Diagnostic shifts in autism spectrum disorder can be linked to the fuzzy nature of the diagnostic boundary: a data-driven approach.

BACKGROUND: Diagnostic shifts at early ages may provide invaluable insights into the nature of separation between autism spectrum disorder (ASD) and typical development. Recent conceptualizations of ASD suggest the condition is only fuzzily separated from non-ASD, with intermediate cases between the two. These intermediate cases may shift along a transition region over time, leading to apparent instability of diagnosis. METHODS: We used a cohort of children with high ASD risk, by virtue of having an older sibling with ASD, assessed at 24 months (N = 212) and 36 months (N = 191). We applied machine learning to empirically characterize the classification boundary between ASD and non-ASD, using variables quantifying developmental and adaptive skills. We computed the distance of children to the classification boundary. RESULTS: Children who switched diagnostic labels from 24 to 36 months, in both directions, (dynamic group) had intermediate phenotypic profiles. They were closer to the classification boundary compared to children who had stable diagnoses, both at 24 months (Cohen's d = .52) and at 36 months (d = .75). The magnitude of change in distance between the two time points was similar for the dynamic and stable groups (Cohen's d = .06), and diagnostic shifts were not associated with a large change. At the individual level, a few children in the dynamic group showed substantial change. CONCLUSIONS: Our results suggested that a diagnostic shift was largely due to a slight movement within a transition region between ASD and non-ASD. This fact highlights the need for more vigilant surveillance and intervention strategies. Young children with intermediate phenotypes may have an increased susceptibility to gain or lose their diagnosis at later ages, calling attention to the inherently dynamic nature of early ASD diagnoses.

Distance-based analysis of variance for brain connectivity.

The field of neuroimaging dedicated to mapping connections in the brain is increasingly being recognized as key for understanding neurodevelopment and pathology. Networks of these connections are quantitatively represented using complex structures, including matrices, functions, and graphs, which require specialized statistical techniques for estimation and inference about developmental and disorder-related changes. Unfortunately, classical statistical testing procedures are not well suited to high-dimensional testing problems. In the context of global or regional tests for differences in neuroimaging data, traditional analysis of variance (ANOVA) is not directly applicable without first summarizing the data into univariate or low-dimensional features, a process that might mask the salient features of high-dimensional distributions. In this work, we consider a general framework for two-sample testing of complex structures by studying generalized within-group and between-group variances based on distances between complex and potentially high-dimensional observations. We derive an asymptotic approximation to the null distribution of the ANOVA test statistic, and conduct simulation studies with scalar and graph outcomes to study finite sample properties of the test. Finally, we apply our test to our motivating study of structural connectivity in autism spectrum disorder.

Bayesian regression-based developmental norms for the Benton Facial Recognition Test in males and females.

Face identity recognition is important for social interaction and is impaired in a range of clinical disorders, including several neurodevelopmental disorders. The Benton Facial Recognition Test (BFRT; Benton & Van Allen, 1968), a widely used assessment of identity recognition, is the only standardized test of face identity perception, as opposed to face memory, that has been normed on children and adolescents. However, the existing norms for the BFRT are suboptimal, with several ages not represented and no established time limit (which can lead to inflated scores by allowing individuals with prosopagnosia to use feature matching). Here we address these issues with a large normative dataset of children and adolescents (ages 5-17, N = 398) and adults (ages 18-55; N = 120) who completed a time-limited version of the BFRT. Using Bayesian regression, we demonstrate that face identity perception increases asymptotically from childhood through adulthood, and provide continuous norms based on age and sex that can be used to calculate standard scores. We show that our time limit of 16 seconds per item yields scores comparable to the existing norms without time limits from the non-prosopagnostic samples. We also find that females (N = 156) score significantly higher than males (N = 362), supporting the existence of a female superiority effect for face identification. Overall, these results provide more robust norms for the BFRT and promote future research on face identity perception in developmental populations.

