STUDY OBJECTIVES:Excessive daytime sleepiness (EDS), influenced by environmental and social-behavioral factors, is reported by a subset of patients with sleep apnea-a group that may be at elevated cardiovascular risk. However, it is unclear whether sleep apnea with and without EDS have distinct genetic underpinnings. In this study, we perform gene-by-EDS interaction analyses for apnea hypopnea index, a diagnostic marker of sleep apnea severity, to understand EDS's influence on its underlying genetic risk. METHODS:Discovery interaction analyses for common variants and gene-based rare variants were conducted respectively using multi-ethnic Trans-Omics for Precision Medicine (N = 11 619) data, followed by replication and subsequent meta-analysis in additional Trans-Omics for Precision Medicine-imputed data (N = 8904). The 1 degree-of-freedom (1df) G × E test and the 2df joint G,G × E tests were utilized. Sex-stratified analyses were additionally performed. RESULTS:Discovery analysis revealed two common intronic variants-rs13118183 (CCDC3) and rs281851 (MARCHF1)-and three rare variant gene sets mapped to SCUBE2, TMEM26, and CPS4FL-to exhibit interaction with EDS. Meta-analysis revealed EDS interaction with 11 rare variant gene sets mapped to UBLCP1, MED31, RAP1GAP, CPNE5, MYMX, YY1, ZNF773, YBEY, IQCB1, PI4K2B, and CORO1A. CONCLUSION:Genetic loci reveal connections to cardiovascular risk, insulin resistance, thiamine deficiency, and resveratrol mechanism. Discovered genetic signals may offer insight into pertinent biological pathways for sleep apnea patients with an excessively sleepy subtype. Statement of Significance Sleep apnea is a complex sleep disorder. Exemplifying this is the disparately varying estimates of presence of excessive daytime sleepiness (EDS) in patients, and persistent EDS that lingers despite treatment. Some data indicate that the excessively sleepy subtype of sleep apnea carries heightened cardiovascular risk. Whether EDS influences genetic risk factors underlying sleep apnea has not yet been investigated. This study addresses this gap, as the first genome-wide gene × EDS interaction study for apnea hypopnea index, the standard sleep apnea severity metric. Genetic loci that have been previously unconsidered for sleep apnea are revealed. Discovered interaction signals highlight pathways in metabolism, genes associated with cardiometabolic traits, and therapeutic agents influencing obesity, blood pressure, oxidative stress, and apnea hypopnea index.

A significant component in regulatory approval of diagnostic artificial intelligence (AI) models is the use of reader study evidence, diagnostic performance studies of radiologists on a curated set of images. Retrospective MRMC (multi-reader, multi-case) reader studies of AI use are commonly conducted in the field of medical AI, though they tend to overestimate AI performance and underestimate human performance through the use of aggregated metrics. Further, there has been limited investigation into the impact of diagnostic AI models on radiological decision-making, with few if any cognitive modelling studies undertaken. This has serious implications, as the evidence base needed for implementation of models is lacking verified cognitive research. As such, this paper explores a lost opportunity to model diagnostic reader studies to ask: what can reader studies teach us about radiologist adaptation to AI models? In this study, we used a 2021 reader study of a commercially available Australian diagnostic system by Harrison.ai, consisting of 20 expert radiologists reading 1163 cases. We fit hierarchical signal detection models at the individual and pathology level to a reader study of 20 expert radiologists over 1163 cases and 127 pathologies. Our modelling results indicated that radiologists at the pathology and individual level had increased discriminability with AI, though with a more liberal response bias, leading to higher false positive rates. Further exploratory analysis based on these results indicated that disease coverage rates became homogenised with AI, as well as different AI-generated segmentations potentially influencing correct rejection rates. We provide real-world interpretations of these findings.

