The research, led by Berkan Koyak and N. Ahmad Aziz at University Hospital Bonn and the German Center for Neurodegenerative Diseases, together with computer scientists at TU Dortmund University, harnessed a technology more familiar from animated avatars than from neurology clinics. Their application, ExpressionTracker, is built on Apple’s ARKit framework and adapted from software originally developed for photorealistic three-dimensional avatar creation. Using the TrueDepth sensor of an iPad Pro, the system fits a personalised three-dimensional facial mesh to each participant and quantifies 52 predefined blendshape coefficients, each corresponding to a named facial movement such as lowering the lower lip or widening the eyes. The coefficients are normalised from zero to one relative to each individual’s own neutral expression, providing person-specific calibration of baseline facial anatomy.

The protocol was strikingly simple. Seventy participants, 34 people with Parkinson’s disease and 36 healthy controls, sat 30 centimetres from the tablet’s front camera in a room lit only by fluorescent ceiling lights. After establishing a neutral baseline, they performed 52 voluntary facial movements guided by animated demonstrations, with a four-second relaxation pause after each task. The entire assessment took five to ten minutes. From the 53 recorded poses, the researchers derived three measurement conditions: neutral resting weights, peak voluntary activation, and the delta between the two, which captures movement amplitude. On top of these 156 base features, they engineered aggregation statistics, multiplicative interaction terms pairing facial effectors across different tasks, and 60 left-minus-right asymmetry indices, producing a feature space of 261 dimensions.

The results went well beyond the expected finding that people with Parkinson’s move their faces less. Six features survived stringent multiple-testing correction with moderate-to-large effect sizes, and the most informative were not simple amplitude measures. The dynamic range of right lower lip depression was markedly reduced in patients, differing between groups in both the delta and activated conditions with a rank-biserial correlation of up to 0.722. Both outer brows rested significantly higher in patients, a bilaterally elevated resting brow position that had not previously been systematically described in Parkinson’s disease. An asymmetry index for the left mouth frown was significantly greater in patients, and an interaction feature combining right eye widening with right lower lip depression showed one of the largest between-group effects of all, hinting at a right-sided hemihypomimia in this cohort.

That interaction result is scientifically important because facial expression is inherently synergistic, requiring coordinated combinations of activated and relaxed muscle groups. Prior automated approaches have almost exclusively measured how far individual landmarks move, ignoring both lateralised impairment and cross-effector coupling. Here, the interaction between mouth-left and mouth-smile-right features explained 42 percent of the variance in clinician-rated hypomimia, the strongest association in the entire correlation analysis, while mouth-smile-right alone explained 38 percent. Reduced voluntary eyebrow depression correlated with more advanced Hoehn and Yahr stage, and reduced capacity for lip pursing correlated with higher levodopa equivalent dose. These patterns suggest that blendshape analysis captures distinct, anatomically plausible dimensions of motor dysfunction that a single global expressivity rating cannot resolve.

The lower face’s prominence in the findings is anatomically coherent. The upper face benefits from bilateral cortical motor innervation and appears comparatively resilient to unilateral nigrostriatal degeneration, whereas the lower face relies predominantly on contralateral corticobulbar input and is more vulnerable to Parkinson’s pathology. The predominance of perioral features also aligns with earlier automated video studies showing that reduced lower lip movement is among the strongest correlates of dopaminergic loss and limb bradykinesia. The elevated resting brow, by contrast, echoes a well-established compensatory frontalis hyperactivity seen in progressive supranuclear palsy, and the authors caution that its presence in Parkinson’s disease requires validation before any neuroanatomical interpretation.

To test diagnostic value, the team ran eight machine-learning classifiers through a rigorously nested cross-validation pipeline in which every data-dependent step, including age and sex residualisation, imputation, scaling, elastic-net feature selection, and hyperparameter tuning, was fitted only on training folds. The gradient boosting machine performed best, achieving an area under the receiver operating characteristic curve of 0.834, followed by XGBoost at 0.810 and LightGBM at 0.804. At a fixed decision threshold, the best model reached an accuracy of 0.786, sensitivity of 0.735, and specificity of 0.833. Crucially, the engineered ARKit features outperformed an amplitude-only baseline, which peaked at an AUC of 0.752, and a demographic confounder-only baseline using age and sex, which peaked at 0.681, demonstrating that the diagnostic signal is not an artefact of cohort demographics. A label-permutation test with 1000 shuffles confirmed that discrimination exceeded chance for seven of the eight classifiers.

