The Impact of Morningness-Eveningness on Depression through a Serial Mediation Model of Resilience and Anxiety.

OBJECTIVE: Resilience has been recently considered one of the possible mechanisms for the association between morningness-eveningness and depression. Meanwhile, anxiety is closely associated with mood disorder, but its association with morningness-eveningness is unclear. Therefore, this study aimed to explore the mediating effects of resilience and anxiety on morningness-eveningness and depression as the possible mechanisms. METHODS: This study included patient group and nonpatient group. Patient group consists of 743 patients with mood disorders [Major Depressive Disorder (MDD), 233; Bipolar Disorder Ⅰ (BDⅠ), 113; Bipolar Disorder Ⅱ (BDⅡ), 397] whereas nonpatient group consists of 818 individuals without mood disorder. The Composite Scale of Morningness, Connor-Davidson Resilience Scale, Self-Rating Depression Scale, and Beck Anxiety Inventory were used to evaluate morningness-eveningness, resilience, anxiety, and depression, respectively. RESULTS: Our model provided a good fit for the data. The association between morningness-eveningness and depression symptoms was partially serially mediated by resilience and anxiety in both the patient and nonpatient groups. The patient group exhibited significantly stronger morningness-eveningness toward resilience and anxiety than the nonpatient group. In the indirect effect of morningness-eveningness on depression, group differences exist only through each mediation of resilience and anxiety, not through serial mediation. CONCLUSION: Our results expand on the mechanism underlying the association between morningness-eveningness and depression. They highlight the importance of morningness-eveningness modification to increase resilience and the need to consider anxiety jointly in this process.

Development and Validation of the Mental Health Scale for Maltreated Children.

OBJECTIVE: This study aimed to develop and validate a comprehensive self-report questionnaire to assess emotional and behavioral problems and psychological trauma in maltreated children. METHODS: The Mental Health Scale for Maltreated Children (MHS-MC) was constructed to encompass five major symptoms (depression, anxiety, inattention/hyperactivity/impulsivity, aggression/defiance, and psychological trauma) prevalent in maltreated children. Critical items and ego-resilience subscale were also devised to increase clinical utility. After informed consent, 205 children (maltreated children, n=157, 76.6%) were recruited nationwide, and they answered a package of self-report measures, including the MHS-MC. Reliability, construct validity, concurrent validity, and criterion-related validity were examined to explore the psychometric properties. RESULTS: The reliability was good to excellent. Confirmatory factor analysis yielded a five-factorial solution for the symptom subscales supporting construct validity. In logistic regression, the total scores of the MHS-MC predicted membership in the maltreated group. Criterion-related validity was generally satisfactory in that all subscales of the MHS-MC showed significant correlations with relevant measures in the expected direction. CONCLUSION: This is the first attempt to develop a comprehensive psychological scale based on nationwide data collected from maltreated Korean children. We hope that the continued standardization of this scale will contribute to evidence-based clinical and policy decisionmaking for maltreated children.

Meaning-making while staying connected matters in psychological adaptation during pandemic: a longitudinal moderated mediation study.

Adversity may bring about both negative and positive changes in psychological adaptation. Although there is mounting evidence regarding the psychological distress during the pandemic, the other side of posttraumatic change, posttraumatic growth (PTG) and its predictors are relatively underexamined. Moreover, there is a paucity of longitudinal investigations that examined intra- and interpersonal predictors responsible for both sides of psychological adaptation. Therefore, this study comprehensively examined the longitudinal relationship among cognitive processing, social support, and adaptation during the pandemic using a moderated mediation model. Specifically, it was tested whether two types of event-related rumination mediated the link between perceived stress and ambilateral adaptational outcomes, and whether social support moderated the mediating pathways of ruminations on adaptation. After informed consent, a representative sample of adults was followed up for over a year, and answered a package of online questionnaires. The results showed that intrusive rumination prospectively predicted greater psychological distress and less PTG in response to stress, whereas deliberate rumination led to less psychological distress and more PTG over time. As predicted, the indirect protective effect of deliberate rumination was stronger when perceived social support was higher. This longitudinal study highlighted the core factors responsible for continued suffering and personal growth during the pandemic. These results have both practical and clinical implications for mental healthcare in the post-COVID era, when the heterogeneity of psychological adaptation increases and preparation for the next pandemic is warranted.