Deep learning models for medical image classification are susceptible to subgroup performance disparities across demographic attributes such as age, gender, and race. We identify a latent representational mechanism underlying these disparities: in transfer-learned models, the dominant penultimate-layer activation channel under positive predictions is co-activated by both disease-positive samples and privileged demographic groups (male, older patients), producing over-diagnosis; conversely, the dominant channel under negative predictions is co-activated by disadvantaged groups (female, younger patients), producing systematic under-diagnosis. To address this, we propose Neuron Incidence Redistribution (NIR), a lightweight regularization method that penalizes the variance of predicted-probability-weighted mean activations across penultimate-layer neurons, requiring no demographic labels at training time. On HAM10000, TPR disparity drops from 10.81

OBJECTIVE:This study aimed to determine the relative contributions of genetic and environmental factors to phenotypic variations of dental arch shape from the primary to permanent dentition stages. MATERIALS AND METHODS:Maxillary and mandibular digital models of 188 twin pairs (90 monozygotic and 98 dizygotic) in the primary dentition stage, tracked longitudinally through the mixed and permanent dentition stages, were examined. Dental landmarks were recorded at the incisal edges of incisors and the cusp tips of canines and posterior teeth in both arches in MeshLab. Arch shape variation was analysed using Procrustes superimposition and principal component analysis, then partitioned into genetic and environmental variance components using genetic structural equation modelling. RESULTS:The first three principal components were meaningful, explaining 42%-54% of the phenotypic variation in dental arch shape from the primary to permanent dentition stages. The arch depth-to-width ratio (principal component 1) greatly influenced arch shape in both maxillary and mandibular arches across all dentition stages. Other components included canine position, displacement and rotation of incisors and posterior teeth and the steepness of the curve of Spee. An AE model incorporating additive genetic (A) and non-shared environmental (E) components best explained the phenotypic variances in dental arch shape, with narrow-sense heritability estimates ranging from 0.60 to 0.84 for the first three principal components. CONCLUSIONS:The dental arch shape was predominantly influenced by additive genetic and non-shared environmental factors throughout development, with moderate to high heritability estimates across all dentition stages.

A significant component in regulatory approval of diagnostic artificial intelligence (AI) models and the decision to procure an AI model lies in the use of retrospective MRMC (multi-reader, multi-case) reader study evidence; diagnostic performance studies of radiologists on a curated set of images. However, previous conclusions drawn about radiologist-AI performance based on reader studies have been inconsistent and contradictory, often plagued with over-inflated measures of AI performance at the expense of radiologist performance. Our paper addresses a lost opportunity to explore AI reader study datasets through a cognitive lens, allowing unique insights into the psychological mechanisms involved in radiological detection with AI. By grounding our findings in human behaviour, the takeaways for the healthcare system have greater relevance and value than pure diagnostic measures. In this study, we used a 2021 reader study of a commercially available Australian diagnostic system by Harrison.ai, consisting of 20 expert radiologists reading 1163 cases. We fit hierarchical signal detection models at the individual and pathology level to a reader study of 20 expert radiologists over 1163 cases and 127 pathologies. Our modelling results indicated that changes to diagnostic decision-making with AI are due to improvements in discriminability and shifts in decision-thresholding, leading to more detections but higher false positive rates. Further, we used our modelling results as a guide for targeted exploratory analysis, finding changes to disease coverage rates across the radiologist cohort, as well as different AI-derived segmentations potentially influencing correct rejection rates. We note three key interpretations of our findings that have direct implications for AI in our health system: increases in patient false positive diagnoses may change resource planning and increase workforce needs; clinical reporting may become homogenised and less diverse; and AI explainability could influence the certainty of diagnoses made for patients.

OBJECTIVE:This study aimed to estimate the relative contributions of genetic and environmental factors to phenotypic variations of dental arch traits from primary to permanent dentition stages. METHODS:Digital dental models of 188 Australian twin pairs (90 monozygotic and 98 dizygotic) in the primary dentition stage, followed up through the mixed and permanent dentition stages, were included in the study. Landmarks were identified on both maxillary and mandibular dental arches in MeshLab for measuring intercanine widths, intermolar widths, arch lengths, overjet, overbite and molar relationships. Genetic structural equation modelling was performed on the quantitative twin data of dental arch traits. RESULTS:The phenotypic variance of dental arch traits was generally best explained by a model incorporating additive genetic (A) and non-shared environmental (E) components, an AE model. However, the variance of overjet in the primary dentition was best explained by shared environmental (C) and non-shared environmental (E) components. Heritability estimates were high for intra-arch traits (0.65-0.88), but low to moderate for inter-arch traits (0.21-0.51). While heritability estimates fluctuated for most traits from primary to permanent dentition stages, the estimates for arch lengths and intermolar widths were mostly above 0.8 throughout development. LIMITATION:Only twins of European descent were included in this study. CONCLUSIONS:Dental arch traits were mostly influenced by additive genetic and non-shared environmental factors during development. Except for arch lengths and intermolar widths, genetic and environmental influences on dental arch traits fluctuated during development, with the genetic influence at its lowest during the mixed dentition stage.