Interpretability analysis using SHAP values reinforced the central message. The most influential feature in the best model was the dynamic range of right lower lip depression, followed by baseline lateral gaze position, the asymmetry of inward gaze change, right eye widening, and left downward gaze. Notably, engineered OpenFace action-unit features computed on the same frames failed to discriminate patients from controls, a contrast the authors interpret cautiously since frame selection was performed by ARKit. The comparison nevertheless underscores a key methodological advantage: ARKit blendshape coefficients are both person-referenced and anatomically interpretable, and the TrueDepth sensor’s active depth encoding is by design less dependent on ambient lighting and skin texture than pixel-based or convolutional approaches known to be sensitive to those factors.

The authors are candid about limitations. The between-group and correlation analyses retain dependence on clinician-rated scores, all clinical assessments came from a single unblinded rater, and apathy and depression, which can mimic hypomimia, were not systematically assessed. The method requires frontal head pose, the sample was modest and single-site without external validation or test-retest data, and all participants had established diagnoses, leaving performance in prodromal populations unknown. The protocol captured only cued voluntary movements, not the spontaneous expressions also affected in Parkinson’s disease. And ARKit itself is proprietary and closed-source, designed for consumer animation rather than clinical measurement, with blendshape coefficients that saturate at their upper bound in the most severely affected individuals.

Even with these caveats, the study makes a compelling case that parkinsonian hypomimia is a multidimensional phenomenon encompassing diminished amplitude, bilaterally elevated resting brow tone, perioral cross-task interaction, and hemifacial asymmetry, all extractable in under ten minutes from a consumer tablet. With external validation in larger and more diverse cohorts, ExpressionTracker could become a candidate digital biomarker for community screening, longitudinal disease monitoring, and drug-efficacy assessment in clinical trials, precisely the objective, standardised measurement infrastructure that disease-modifying therapy development has been waiting for.

Subject of Research: Objective quantification of Parkinsonian hypomimia using facial blendshape analysis

Article Title: Interaction- and asymmetry-aware facial blendshape analysis for objective quantification of Parkinsonian hypomimia

Article References: Koyak, B., Menzel, T., Rodemann, M., Hennes, G., Spottke, A., Saxler, E., Bedarf, J., Faber, J., Sommerauer, M., Weydt, P., Reuter, M., Botsch, M., Wuellner, U., & Aziz, N. A. (2026). Interaction- and asymmetry-aware facial blendshape analysis for objective quantification of Parkinsonian hypomimia. npj Parkinson's Disease, 12(1), Article 231. https://doi.org/10.1038/s41531-026-01579-2

Image Credits: AI Generated

DOI: 10.1038/s41531-026-01579-2

Keywords: Parkinson's disease, hypomimia, facial blendshapes, ARKit, digital biomarker, machine learning, facial asymmetry, ExpressionTracker, cross-validation, MDS-UPDRS, computer vision, neurodegeneration

The research, conducted by Natalia Jaguszewska, Grazyna Jasienska, and Urszula M. Marcinkowska of Jagiellonian University Medical College in Cracow, Poland, examined whether facial fluctuating asymmetry, the small, random deviations from perfect bilateral symmetry that arise from developmental imprecision, was related to multiple dimensions of female sexuality. The team measured not only long-term sociosexual orientation, including the number of sexual partners and openness to casual sex, but also current sexual activity and desire, captured through questions adapted from the Female Sexual Function Index. Seventy-five women aged between twenty and thirty-seven and a half years took part, most of them in relationships, and all were photographed three times across a single menstrual cycle under standardized conditions.