Psychometric properties and measurement invariance of the short form of grit scale in Korean adolescents.

This study aimed to identify the factor structure of the Korean version of the Short Grit Scale (Grit-S) and examine its cross-sectional and longitudinal measurement invariance (MI). Data from the Korean Children and Youth Panel Survey 2018 were analyzed, which included two cohorts, comprising 2,327 and 2,325 fourth-year elementary and first-year middle school students, respectively. It was found that the two-factor model fit the data well for the elementary and middle school samples. The results of the cross-sectional MI tests across genders indicated that the full threshold and loading invariance were also supported for the elementary school sample, and the partial threshold and loading invariance were supported for the middle school sample. The analyses of the longitudinal MI revealed that the partial threshold and loading invariance were supported for both samples. The reliability analysis revealed satisfactory McDonald's Omega values for both samples at each time point and moderate stability coefficients over time. Based on these findings, it was concluded that the Korean version of the Grit-S demonstrated satisfactory psychometric properties and exhibited MI across gender and time in Korean adolescents.

An Efficient Deep Neural Network to Classify Large 3D Images With Small Objects.

3D imaging enables accurate diagnosis by providing spatial information about organ anatomy. However, using 3D images to train AI models is computationally challenging because they consist of 10x or 100x more pixels than their 2D counterparts. To be trained with high-resolution 3D images, convolutional neural networks resort to downsampling them or projecting them to 2D. We propose an effective alternative, a neural network that enables efficient classification of full-resolution 3D medical images. Compared to off-the-shelf convolutional neural networks, our network, 3D Globally-Aware Multiple Instance Classifier (3D-GMIC), uses 77.98%-90.05% less GPU memory and 91.23%-96.02% less computation. While it is trained only with image-level labels, without segmentation labels, it explains its predictions by providing pixel-level saliency maps. On a dataset collected at NYU Langone Health, including 85,526 patients with full-field 2D mammography (FFDM), synthetic 2D mammography, and 3D mammography, 3D-GMIC achieves an AUC of 0.831 (95% CI: 0.769-0.887) in classifying breasts with malignant findings using 3D mammography. This is comparable to the performance of GMIC on FFDM (0.816, 95% CI: 0.737-0.878) and synthetic 2D (0.826, 95% CI: 0.754-0.884), which demonstrates that 3D-GMIC successfully classified large 3D images despite focusing computation on a smaller percentage of its input compared to GMIC. Therefore, 3D-GMIC identifies and utilizes extremely small regions of interest from 3D images consisting of hundreds of millions of pixels, dramatically reducing associated computational challenges. 3D-GMIC generalizes well to BCS-DBT, an external dataset from Duke University Hospital, achieving an AUC of 0.848 (95% CI: 0.798-0.896).

Network structure of symptomatology of adult attention-deficit hyperactivity disorder in patients with mood disorders.

Patients with mood disorders commonly manifest comorbid psychiatric disorders, including attention-deficit/hyperactivity disorder (ADHD). However, few studies have evaluated ADHD symptoms in this population. The current study aimed to explore the network structure of ADHD symptomology and identify central symptoms in patients with mood disorders. The Korean version of the Adult ADHD Self-Report Scale was used to assess the overall ADHD symptoms in 1,086 individuals diagnosed with mood disorders (major depressive disorder [n = 373], bipolar I disorder [n = 314], and bipolar II disorder [n = 399]). We used exploratory graph analysis to detect the number of communities, and the network structure was analyzed using regularized partial correlation models. We identified the central ADHD symptom using centrality indices. Network comparison tests were conducted with different subgroups of patients with mood disorders, including three mood diagnosis groups, between the patients who met the diagnostic criteria for ADHD [ADHD-suspected, n = 259] in their self-report and the others [ADHD-non-suspected, n = 827], and groups with high [n = 503] versus low [n = 252] levels of depressive state. The network analysis detected four communities: disorganization, agitation/restlessness, hyperactivity/impulsivity, and inattention. The centrality indices indicated that "feeling restless" was the core ADHD symptom. The result was replicated in the subgroup analyses within our clinically diverse population of mood disorders, encompassing three presentations: Patients with suspected ADHD, patients without suspected ADHD, and patients with a high depressive state. Our findings reveal that "feeling restless" is the central ADHD symptom. The treatment intervention for "feeling restless" may thus play a pivotal role in tackling ADHD symptoms in adult patients with mood disorders.