BACKGROUND:The burden of inflammatory bowel disease (IBD) is often reported on from a system or cost viewpoint. We created and explored a novel patient-perceived burden of disease (PPBoD) score in a large Australasian cohort. AIM:To create and explore a novel patient-perceived burden of disease (PPBoD) score in a large Australasian cohort, and correlate PPBoD scores with demographics, disease and treatment factors. METHODS:The Crohn Colitis Care Registry was interrogated in October 2023. Data from adults with IBD with an outpatient care encounter in the last 14 months among 17 centres were included. A novel PPBoD score was designed for ulcerative colitis (UC), Crohn disease (CD) and IBD-unclassified (IBDU). Correlations between PPBoD scores and demographics, disease and treatment factors were examined. RESULTS:Of those with adequate data, 46.7% (2653/5685) had no PPBoD, 34.6% (1969/5685) had mild, 11.3% (641/5685) had moderate and 7.4% (422/5685) had significant PPBoD. New Zealanders were more likely to have higher PPBoD compared to Australians (P = 0.047). Greater PPBoD was seen in patients with CD and IBDU compared to patients with UC (P < 0.001) and females were more likely to have significant PPBoD (8.7%) than males (6.1%) (P < 0.001). People with no or mild PPBoD were more likely to be on advanced therapies (55.7% and 59.5% respectively) than those with significant PPBoD (46.3%) (P < 0.001). The proportion of people on advanced therapies in Australia was higher than in New Zealand (61.2% vs 38.5% respectively, P < 0.001). Steroid usage was significantly higher in people with greater PPBoD (significant BoD 7.1% vs no BoD 1.1%; P < 0.001). CONCLUSION:Most of this real-world care cohort had no or mild PPBoD. Data suggest that higher PPBoD levels may be resolved by appropriate therapeutic escalations.

PURPOSE:Pituitary tumours are relatively common, and familial in approximately 5% of cases. However, germline genetic contributions to pituitary tumour development are incompletely characterised. Preliminary evidence suggests pituitary tumours may be promoted by variants in pituitary organogenesis genes. Our study aimed to identify rare germline variants in pituitary organogenesis genes that may contribute to pituitary tumour development. METHODS:A familial case of pituitary disease was investigated. We also examined 36 pituitary organogenesis genes in 134 individuals with pituitary tumours using a targeted next-generation sequencing panel, identifying and characterising variants with a population allele frequency < 0.05%. RESULTS:One patient with a prolactin-secreting pituitary tumour and his daughter with combined pituitary hormone deficiency shared a rare germline variant in FGFR1, c.386 A > C, p.(D129A). In our broader study, we identified an additional individual with the FGFR1 D129A variant and demonstrated enrichment compared to a control population derived from the Genome Aggregation Database (gnomAD). We also observed 66 rare germline variants in pituitary organogenesis genes amongst 54/134 individuals (40%). However, compared to control data, the study cohort exhibited no enrichment for other rare variants in FGFR1, FGF-related genes, or other pituitary embryogenesis genes. CONCLUSION:Our results suggest that the FGFR1 D129A variant may be associated with pituitary tumorigenesis but the role of other pituitary embryogenesis genes remains unclear. Additional independent cohorts and functional studies are required.