System-level matching of structural and functional connectomes in the human brain.

The brain can be considered as an information processing network, where complex behavior manifests as a result of communication between large-scale functional systems such as visual and default mode networks. As the communication between brain regions occurs through underlying anatomical pathways, it is important to define a "traffic pattern" that properly describes how the regions exchange information. Empirically, the choice of the traffic pattern can be made based on how well the functional connectivity between regions matches the structural pathways equipped with that traffic pattern. In this paper, we present a multimodal connectomics paradigm utilizing graph matching to measure similarity between structural and functional connectomes (derived from dMRI and fMRI data) at node, system, and connectome level. Through an investigation of the brain's structure-function relationship over a large cohort of 641 healthy developmental participants aged 8-22 years, we demonstrate that communicability as the traffic pattern describes the functional connectivity of the brain best, with large-scale systems having significant agreement between their structural and functional connectivity patterns. Notably, matching between structural and functional connectivity for the functionally specialized modular systems such as visual and motor networks are higher as compared to other more integrated systems. Additionally, we show that the negative functional connectivity between the default mode network (DMN) and motor, frontoparietal, attention, and visual networks is significantly associated with its underlying structural connectivity, highlighting the counterbalance between functional activation patterns of DMN and other systems. Finally, we investigated sex difference and developmental changes in brain and observed that similarity between structure and function changes with development.

Whole brain white matter connectivity analysis using machine learning: An application to autism.

In this paper, we propose an automated white matter connectivity analysis method for machine learning classification and characterization of white matter abnormality via identification of discriminative fiber tracts. The proposed method uses diffusion MRI tractography and a data-driven approach to find fiber clusters corresponding to subdivisions of the white matter anatomy. Features extracted from each fiber cluster describe its diffusion properties and are used for machine learning. The method is demonstrated by application to a pediatric neuroimaging dataset from 149 individuals, including 70 children with autism spectrum disorder (ASD) and 79 typically developing controls (TDC). A classification accuracy of 78.33% is achieved in this cross-validation study. We investigate the discriminative diffusion features based on a two-tensor fiber tracking model. We observe that the mean fractional anisotropy from the second tensor (associated with crossing fibers) is most affected in ASD. We also find that local along-tract (central cores and endpoint regions) differences between ASD and TDC are helpful in differentiating the two groups. These altered diffusion properties in ASD are associated with multiple robustly discriminative fiber clusters, which belong to several major white matter tracts including the corpus callosum, arcuate fasciculus, uncinate fasciculus and aslant tract; and the white matter structures related to the cerebellum, brain stem, and ventral diencephalon. These discriminative fiber clusters, a small part of the whole brain tractography, represent the white matter connections that could be most affected in ASD. Our results indicate the potential of a machine learning pipeline based on white matter fiber clustering.

The impact of in-scanner head motion on structural connectivity derived from diffusion MRI.

Multiple studies have shown that data quality is a critical confound in the construction of brain networks derived from functional MRI. This problem is particularly relevant for studies of human brain development where important variables (such as participant age) are correlated with data quality. Nevertheless, the impact of head motion on estimates of structural connectivity derived from diffusion tractography methods remains poorly characterized. Here, we evaluated the impact of in-scanner head motion on structural connectivity using a sample of 949 participants (ages 8-23 years old) who passed a rigorous quality assessment protocol for diffusion magnetic resonance imaging (dMRI) acquired as part of the Philadelphia Neurodevelopmental Cohort. Structural brain networks were constructed for each participant using both deterministic and probabilistic tractography. We hypothesized that subtle variation in head motion would systematically bias estimates of structural connectivity and confound developmental inference, as observed in previous studies of functional connectivity. Even following quality assurance and retrospective correction for head motion, eddy currents, and field distortions, in-scanner head motion significantly impacted the strength of structural connectivity in a consistency- and length-dependent manner. Specifically, increased head motion was associated with reduced estimates of structural connectivity for network edges with high inter-subject consistency, which included both short- and long-range connections. In contrast, motion inflated estimates of structural connectivity for low-consistency network edges that were primarily shorter-range. Finally, we demonstrate that age-related differences in head motion can both inflate and obscure developmental inferences on structural connectivity. Taken together, these data delineate the systematic impact of head motion on structural connectivity, and provide a critical context for identifying motion-related confounds in studies of structural brain network development.