3D-Printed Auxetic Skin Scaffold for Decreasing Burn Wound Contractures at Joints.

For patients with severe burns that consist of contractures induced by fibrous scar tissue formation, a graft must adhere completely to the wound bed to enable wound healing and neovascularization. However, currently available grafts are insufficient for scar suppression owing to their nonuniform pressure distribution in the wound area. Therefore, considering the characteristics of human skin, which is omnidirectionally stretched via uniaxial stretching, we proposed an auxetic skin scaffold with a negative Poisson's ratio (NPR) for tight adherence to the skin scaffold on the wound bed site. Briefly, a skin scaffold with the NPR effect was fabricated by creating a fine pattern through 3D printing. Electrospun layers were also added to improve adhesion to the wound bed. Fabricated skin scaffolds displayed NPR characteristics (-0.5 to -0.1) based on pulling simulation and experiment. Finger bending motion tests verified the decreased marginal forces (<50%) and deformation (<60%) of the NPR scaffold. In addition, the filling of human dermal fibroblasts in most areas (>95%) of the scaffold comprising rarely dead cells and their spindle-shaped morphologies revealed the high cytocompatibility of the developed scaffold. Overall, the developed skin scaffold may help reduce wound strictures in the joints of patients with burns as it exerts less pressure on the wound margin.

Validation of the Short Form of the Mood Instability Questionnaire-Trait (MIQ-T-SF) in the Korean General Population.

OBJECTIVE: Mood instability (MI) is a clinically significant trait associated with psychiatric disorders. However, there are no concise measurements to evaluate MI. The initial Mood Instability Questionnaire-Trait (MIQ-T) was developed to fill this gap. The current study aimed to create a short form of MIQ-T (MIQ-T-SF) that measures MI with high validity and reliability in the Korean general population. METHODS: Of the 59 items in the MIQ-T, 17 items were chosen for the MIQ-T-SF following the factor analysis process. In total, 540 participants completed the MIQ-T-SF. Cronbach's alpha and McDonald's omega were used to evaluate reliability. Exploratory factor analysis (EFA) and confirmatory factor analysis (CFA) were used to determine construct validity. Concurrent validity was confirmed via comparisons with Personality Assessment Inventory-Borderline Features Scale. Measurement invariance across gender and age groups was confirmed before analyzing differences in scores using Kruskal-Wallis test. RESULTS: The MIQ-T-SF displayed expected correlations and high internal consistency (α=0.71-0.90, Ωt=0.72-0.92). Using EFA and CFA, a five-factor structure was confirmed. Measurement invariance was supported, and gender differences were observed. CONCLUSION: The MIQ-T-SF is an accurate and reliable method to detect MI in the Korean general population. The study's results offer new perspectives for future studies on MI.

Genetic network structure of 13 psychiatric disorders in the general population.

Psychiatric disorders frequently co-occur and share common symptoms and genetic backgrounds. Previous research has used genome-wide association studies to identify the interrelationships among psychiatric disorders and identify clusters of disorders; however, these methods have limitations in terms of their ability to examine the relationships among disorders as a network structure and their generalizability to the general population. In this study, we explored the network structure of the polygenic risk score (PRS) for 13 psychiatric disorders in a general population (276,249 participants of European ancestry from the UK Biobank) and identified communities and the centrality of the network. In this network, the nodes represented a PRS for each psychiatric disorder and the edges represented the connections between nodes. The psychiatric disorders comprised four robust communities. The first community included attention-deficit hyperactivity disorder, autism spectrum disorder, major depressive disorder, and anxiety disorder. The second community consisted of bipolar I and II disorders, schizophrenia, and anorexia nervosa. The third group included Tourette's syndrome and obsessive-compulsive disorder. Cannabis use disorder, alcohol use disorder, and post-traumatic stress disorder make up the fourth community. The PRS of schizophrenia had the highest values for the three metrics (strength, betweenness, and closeness) in the network. Our findings provide a comprehensive genetic network of psychiatric disorders and biological evidence for the classification of psychiatric disorders.

A Latent Class Analysis of Suicidal Behaviors in Adolescents.