The technical rigor of the facial measurements is one of the study’s standout features. Researchers placed 124 anatomical landmarks on each photograph using PSYCHOMORPH software, with the person doing the delineation blind to the participants’ sexuality responses. The landmark configurations were then Procrustes-aligned using the R package geomorph, and asymmetry was computed as the Euclidean distance between the original and mirrored facial configurations, following established analytical methods. By averaging facial shape measurements across three separate photographic sessions, taken in the follicular phase, around mid-cycle, and in the late luteal phase, the team substantially reduced measurement error, a common weakness in asymmetry research.

At first glance, the raw correlations seemed to support the classic evolutionary prediction. Women with more asymmetrical faces reported fewer sexual partners in the past year, a statistical association of rho equal to minus 0.23. More asymmetrical women also scored lower on the attitude subscale of the revised Sociosexual Orientation Inventory, which measures views on casual, uncommitted sex, with rho equal to minus 0.25, and showed a weaker, borderline trend toward lower sexual desire, at rho equal to minus 0.21. On the surface, this looked like exactly the pattern the mate-selection framework would predict: symmetrical women, presumably signaling better biological condition, enjoying more open and active sexual lives.

But then the researchers did something that many previous studies in this field had not done adequately: they controlled for confounding variables. When place of residence, urban versus rural, and relationship status were added to the regression models, every single one of those associations vanished. The p-values for the asymmetry-sexuality relationships ranged from 0.42 to 0.91, nowhere near conventional significance thresholds. The apparent biological signal, it turned out, was almost certainly not a signal about faces and sexuality at all, but a statistical echo of where the women lived and whether they were partnered.

The confounding structure was substantial. Women from rural areas had significantly higher facial asymmetry than their urban counterparts, were less sexually active, reported lower sexual desire, and scored lower on most aspects of sociosexuality, including attitudes toward casual sex, numbers of one-night stands, and numbers of uncommitted partners. This pattern is consistent with earlier findings showing that people of lower socioeconomic status, who are disproportionately located in rural areas, tend to show higher fluctuating asymmetry, possibly reflecting poorer developmental conditions, more physically demanding labor, and greater exposure to stressors. Rural women in Poland also tend to be more conservative and religious, predominantly Catholic, which may contribute to lower reported sexual openness, and they were more often married or in a relationship, further shaping their sexual behavior.

The study’s authors are careful to note that place of residence should be treated as a contextual control variable rather than a confirmed causal mechanism. They did not have direct evidence linking specific environmental exposures, such as endocrine-disrupting chemicals, differences in physical workload, or farm environments, to facial asymmetry in their dataset. Still, the broader lesson is hard to escape: if a study had stopped at the simple correlations, it would have published a headline-grabbing claim that facially symmetrical women are more sexually open and active, a claim the fuller analysis could not support. The researchers explicitly raise the question of whether previously reported asymmetry-sexuality relationships in the literature would survive similar statistical controls.

The null result also lands in the middle of an ongoing debate about what facial symmetry actually tells us. While symmetrical faces have long been rated as more attractive in many studies, recent work has failed to find a correlation between symmetry and attractiveness, suggesting that other features, such as facial averageness or distinctiveness, may be better predictors of perceived beauty. Some researchers have even shown that preferences for symmetry depend on how a study is designed. A 2025 scoping review by Obrochta and colleagues highlighted the weak relationship between facial fluctuating asymmetry and health more generally, casting further doubt on the assumption that the face is a reliable billboard of biological quality.

From an evolutionary standpoint, the original hypothesis had a certain logic. Even though the number of sexual partners is not a reliable indicator of female reproductive success in the way it may be for men, women could in principle obtain indirect reproductive benefits from multiple partners, and if symmetry signals good genes, symmetrical women might be expected to secure more mates through intersexual selection. Earlier studies had linked body asymmetry to fewer lifetime partners and fewer extrapair copulations, and facial asymmetry to a later age at first reproduction. Yet the present data found no evidence for the first step of the proposed mate-value pathway: partnered women did not differ in facial asymmetry from single women, and asymmetry was unrelated to partner numbers even in the regression models.