Deformable brain templates are an important tool in many neuroimaging analyses. Conditional templates (e.g., age-specific templates) have advantages over single population templates by enabling improved registration accuracy and capturing common processes in brain development and degeneration. Conventional methods require large, evenly spread cohorts to develop conditional templates, limiting their ability to create templates that could reflect richer combinations of clinical and demographic variables. More recent deep-learning methods, which can infer relationships in very high-dimensional spaces, open up the possibility of producing conditional templates that are jointly optimised for these richer sets of conditioning parameters. We have built on recent deep-learning template generation approaches using a diffeomorphic (topology-preserving) framework to create a purely geometric method of conditional template construction that learns diffeomorphisms between: (i) a global or group template and conditional templates, and (ii) conditional templates and individual brain scans. We evaluated our method, as well as other recent deep-learning approaches, on a data set of cognitively normal (CN) participants from the Alzheimer's Disease Neuroimaging Initiative (ADNI), using age as the conditioning parameter of interest. We assessed the effectiveness of these networks at capturing age-dependent anatomical differences. Our results demonstrate that while the assessed deep-learning methods have a number of strengths, they require further refinement to capture morphological changes in ageing brains with an acceptable degree of accuracy. The volumetric output of our method, and other recent deep-learning approaches, across four brain structures (grey matter, white matter, the lateral ventricles and the hippocampus), was measured and showed that although each of the methods captured some changes well, each method was unable to accurately track changes in all of the volumes. However, as our method is purely geometric, it was able to produce T1-weighted conditional templates with high spatial fidelity and with consistent topology as age varies, making these conditional templates advantageous for spatial registrations. The use of diffeomorphisms in these deep-learning methods represents an important strength of these approaches, as they can produce conditional templates that can be explicitly linked, geometrically, across age as well as to fixed, unconditional templates or brain atlases. The use of deep learning in conditional template generation provides a framework for creating templates for more complex sets of conditioning parameters, such as pathologies and demographic variables, in order to facilitate a broader application of conditional brain templates in neuroimaging studies. This can aid researchers and clinicians in their understanding of how brain structure changes over time and under various interventions, with the ultimate goal of improving the calibration of treatments and interventions in personalised medicine. The code to implement our conditional brain template network is available at: github.com/lwhitbread/deep-diff.

OBJECTIVES:Artificial intelligence (AI) has demonstrated the potential to improve efficiency and reliability of radiographic scoring in rheumatoid arthritis but lacks sufficient evidence to justify clinical use. We developed and rigorously validated a deep learning model to automate radiographic scoring against two external test sets, drawing upon state of the art reporting guidelines to clarify present barriers to implementation. METHODS:AI algorithms were trained to predict the Sharp van der Heijde score in hands and feet using a cohort of 157 patients and 1470 radiographs. External replication was undertaken in test datasets from two hospitals (n=253 patients, 589 radiographs). Alongside standard performance metrics to measure error and agreement, we reported subgroup performance, conducted an exploratory analysis of error, and demonstrated relationships with functional outcomes. RESULTS:Our AI system underperformed compared to manual scoring, with lower agreement between the AI and consensus score than between the two manual scorers. The AI system was better at ranking scores than achieving absolute agreement, with intraclass correlation coefficients ranging from 0.03 to 0.27 while Spearman's correlation coefficients were consistently higher, ranging from 0.16 to 0.55. CONCLUSIONS:The performance of the AI systems developed for automating radiographic scoring in RA is insufficient to justify use in research or clinical practice. Large, diverse, and thoroughly described longitudinal datasets will be indispensable in the development and rigorous evaluation of algorithms. Achieving this is key to the ongoing precise evaluation of clinical outcomes in rheumatoid arthritis to enable further improvements to patient care.

In both the SAC305 and eutectic Sn-Bi solders, the electromigration behavior is dominated by one element preferentially migrating in the solder towards the anode. In the case of SAC305 solder, which contains 0.5 wt. % Cu, the copper atoms migrate to form CU6Sn5 intermetallic phase at the anode. In the case of eutectic Sn-Bi solder, bismuth atoms migrate to the anode forming a layer of the Bi-rich phase. In both cases, the phases accumulating at the anode have higher electrical resistivities than the original solder compositions and occupy large volume factions of the solder joints. This combination of higher resistivity and large volume fraction of the phase forming at the anode raises the electrical resistance of the joint as a function of time during electromigration stressing. The paper will discuss and compare the metallurgical and the electromigration behavior of the two solder alloys in the early stage of electromigration before enough voids have formed at the cathode interface to coalesce and start dominating the joint resistance. It will be shown that in both the alloys the rate of change of solder joint resistance as a function of current density and temperature can be concisely presented on Arrhenius plots and be used to predict the life of solder joints under any current density and temperature condition.