Harmonization of multi-site diffusion tensor imaging data.

Diffusion tensor imaging (DTI) is a well-established magnetic resonance imaging (MRI) technique used for studying microstructural changes in the white matter. As with many other imaging modalities, DTI images suffer from technical between-scanner variation that hinders comparisons of images across imaging sites, scanners and over time. Using fractional anisotropy (FA) and mean diffusivity (MD) maps of 205 healthy participants acquired on two different scanners, we show that the DTI measurements are highly site-specific, highlighting the need of correcting for site effects before performing downstream statistical analyses. We first show evidence that combining DTI data from multiple sites, without harmonization, may be counter-productive and negatively impacts the inference. Then, we propose and compare several harmonization approaches for DTI data, and show that ComBat, a popular batch-effect correction tool used in genomics, performs best at modeling and removing the unwanted inter-site variability in FA and MD maps. Using age as a biological phenotype of interest, we show that ComBat both preserves biological variability and removes the unwanted variation introduced by site. Finally, we assess the different harmonization methods in the presence of different levels of confounding between site and age, in addition to test robustness to small sample size studies.

Assessing connectivity related injury burden in diffuse traumatic brain injury.

Many of the clinical and behavioral manifestations of traumatic brain injury (TBI) are thought to arise from disruption to the structural network of the brain due to diffuse axonal injury (DAI). However, a principled way of summarizing diffuse connectivity alterations to quantify injury burden is lacking. In this study, we developed a connectome injury score, Disruption Index of the Structural Connectome (DISC), which summarizes the cumulative effects of TBI-induced connectivity abnormalities across the entire brain. Forty patients with moderate-to-severe TBI examined at 3 months postinjury and 35 uninjured healthy controls underwent magnetic resonance imaging with diffusion tensor imaging, and completed behavioral assessment including global clinical outcome measures and neuropsychological tests. TBI patients were selected to maximize the likelihood of DAI in the absence of large focal brain lesions. We found that hub-like regions, with high betweenness centrality, were most likely to be impaired as a result of diffuse TBI. Clustering of participants revealed a subgroup of TBI patients with similar connectivity abnormality profiles who exhibited relatively poor cognitive performance. Among TBI patients, DISC was significantly correlated with post-traumatic amnesia, verbal learning, executive function, and processing speed. Our experiments jointly demonstrated that assessing structural connectivity alterations may be useful in development of patient-oriented diagnostic and prognostic tools. Hum Brain Mapp 38:2913-2922, 2017. © 2017 Wiley Periodicals, Inc.

Automated tract extraction via atlas based Adaptive Clustering.

Advancements in imaging protocols such as the high angular resolution diffusion-weighted imaging (HARDI) and in tractography techniques are expected to cause an increase in the tract-based analyses. Statistical analyses over white matter tracts can contribute greatly towards understanding structural mechanisms of the brain since tracts are representative of connectivity pathways. The main challenge with tract-based studies is the extraction of the tracts of interest in a consistent and comparable manner over a large group of individuals without drawing the inclusion and exclusion regions of interest. In this work, we design a framework for automated extraction of white matter tracts. The framework introduces three main components, namely a connectivity based fiber representation, a fiber bundle atlas, and a clustering approach called Adaptive Clustering. The fiber representation relies on the connectivity signatures of fibers to establish an easy correspondence between different subjects. A group-wise clustering of these fibers that are represented by the connectivity signatures is then used to generate a fiber bundle atlas. Finally, Adaptive Clustering incorporates the previously generated clustering atlas as a prior, to cluster the fibers of a new subject automatically. Experiments on the HARDI scans of healthy individuals acquired repeatedly, demonstrate the applicability, reliability and the repeatability of our approach in extracting white matter tracts. By alleviating the seed region selection and the inclusion/exclusion ROI drawing requirements that are usually handled by trained radiologists, the proposed framework expands the range of possible clinical applications and establishes the ability to perform tract-based analyses with large samples.