The authors acknowledge the limitations of their sample, which was too small to support formal mediation analyses, and they call for larger, longitudinal studies that could disentangle whether variables like residence and relationship status function merely as statistical confounds or as part of genuine developmental pathways. They also suggest that future work should incorporate independent ratings of facial attractiveness to test whether attractiveness mediates any link between facial shape and sociosexuality. For now, the takeaway is a refreshing dose of scientific humility. The human face may still hold clues to our biology, but this study shows that some of the most tantalizing correlations in evolutionary psychology can evaporate the moment researchers account for the ordinary social geography of people’s lives. Culture, it seems, can masquerade convincingly as biology, and only careful, confounder-aware analysis can tell the difference.

Subject of Research: The relationship between facial fluctuating asymmetry and female sexuality, controlling for residence and relationship status

Article Title: Facial Asymmetry and Sexuality in Women

Article References: Jaguszewska, N., Jasienska, G., & Marcinkowska, U. M. (2026). Facial Asymmetry and Sexuality in Women. Archives of Sexual Behavior. https://doi.org/10.1007/s10508-026-03527-w

Image Credits: AI Generated

DOI: 10.1007/s10508-026-03527-w

Keywords: facial asymmetry, fluctuating asymmetry, sociosexuality, female sexuality, evolutionary psychology, facial attractiveness, mate selection, developmental stability, rural-urban differences, relationship status, Archives of Sexual Behavior, Poland

The zygomatic bone, commonly known as the cheekbone, connects to four neighbouring bones through four sutures: the zygomaticomaxillary suture where it meets the upper jaw, the sphenozygomatic suture where it meets the sphenoid bone deep within the skull, the temporozygomatic suture where it meets the temporal bone, and the frontozygomatic suture where it meets the frontal bone of the forehead. Together, these junctions form the architectural framework of the zygomaticomaxillary complex, a structure that is frequently fractured in facial trauma from road accidents, sports injuries and assaults. When surgeons repair such fractures, they rely on anatomical landmarks to restore the original position of the bone, but until now there has been limited comparative data on the projected geometry of all four sutures measured within a single, consistent imaging protocol.

The research team took a retrospective cross-sectional approach, analysing cranial CT examinations that yielded 762 measurable sides across 381 adult patients. Rather than measuring the sutures directly on individual axial slices, which can distort their true course, the researchers used a technique called maximum intensity projection reformats with standardised thick slabs. This method projects the brightest bone signals along a viewing axis onto a single plane, allowing the entire winding course of each suture to be visualised and measured as a continuous line. On these projected images, the team recorded four parameters for each suture on both sides of every patient: the projected length along the suture’s full course, the end-to-end straight-line distance between its endpoints, the sinuosity, which is the ratio of the full course length to the straight-line distance and captures how winding the suture is, and the angle the suture makes with the Frankfort horizontal plane, a standard cranial reference plane that runs from the lower edge of the eye socket to the upper edge of the ear canal.

The statistical analysis was designed with care for the fact that measurements from the two sides of the same patient are not independent observations. The researchers used generalised estimating equations, a method that accounts for this within-patient correlation, and applied false discovery rate adjustment to control for the many comparisons being made simultaneously. They also calculated patient-level Spearman correlations to explore whether the dimensions of one suture predict those of another within the same individual, and intraclass correlation coefficients to quantify how closely the left and right sides of each suture matched within a single person. This combination of approaches allowed the team to distinguish genuine group-level patterns from individual variability, a distinction that proved central to their findings.

The results revealed that each of the four sutures possesses its own characteristic geometric signature. Among the defined projected measurement courses, the zygomaticomaxillary suture showed the greatest mean length and the greatest end-to-end distance, reflecting its role as the long horizontal junction between the cheekbone and the upper jaw. The orientation measurements were particularly striking: the mean angles to the Frankfort horizontal plane ranged from 15.3 degrees for the frontozygomatic suture, which runs nearly parallel to that reference plane, to 67.1 degrees for the sphenozygomatic suture, which plunges steeply into the depth of the skull. In other words, the four sutures are not variations on a theme but genuinely different structures in terms of size, winding pattern and three-dimensional orientation, each demanding its own reference values in any clinical or anatomical application.