OBJECTIVE: The purpose of the present study was to identify adolescents' suicide subgroups using five indicators (depression, anxiety, suicide ideation, and planned and attempted suicide), and explore the distinctive features of each subgroup. METHODS: This study included 2,258 teenagers from four schools. Both adolescents and their parents, who voluntarily agreed to participate in the study, completed a series of self-reported questionnaires on depression, anxiety, suicide, self-harm, self-esteem, impulsivity, childhood maltreatment, and deviant behaviors. The data were analyzed using latent class analysis, a person-centered method. RESULTS: Four classes were detected: "high risk for suicide without distress," "high risk for suicide with distress," "low risk for suicide with distress," and "healthy." The "high risk for suicide with distress" class was the most severe on all evaluated psychosocial risk factors, namely, impulsivity, low self-esteem, self-harming behaviours, deviant behaviour problems, and childhood maltreatment, followed by "high risk for suicide without distress." CONCLUSION: This study identified two high risk subgroups for adolescent' suicidality, "high risk for suicide with or without distress." Both high risk subgroups for suicide showed higher scores for all psychosocial risk factors than low risk subgroups for suicide. Our findings suggest that special attention needs to be paid to the latent class "high risk for suicide without distress," as this group's "cry for help" might be relatively difficult to detect. Specific interventions for each group (e.g., distress safety plans for "suicidal potential with or without emotional distress") need to be developed and implemented.

A Competition, Benchmark, Code, and Data for Using Artificial Intelligence to Detect Lesions in Digital Breast Tomosynthesis.

Importance: An accurate and robust artificial intelligence (AI) algorithm for detecting cancer in digital breast tomosynthesis (DBT) could significantly improve detection accuracy and reduce health care costs worldwide. Objectives: To make training and evaluation data for the development of AI algorithms for DBT analysis available, to develop well-defined benchmarks, and to create publicly available code for existing methods. Design, Setting, and Participants: This diagnostic study is based on a multi-institutional international grand challenge in which research teams developed algorithms to detect lesions in DBT. A data set of 22 032 reconstructed DBT volumes was made available to research teams. Phase 1, in which teams were provided 700 scans from the training set, 120 from the validation set, and 180 from the test set, took place from December 2020 to January 2021, and phase 2, in which teams were given the full data set, took place from May to July 2021. Main Outcomes and Measures: The overall performance was evaluated by mean sensitivity for biopsied lesions using only DBT volumes with biopsied lesions; ties were broken by including all DBT volumes. Results: A total of 8 teams participated in the challenge. The team with the highest mean sensitivity for biopsied lesions was the NYU B-Team, with 0.957 (95% CI, 0.924-0.984), and the second-place team, ZeDuS, had a mean sensitivity of 0.926 (95% CI, 0.881-0.964). When the results were aggregated, the mean sensitivity for all submitted algorithms was 0.879; for only those who participated in phase 2, it was 0.926. Conclusions and Relevance: In this diagnostic study, an international competition produced algorithms with high sensitivity for using AI to detect lesions on DBT images. A standardized performance benchmark for the detection task using publicly available clinical imaging data was released, with detailed descriptions and analyses of submitted algorithms accompanied by a public release of their predictions and code for selected methods. These resources will serve as a foundation for future research on computer-assisted diagnosis methods for DBT, significantly lowering the barrier of entry for new researchers.

A latent profile approach for classifying internet gamers based on motives for online gaming.

Background and aims: Online gaming motives have proven to be useful in differentiating problematic engagement in online gaming. However, the mixture modeling approach for classifying problematic subtypes based on gaming motives remains limited. This study attempted to differentiate heterogeneous online gamers into more homogenous subtypes based on gaming motives using latent profile analysis (LPA). We also compared various psychological and gaming/leisure related variables across the derived profiles. Methods: A total of 674 Korean online game users (mean age = 21.81 years, male = 76%) completed self-report questionnaires, including the Korean version of the Motives for Online Gaming Questionnaire (K-MOGQ). After the LPA, the relationships between latent profile membership and auxiliary variables were explored. Results: Four latent profiles were identified, that were further classified into one problematic (highly motivated-dissatisfied gamer), one highly engaged (highly motivated-satisfied gamer), and two casual (moderately-motivated casual gamer and lowly-motivated casual gamer) gamer profiles. Inter-profile comparisons revealed that highly motivated-dissatisfied gamer had the most pathological profile, characterized by high Internet gaming disorder (IGD) tendency, neuroticism, and impulsivity, but the lowest recreation motive. While highly motivated-satisfied gamer also demonstrated a heightened IGD tendency, they showed positive patterns of psychological and gaming/leisure-related variables, which indicated they could be better considered as high engaged instead of problematic gamers. Discussion and conclusions: These results indicate that the recreation motive, in addition to fantasy or escape motives, is an important factor in differentiating maladaptive online gamers. Classifying online gamers based on gaming motives can contribute to a clearer conceptualization of heterogeneous gamers, paving the way for individualized assessment and treatment planning.