PUNCH: Population Characterization of Heterogeneity.

Neuropsychiatric disorders are notoriously heterogeneous in their presentation, which precludes straightforward and objective description of the differences between affected and typical populations that therefore makes finding reliable biomarkers a challenge. This difficulty underlines the need for reliable methods to capture sample characteristics of heterogeneity using a single continuous measure, incorporating the multitude of scores used to describe different aspects of functioning. This study addresses this challenge by proposing a general method of identifying and quantifying the heterogeneity of any clinical population using a severity measure called the PUNCH (Population Characterization of Heterogeneity). PUNCH is a decision level fusion technique to incorporate decisions of various phenotypic scores, while providing interpretable weights for scores. We provide applications of our framework to simulated datasets and to a large sample of youth with Autism Spectrum Disorder (ASD). Next we stratify PUNCH scores in our ASD sample and show how severity moderates findings of group differences in diffusion weighted brain imaging data; more severely affected subgroups of ASD show expanded differences compared to age and gender matched healthy controls. Results demonstrate the ability of our measure in quantifying the underlying heterogeneity of the clinical samples, and suggest its utility in providing researchers with reliable severity assessments incorporating population heterogeneity.

Inequality-Constrained 3D Morphable Face Model Fitting.

3D morphable model (3DMM) fitting on 2D data is traditionally done via unconstrained optimization with regularization terms to ensure that the result is a plausible face shape and is consistent with a set of 2D landmarks. This paper presents inequality-constrained 3DMM fitting as the first alternative to regularization in optimization-based 3DMM fitting. Inequality constraints on the 3DMM's shape coefficients ensure face-like shapes without modifying the objective function for smoothness, thus allowing for more flexibility to capture person-specific shape details. Moreover, inequality constraints on landmarks increase robustness in a way that does not require per-image tuning. We show that the proposed method stands out with its ability to estimate person-specific face shapes by jointly fitting a 3DMM to multiple frames of a person. Further, when used with a robust objective function, namely gradient correlation, the method can work "in-the-wild" even with a 3DMM constructed from controlled data. Lastly, we show how to use the log-barrier method to efficiently implement the method. To our knowledge, we present the first 3DMM fitting framework that requires no learning yet is accurate, robust, and efficient. The absence of learning enables a generic solution that allows flexibility in the input image size, interchangeable morphable models, and incorporation of camera matrix.

Meta-evaluation for 3D Face Reconstruction Via Synthetic Data.

The standard benchmark metric for 3D face reconstruction is the geometric error between reconstructed meshes and the ground truth. Nearly all recent reconstruction methods are validated on real ground truth scans, in which case one needs to establish point correspondence prior to error computation, which is typically done with the Chamfer (i.e., nearest neighbor) criterion. However, a simple yet fundamental question have not been asked: Is the Chamfer error an appropriate and fair benchmark metric for 3D face reconstruction? More generally, how can we determine which error estimator is a better benchmark metric? We present a meta-evaluation framework that uses synthetic data to evaluate the quality of a geometric error estimator as a benchmark metric for face reconstruction. Further, we use this framework to experimentally compare four geometric error estimators. Results show that the standard approach not only severely underestimates the error, but also does so inconsistently across reconstruction methods, to the point of even altering the ranking of the compared methods. Moreover, although non-rigid ICP leads to a metric with smaller estimation bias, it could still not correctly rank all compared reconstruction methods, and is significantly more time consuming than Chamfer. In sum, we show several issues present in the current benchmarking and propose a procedure using synthetic data to address these issues.