The correlations between sutures added another layer of nuance. For dimensional parameters such as length and distance, and for sinuosity, the intersuture correlations were weak to moderate, with the largest Spearman coefficient reaching only 0.40. This suggests that a person with a long zygomaticomaxillary suture cannot be assumed to have correspondingly long other zygomatic sutures; each suture appears to develop largely on its own geometric trajectory. The angular correlations were even smaller, with the largest absolute coefficient at just 0.12, and none of the angular correlations remained statistically significant after false discovery rate adjustment. The orientation of one suture, in short, tells you essentially nothing about the orientation of its neighbours, reinforcing the picture of four independently patterned anatomical features.

Sex and side differences emerged in a pattern that clinicians will find informative. Length and end-to-end distance were greater in males for all four sutures, consistent with the overall sexual dimorphism of the craniofacial skeleton. Side differences, however, were remarkably limited at the group level: the only directional left-right difference detected was in the sphenozygomatic straight-line distance. This near-symmetry at the population level might suggest that the two sides of any individual’s face are mirror images, but the bilateral intraclass correlation coefficients told a more complicated story. These coefficients, which measure how strongly the left and right measurements correspond within the same person, ranged from 0.37 to 0.80 across the different sutures and parameters. Small average side differences, in other words, coexist with considerable individual variation in symmetry, meaning that some people have closely matched pairs of sutures while others show substantial left-right divergence even though the group averages look balanced.

This distinction between group-level symmetry and individual-level agreement has practical implications for the concept of using the uninjured side as a template for reconstructing the injured side, a common strategy in maxillofacial surgery. If bilateral agreement varies substantially from person to person, then the healthy side of a patient with a zygomaticomaxillary complex fracture may be a more or less reliable guide depending on which suture is involved and which parameter is being compared. The authors are careful to stress, however, that their data come from a single centre and a single scanner, and that the measurements were descriptive projected morphometric values obtained from a general adult population rather than from fracture patients. Before these reference values can be used in fracture reduction assessment or surgical planning, they require external validation in independent cohorts and direct evaluation in patients with actual zygomatic fractures.

The study also stands as an example of methodological transparency in modern anatomical research. The protocol was approved by the İstanbul Medipol University Non-Interventional Clinical Research Ethics Committee, and the requirement for individual informed consent was waived because of the retrospective design and the use of fully anonymised archival imaging data. The authors disclosed that a generative artificial intelligence tool was used for English-language editing, assistance with statistical coding and document formatting, but not for autonomous data analysis or interpretation, with all statistical analyses executed, reviewed and interpreted by the researchers themselves. The work received no specific external funding, and the authors declared no competing interests.

For a field in which surgical decisions can hinge on millimetre-scale judgments of bone position, the value of a large, standardised morphometric dataset is difficult to overstate. By demonstrating that the four zygomatic sutures have distinct projected profiles, that their orientations span a remarkable range from nearly horizontal to steeply inclined, and that individual bilateral symmetry is variable even when population averages appear balanced, the study gives anatomists, radiologists and surgeons a firmer quantitative foundation for understanding the midface. The next step, as the authors make clear, is to test whether these normal values can discriminate between restored and malpositioned bones in real fracture cohorts, transforming a descriptive anatomical atlas into a practical instrument for reconstructive care.

Subject of Research: Computed tomography-based morphometry of the four zygomatic sutures in adults

Article Title: Computed tomography-derived projected morphometry of four zygomatic sutures: distinct geometric profiles and bilateral agreement

Article References: Altıntaş, H. M., Köksal, A., Cingöz, G., Eberliköse, H., & Demir, B. T. (2026). Computed tomography-derived projected morphometry of four zygomatic sutures: distinct geometric profiles and bilateral agreement. BMC Medical Imaging. https://doi.org/10.1186/s12880-026-02896-x

Image Credits: AI Generated

DOI: 10.1186/s12880-026-02896-x

Keywords: computed tomography, zygomatic sutures, morphometry, craniofacial anatomy, zygomaticomaxillary complex, facial asymmetry, maxillofacial surgery, maximum intensity projection, bilateral agreement, Frankfort horizontal plane, facial trauma, BMC Medical Imaging