians and Germans (both at 2%). Finland also showed better discrimination between the monosodium glutamate (umami) and sodium chloride (cooking salt) solutions. Overall, the umami taste was rated less pleasant than the salt, females preferring umami more than males, while it was the other way around for salt. Interestingly, a similar percentage of individuals with very low sensitivity for both umami and salt was detected in the countries.

Health and disease phenotyping in old age using a cluster network analysis

Human ageing is a complex trait that involves the synergistic action of numerous biological processes that interact to form a complex network. Here we performed a network analysis to examine the interrelationships between physiological and psychological functions, disease, disability, quality of life, lifestyle and behavioural risk factors for ageing in a cohort of 3,270 subjects aged ≥55 years. We considered associations between numerical and categorical descriptors using effect-size measures for each variable pair and identified clusters of variables from the resulting pairwise effect-size network and minimum spanning tree. We show, by way of a correspondence analysis between the two sets of clusters, that they correspond to coarse-grained and fine-grained structure of the network relationships. The clusters obtained from the minimum spanning tree mapped to various conceptual domains and corresponded to physiological and syndromic states. Hierarchical ordering of these clusters identified six common themes based on interactions with physiological systems and common underlying substrates of age-associated morbidity and disease chronicity, functional disability, and quality of life. These findings provide a starting point for indepth analyses of ageing that incorporate immunologic, metabolomic and proteomic biomarkers, and ultimately offer low-level-based typologies of healthy and unhealthy ageing.

Trait patterns of mungbean, black gram and rice bean

Food legumes like mungbean (Vigna radiata var. radiata), black gram (Vigna mungo), and rice bean (Vigna umbellata) have potential to diversify diets and agro-ecosystems of the tropics. Access to genetic diversity and phenotypic data are key issues for crop improvement. Large germplasm collections like the one at the World Vegetable Center may be used in detecting trait relationships and to identify accessions of interest for breeding. The current study provides results of more than 30 years of field characterization, including 9 numeric and 15 categorical descriptors and more than 10,000 accessions or sub-accessions. Core collections were established in black gram and rice bean. A positive correlation was detected between yield components such as seed size and pod length, and between these yield components and less desirable traits such as plant height and late maturity. High within-accession diversity was identified in several of the accessions.

Graphical Displays for Understanding SEM Model Similarity

Relationships between structural equation models are almost exclusively characterized as categorical. However, a simple yes–no answer as to whether 2 models are equivalent or nested (or other categorical descriptors of relationships) cannot address whether the models tend to fit similarly across a wide range of data, or how the characteristics of models and data affect the differences in fit. Therefore, such simple answers have limited value for model selection and evaluation. In this article we proposed a quantitative framework that allows for graphical display of results, incorporates the categorical distinctions in model similarity, is applicable to any model pairs, and offers a new method to generate data for studying model similarity. The framework is straightforward to implement and does not require complex analyses. The information this framework provides is diagnostic and can help researchers identify and explain how key characteristics in data and model parameters lead to (dis)similarity in fit.

Quantitative imaging: Correlating image features with the segmentation accuracy of PET based tumor contours in the lung

The purpose of this study was to investigate the correlation between image features extracted from PET images and the accuracy of manually drawn lesion contours in the lung. Such correlations are interesting in that they could potentially be used in predictive models to help guide physician contouring. In this work, 26 synthetic PET datasets were created using an anthropomorphic phantom and Monte Carlo simulation. Manual contours of simulated lesions were provided by 10 physicians. Contour accuracy was quantified using five commonly used similarity metrics which were then correlated with several features extracted from the images. Features were sub-divided into three groups using intensity, geometry, and texture as categorical descriptors. When averaged among the participants, the results showed relatively strong correlations with complexity and contrastI (r≥0.65, p<0.001), and moderate correlations with several other image features (r≥0.5, p<0.01). The predictive nature of these correlations was improved through stepwise regression and the creation of multi-feature models. Imaging features were also correlated with the standard deviation of contouring error in order to investigate inter-observer variability. Several features were consistently identified as influential including integral of mean curvature and complexity. These relationships further the understanding as to what causes variation in the contouring of PET positive lesions.