Comparison of Human Experts and AI in Predicting Autism from Facial Behavior.

Advances in computational behavior analysis via artificial intelligence (AI) promise to improve mental healthcare services by providing clinicians with tools to assist diagnosis or measurement of treatment outcomes. This potential has spurred an increasing number of studies in which automated pipelines predict diagnoses of mental health conditions. However, a fundamental question remains unanswered: How do the predictions of the AI algorithms correspond and compare with the predictions of humans? This is a critical question if AI technology is to be used as an assistive tool, because the utility of an AI algorithm would be negligible if it provides little information beyond what clinicians can readily infer. In this paper, we compare the performance of 19 human raters (8 autism experts and 11 non-experts) and that of an AI algorithm in terms of predicting autism diagnosis from short (3-minute) videos of N = 42 participants in a naturalistic conversation. Results show that the AI algorithm achieves an average accuracy of 80.5%, which is comparable to that of clinicians with expertise in autism (83.1%) and clinical research staff without specialized expertise (78.3%). Critically, diagnoses that were inaccurately predicted by most humans (experts and non-experts, alike) were typically correctly predicted by AI. Our results highlight the potential of AI as an assistive tool that can augment clinician diagnostic decision-making.

Predicting Autism from Head Movement Patterns during Naturalistic Social Interactions.

Autism spectrum disorder (ASD) is a neurodevelopmental condition characterized in part by difficulties in verbal and nonverbal social communication. Evidence indicates that autistic people, compared to neurotypical peers, exhibit differences in head movements, a key form of nonverbal communication. Despite the crucial role of head movements in social communication, research on this nonverbal cue is relatively scarce compared to other forms of nonverbal communication, such as facial expressions and gestures. There is a need for scalable, reliable, and accurate instruments for measuring head movements directly within the context of social interactions. In this study, we used computer vision and machine learning to examine the head movement patterns of neurotypical and autistic individuals during naturalistic, face-to-face conversations, at both the individual (monadic) and interpersonal (dyadic) levels. Our model predicts diagnostic status using dyadic head movement data with an accuracy of 80%, highlighting the value of head movement as a marker of social communication. The monadic data pipeline had lower accuracy (69.2%) compared to the dyadic approach, emphasizing the importance of studying back-and-forth social communication cues within a true social context. The proposed classifier is not intended for diagnostic purposes, and future research should replicate our findings in larger, more representative samples.

Development of a phenotype ontology for autism spectrum disorder by natural language processing on electronic health records.

BACKGROUND: Autism spectrum disorder (ASD) is a complex neurodevelopmental condition characterized by restricted, repetitive behavior, and impaired social communication and interactions. However, significant challenges remain in diagnosing and subtyping ASD due in part to the lack of a validated, standardized vocabulary to characterize clinical phenotypic presentation of ASD. Although the human phenotype ontology (HPO) plays an important role in delineating nuanced phenotypes for rare genetic diseases, it is inadequate to capture characteristic of behavioral and psychiatric phenotypes for individuals with ASD. There is a clear need, therefore, for a well-established phenotype terminology set that can assist in characterization of ASD phenotypes from patients' clinical narratives. METHODS: To address this challenge, we used natural language processing (NLP) techniques to identify and curate ASD phenotypic terms from high-quality unstructured clinical notes in the electronic health record (EHR) on 8499 individuals with ASD, 8177 individuals with non-ASD psychiatric disorders, and 8482 individuals without a documented psychiatric disorder. We further performed dimensional reduction clustering analysis to subgroup individuals with ASD, using nonnegative matrix factorization method. RESULTS: Through a note-processing pipeline that includes several steps of state-of-the-art NLP approaches, we identified 3336 ASD terms linking to 1943 unique medical concepts, which represents among the largest ASD terminology set to date. The extracted ASD terms were further organized in a formal ontology structure similar to the HPO. Clustering analysis showed that these terms could be used in a diagnostic pipeline to differentiate individuals with ASD from individuals with other psychiatric disorders. CONCLUSION: Our ASD phenotype ontology can assist clinicians and researchers in characterizing individuals with ASD, facilitating automated diagnosis, and subtyping individuals with ASD to facilitate personalized therapeutic decision-making.