Conformal coatings have traditionally been tested by determining the mean time to failure of conformally coated hardware exposed to corrosive test environments. This test approach has serious shortcomings: The test temperatures are most often too high. At these high temperatures, the conformal coating properties may be quite different from those at the application temperatures. In addition, the times to failure are unacceptably long extending into many months. Overcoming these shortcomings is an iNEMI championed test that involves exposing conformally coated thin films of copper and silver to sulfur vapors at 40-50 oC in flowers of sulfur (FoS) chamber and using the corrosion rates of the coated metal thin films as a measure of the corrosion protection capabilities of the conformal coatings. The test temperatures are similar to the application temperatures, the test durations are no more than a week and can be conducted under various temperature and humidity conditions. The purpose of this paper was to determine if testing in the industry-standard mixed-flowing gas corrosion chamber would give similar results as those using the FoS chamber. Acrylic, fluorinated acrylate, and atomic layer deposition conformal coatings were tested in three environments: (a) flowers of sulfur (FoS), (b) mixed-flowing gas (MFG), and (c) iodine vapor. The performance of the coatings tested in the FoS and the MFG corrosion chambers were quantitatively similar. The iodine vapor test results were in qualitative agreement with the FoS and MFG test results. In addition, we present early results pointing to the utility of terahertz-frequency imaging as a technique for measuring conformal-coating thickness nondestructively.

BackgroundAccurate outcome predictions for patients who had ischaemic stroke with successful reperfusion after endovascular thrombectomy (EVT) may improve patient treatment and care. Our study developed prediction models for key clinical outcomes in patients with successful reperfusion following EVT in an Australian population.MethodsThe study included all patients who had ischaemic stroke with occlusion in the proximal anterior cerebral circulation and successful reperfusion post-EVT over a 7-year period. Multivariable logistic regression and Cox regression models, incorporating bootstrap and multiple imputation techniques, were used to identify predictors and develop models for key clinical outcomes: 3-month poor functional status; 30-day, 1-year and 3-year mortality; survival time.ResultsA total of 978 patients were included in the analyses. Predictors associated with one or more poor outcomes include: older age (ORs for every 5-year increase: 1.22–1.40), higher premorbid functional modified Rankin Scale (ORs: 1.31–1.75), higher baseline National Institutes of Health Stroke Scale (ORs: 1.05–1.07) score, higher blood glucose (ORs: 1.08–1.19), larger core volume (ORs for every 10 mL increase: 1.10–1.22), pre-EVT thrombolytic therapy (ORs: 0.44–0.56), history of heart failure (outcome: 30-day mortality, OR=1.87), interhospital transfer (ORs: 1.42 to 1.53), non-rural/regional stroke onset (outcome: functional dependency, OR=0.64), longer onset-to-groin puncture time (outcome: 3-year mortality, OR=1.08) and atherosclerosis-caused stroke (outcome: functional dependency, OR=1.68). The models using these predictors demonstrated moderate predictive abilities (area under the receiver operating characteristic curve range: 0.752–0.796).ConclusionOur models using real-world predictors assessed at hospital admission showed satisfactory performance in predicting poor functional outcomes and short-term and long-term mortality for patients with successful reperfusion following EVT. These can be used to inform EVT treatment provision and consent.