Head Movement Patterns during Face-to-Face Conversations Vary with Age.

Advances in computational behavior analysis have the potential to increase our understanding of behavioral patterns and developmental trajectories in neurotypical individuals, as well as in individuals with mental health conditions marked by motor, social, and emotional difficulties. This study focuses on investigating how head movement patterns during face-to-face conversations vary with age from childhood through adulthood. We rely on computer vision techniques due to their suitability for analysis of social behaviors in naturalistic settings, since video data capture can be unobtrusively embedded within conversations between two social partners. The methods in this work include unsupervised learning for movement pattern clustering, and supervised classification and regression as a function of age. The results demonstrate that 3-minute video recordings of head movements during conversations show patterns that distinguish between participants that are younger vs. older than 12 years with 78% accuracy. Additionally, we extract relevant patterns of head movement upon which the age distinction was determined by our models.

Calculating genetic risk for dysfunction in pleiotropic biological processes using whole exome sequencing data.

BACKGROUND: Numerous genes are implicated in autism spectrum disorder (ASD). ASD encompasses a wide-range and severity of symptoms and co-occurring conditions; however, the details of how genetic variation contributes to phenotypic differences are unclear. This creates a challenge for translating genetic evidence into clinically useful knowledge. Sleep disturbances are particularly prevalent co-occurring conditions in ASD, and genetics may inform treatment. Identifying convergent mechanisms with evidence for dysfunction that connect ASD and sleep biology could help identify better treatments for sleep disturbances in these individuals. METHODS: To identify mechanisms that influence risk for ASD and co-occurring sleep disturbances, we analyzed whole exome sequence data from individuals in the Simons Simplex Collection (n = 2380). We predicted protein damaging variants (PDVs) in genes currently implicated in either ASD or sleep duration in typically developing children. We predicted a network of ASD-related proteins with direct evidence for interaction with sleep duration-related proteins encoded by genes with PDVs. Overrepresentation analyses of Gene Ontology-defined biological processes were conducted on the resulting gene set. We calculated the likelihood of dysfunction in the top overrepresented biological process. We then tested if scores reflecting genetic dysfunction in the process were associated with parent-reported sleep duration. RESULTS: There were 29 genes with PDVs in the ASD dataset where variation was reported in the literature to be associated with both ASD and sleep duration. A network of 108 proteins encoded by ASD and sleep duration candidate genes with PDVs was identified. The mechanism overrepresented in PDV-containing genes that encode proteins in the interaction network with the most evidence for dysfunction was cerebral cortex development (GO:0,021,987). Scores reflecting dysfunction in this process were associated with sleep durations; the largest effects were observed in adolescents (p = 4.65 × 10-3). CONCLUSIONS: Our bioinformatic-driven approach detected a biological process enriched for genes encoding a protein-protein interaction network linking ASD gene products with sleep duration gene products where accumulation of potentially damaging variants in individuals with ASD was associated with sleep duration as reported by the parents. Specifically, genetic dysfunction impacting development of the cerebral cortex may affect sleep by disrupting sleep homeostasis which is evidenced to be regulated by this brain region. Future functional assessments and objective measurements of sleep in adolescents with ASD could provide the basis for more informed treatment of sleep problems in these individuals.

Modeling of Glioma Growth With Mass Effect by Longitudinal Magnetic Resonance Imaging.

It is well-known that expanding glioblastomas typically induce significant deformations of the surrounding parenchyma (i.e., the so-called "mass effect"). In this study, we evaluate the performance of three mathematical models of tumor growth: 1) a reaction-diffusion-advection model which accounts for mass effect (RDAM), 2) a reaction-diffusion model with mass effect that is consistent only in the case of small deformations (RDM), and 3) a reaction-diffusion model that does not include the mass effect (RD). The models were calibrated with magnetic resonance imaging (MRI) data obtained during tumor development in a murine model of glioma (n = 9). We obtained T2-weighted and contrast-enhanced T1-weighted MRI at 6 time points over 10 days to determine the spatiotemporal variation in the mass effect and the volume fraction of tumor cells, respectively. We calibrated the three models using data 1) at the first four, 2) only at the first and fourth, and 3) only at the third and fourth time points. Each of these calibrations were run forward in time to predict the volume fraction of tumor cells at the conclusion of the experiment. The diffusion coefficient for the RDAM model (median of 10.65 × 10 -3 mm 2· d -1) is significantly less than those for the RD and RDM models (17.46 × 10 -3 mm 2· d -1 and 19.38 × 10 -3 mm 2· d -1, respectively). The error in the tumor volume fraction for the RD, RDM, and RDAM models have medians of 40.2%, 32.1%, and 44.7%, respectively, for the calibration using data from the first four time points. The RDM model most accurately predicts tumor growth, while the RDAM model presents the least variation in its estimates of the diffusion coefficient and proliferation rate. This study demonstrates that the mathematical models capture both tumor development and mass effect observed in experiments.

Computational Measurement of Motor Imitation and Imitative Learning Differences in Autism Spectrum Disorder: Computational Motor Imitation Measurement in ASD.

Motor imitation is a critical developmental skill area that has been strongly and specifically linked to autism spectrum disorder (ASD). However, methodological variability across studies has precluded a clear understanding of the extent and impact of imitation differences in ASD, underscoring a need for more automated, granular measurement approaches that offer greater precision and consistency. In this paper, we investigate the utility of a novel motor imitation measurement approach for accurately differentiating between youth with ASD and typically developing (TD) youth. Findings indicate that youth with ASD imitate body movements significantly differently from TD youth upon repeated administration of a brief, simple task, and that a classifier based on body coordination features derived from this task can differentiate between autistic and TD youth with 82% accuracy. Our method illustrates that group differences are driven not only by interpersonal coordination with the imitated video stimulus, but also by intrapersonal coordination. Comparison of 2D and 3D tracking shows that both approaches achieve the same classification accuracy of 82%, which is highly promising with regard to scalability for larger samples and a range of non-laboratory settings. This work adds to a rapidly growing literature highlighting the promise of computational behavior analysis for detecting and characterizing motor differences in ASD and identifying potential motor biomarkers.

Can Facial Pose and Expression Be Separated with Weak Perspective Camera?

Separating facial pose and expression within images requires a camera model for 3D-to-2D mapping. The weak perspective (WP) camera has been the most popular choice; it is the default, if not the only option, in state-of-the-art facial analysis methods and software. WP camera is justified by the supposition that its errors are negligible when the subjects are relatively far from the camera, yet this claim has never been tested despite nearly 20 years of research. This paper critically examines the suitability of WP camera for separating facial pose and expression. First, we theoretically show that WP causes pose-expression ambiguity, as it leads to estimation of spurious expressions. Next, we experimentally quantify the magnitude of spurious expressions. Finally, we test whether spurious expressions have detrimental effects on a common facial analysis application, namely Action Unit (AU) detection. Contrary to conventional wisdom, we find that severe pose-expression ambiguity exists even when subjects are not close to the camera, leading to large false positive rates in AU detection. We also demonstrate that the magnitude and characteristics of spurious expressions depend on the point distribution model used to model the expressions. Our results suggest that common assumptions about WP need to be revisited in facial expression modeling, and that facial analysis software should encourage and facilitate the use of the true camera model whenever possible.