Objective In this prospective cohort study, we provide several prognostic models to predict functional status as measured by the modified Health Assessment Questionnaire (mHAQ). The early adoption of the treat-to-target strategy in this cohort offered a unique opportunity to identify predictive factors using longitudinal data across 20 years. Methods A cohort of 397 patients with early RA was used to develop statistical models to predict mHAQ score measured at baseline, 12 months, and 18 months post diagnosis, as well as serially measured mHAQ. Demographic data, clinical measures, autoantibodies, medication use, comorbid conditions, and baseline mHAQ were considered as predictors. Results The discriminative performance of models was comparable to previous work, with an area under the receiver operator curve ranging from 0.64 to 0.88. The most consistent predictive variable was baseline mHAQ. Patient-reported outcomes including early morning stiffness, tender joint count (TJC), fatigue, pain, and patient global assessment were positively predictive of a higher mHAQ at baseline and longitudinally, as was the physician global assessment and C-reactive protein. When considering future function, a higher TJC predicted persistent disability while a higher swollen joint count predicted functional improvements with treatment. Conclusion In our study of mHAQ prediction in RA patients receiving treat-to-target therapy, patient-reported outcomes were most consistently predictive of function. Patients with high disease activity due predominantly to tenderness scores rather than swelling may benefit from less aggressive treatment escalation and an emphasis on non-pharmacological therapies, allowing for a more personalized approach to treatment. Key Points • Long-term use of the treat-to-target strategy in this patient cohort offers a unique opportunity to develop prognostic models for functional outcomes using extensive longitudinal data. • Patient reported outcomes were more consistent predictors of function than traditional prognostic markers. • Tender joint count and swollen joint count had discordant relationships with future function, adding weight to the possibility that disease activity may better guide treatment when the components are considered separately.

Obstructive sleep apnea (OSA) is a multifactorial sleep disorder characterized by a strong genetic basis. Excessive daytime sleepiness (EDS) is a symptom that is reported by a subset of OSA patients, persisting even after treatment with continuous positive airway pressure (CPAP). It is recognized as a clinical subtype underlying OSA carrying alarming heightened cardiovascular risk. Thus, conceptualizing EDS as an exposure variable, we sought to investigate EDS’s influence on genetic variation linked to apnea-hypopnea index (AHI), a diagnostic measure of OSA severity. This study serves as the first large-scale genome-wide gene x environment interaction analysis for AHI, investigating the interplay between its genetic markers and EDS across and within specific sex. Our work pools together whole genome sequencing data from seven cohorts, enabling a diverse dataset (four population backgrounds) of over 11,500 samples. Among the total 16 discovered genetic targets with interaction evidence with EDS, eight are previously unreported for OSA, including CCDC3, MARCHF1, and MED31 identified in all sexes; TMEM26, CPSF4L, and PI4K2B identified in males; and RAP1GAP and YY1 identified in females. We discuss connections to insulin resistance, thiamine deficiency, and resveratrol use that may be worthy of therapeutic consideration for excessively sleepy OSA patients.

Abstract Background The prevalence and burden of inflammatory bowel diseases (IBD) including Crohn’s disease (CD) are rising globally. We present a novel score to evaluate the patient-perceived burden of disease (PPBoD) in CD, and explored it in a large real-world Australasian cohort. Methods The Crohn’s Colitis Care (CCCare) Clinical Registry was interrogated in October 2023. Adults with CD across 17 IBD centres with an outpatient encounter in the last 14 months were included. A novel PPBoD score was designed for CD, which included patient-reported components from the Harvey-Bradshaw index (abdominal pain and patient-rated general wellbeing) as well as nocturnal bowel actions, defecation urgency and stool frequency. The PPBoD score was calculated as detailed in figure 1. A total score of 0 was defined as no PPBoD, 1-2 as mild, 3-4 as moderate and ≥ 5 as significant PPBoD. Correlations amongst PPBoD and demographics, disease and treatment factors were explored. Results A total of 3461 people with CD were assessed in the last 14 months and 3233 (93.4%) had adequate data to calculate PPBoD. Of these, 80.0% had either no or mild PPBoD (table 1). While gender varied significantly between PPBoD categories, age and BMI did not. People with lower PPBoD were more likely to be receiving advanced therapies and had lower rates of steroid use than those with higher PPBoD. There were no significant differences in the use of immunomodulators and/or aminosalicylates across PPBoD categories. The cohort was geographically dispersed across Australia (n = 2414, 74.7%) and New Zealand (n = 819, 25.3%). There was significantly higher PPBoD in New Zealand compared to Australia but notably people in New Zealand were less likely to be receiving advanced therapies (p < 0.001). In the subset of 1074 people (33.2%) with a recent faecal calprotectin, those with no PPBoD were more likely to have biochemical remission (faecal calprotectin < 100 μg/g). Data were available to assess endoscopic and radiological remission in 1049 people (32.4%); those with no PPBoD were more likely to be in remission. Less than 1% of people with no PPBoD had any days out of role due to CD, whereas those with higher PPBoD had more days out of role (Table 1). Conclusion We present a novel consumer-focused score to quantify PPBoD in CD. Within this geographically dispersed cohort, the majority had either no or mild PPBoD. Advanced therapy use appeared to be protective against high PPBoD. Further studies are required to validate this novel score to assess PPBoD in CD.

Objectives This study aimed to determine the genetic and environmental contributions to phenotypic variations of palatal morphology during development.Methods Longitudinal three-dimensional digital maxillary dental casts of 228 twin pairs (104 monozygotic and 124 dizygotic) at primary, mixed, and permanent dentition stages were included in this study. Landmarks were placed on the casts along the midpoints of the dento-gingival junction on the palatal side of each tooth and the mid-palatine raphe using MeshLab. Palatal widths, depths, length, area, and volume were measured using those landmarks. Univariate genetic structural equation modelling was performed on twin data at each stage of dental development.Results Except for anterior depth, all palatal dimensions increased significantly from the primary to permanent dentition stages. The phenotypic variance for most of the palatal dimensions during development was best explained by a model, including additive genetic and non-shared environment variance components. Variance in volume and area in the primary dentition stage was best explained by a model including additive genetic, shared environment, and non-shared environment variance components. For posterior palatal depth and width, narrow-sense heritability estimates were above 0.8 for all dental developmental stages. In contrast, heritability estimates for other palatal traits fluctuated during development.Limitation This study was limited to twins of European ancestry.Conclusions Additive genetic and non-shared environmental factors primarily influenced palatal morphology during development. While the genetic influence on different aspects of the palate varied throughout development, it was particularly strong in the posterior region of the palate and during the permanent dentition stage.

Purpose To investigate the issues of generalizability and replication of deep learning models by assessing performance of a screening mammography deep learning system developed at New York University (NYU) on a local Australian dataset. Materials and Methods In this retrospective study, all individuals with biopsy or surgical pathology-proven lesions and age-matched controls were identified from a South Australian public mammography screening program (January 2010 to December 2016). The primary outcome was deep learning system performance-measured with area under the receiver operating characteristic curve (AUC)-in classifying invasive breast cancer or ductal carcinoma in situ (n = 425) versus no malignancy (n = 490) or benign lesions (n = 44). The NYU system, including models without (NYU1) and with (NYU2) heatmaps, was tested in its original form, after training from scratch (without transfer learning), and after retraining with transfer learning. Results The local test set comprised 959 individuals (mean age, 62.5 years ± 8.5 [SD]; all female). The original AUCs for the NYU1 and NYU2 models were 0.83 (95% CI: 0.82, 0.84) and 0.89 (95% CI: 0.88, 0.89), respectively. When NYU1 and NYU2 were applied in their original form to the local test set, the AUCs were 0.76 (95% CI: 0.73, 0.79) and 0.84 (95% CI: 0.82, 0.87), respectively. After local training without transfer learning, the AUCs were 0.66 (95% CI: 0.62, 0.69) and 0.86 (95% CI: 0.84, 0.88). After retraining with transfer learning, the AUCs were 0.82 (95% CI: 0.80, 0.85) and 0.86 (95% CI: 0.84, 0.88). Conclusion A deep learning system developed using a U.S. dataset showed reduced performance when applied "out of the box" to an Australian dataset. Local retraining with transfer learning using available model weights improved model performance. Keywords: Screening Mammography, Convolutional Neural Network (CNN), Deep Learning Algorithms, Breast Cancer Supplemental material is available for this article. © RSNA, 2024 See also commentary by Cadrin-Chênevert in this issue.