Understanding of Depressive Symptomatology across Major Depressive Disorder and Bipolar Disorder: A Network Analysis.

Background and Objectives: Depressive symptoms are prominent in both major depressive disorder (MDD) and bipolar disorder (BD). However, comparative research on the network structure of depressive symptoms in these two diagnostic groups has been limited. This study aims to compare the network structure of depressive symptoms in MDD and BD, providing a deeper understanding of the depressive symptomatology of each disorder. Materials and Methods: The Zung Self-Rating Depressive Scale, a 20-item questionnaire, was administered to assess the depressive symptoms in individuals with MDD (n = 322) and BD (n = 516). A network analysis was conducted using exploratory graph analysis (EGA), and the network structure was analyzed using regularized partial correlation models. To validate the dimensionality of the Zung SDS, principal component analysis (PCA) was adopted. Centrality measures of the depressive symptoms within each group were assessed, followed by a network comparison test between the two groups. Results: In both diagnostic groups, the network analysis revealed four distinct categories, aligning closely with the PCA results. "Depressed affect" emerged as the most central symptom in both MDD and BD. Furthermore, non-core symptoms, "Personal devaluation" in MDD and "Confusion" in BD, displayed strong centrality. The network comparison test did not reveal significant differences in the network structure between MDD and BD. Conclusions: The absence of significant differences in the network structures between MDD and BD suggests that the underlying mechanisms of depressive symptoms may be similar across these disorders. The identified central symptoms, including "Depressed affect", in both disorders and the distinct non-core symptoms in each highlight the complexity of the depressive symptomatology. Future research should focus on validating these symptoms as therapeutic targets and incorporate various methodologies, including non-metric dimension reduction techniques or canonical analysis.

PCL/Sodium-Alginate Based 3D-Printed Dual Drug Delivery System with Antibacterial Activity for Osteomyelitis Therapy.

Chronic osteomyelitis is mostly caused by bacteria such as S. aureus, and is often treated with oral antibiotics or injections to suppress the bacteria. In severe cases, however, surgical treatment using antibiotic beads and metal supports may be required. In these surgeries, bacterial attachment to the metal may lead to biofilm formation and reduce antibiotics' penetration to the bacteria. Reoperation must be performed to prevent bacterial inflammatory reactions and antibiotic resistance. Thus, in this study, we developed a dual-drug-releasing PCL/sodium-alginate-based 3D-printed scaffold to effectively treat osteomyelitis by removing the biofilm. We proposed an antibiotic-loaded biodegradable polymer scaffold using 3D printing, which was encapsulated by a second antibiotic-containing hydrogel. Then, we successfully established a dual-drug-based scaffold that consisted of a cefazolin (CFZ)-containing polycaprolactone 3D scaffold and a rifampicin (RFP)-loaded alginate hydrogel encapsulating the 3D scaffold. Our scaffold showed a synergistic effect, whereby biofilm formation was inhibited by RFP, which is an external drug, and bacterial activity was inhibited by CFZ, which is an internal drug that increases antibacterial activity. We also confirmed that the dual-drug-based scaffold did not affect the proliferation of human osteoblasts. Our findings suggest that this dual drug delivery system may serve as a new therapeutic treatment for osteomyelitis that overcomes the limitations of individual drugs.

The Relationship Between Employees' Daily Customer Injustice and Customer-Directed Sabotage: Cross-Level Moderation Effects of Emotional Stability and Attentiveness.

Customer injustice has received considerable attention in the field of organizational behavior because it generates a variety of negative outcomes. Among possible negative consequences, customer-directed sabotage is the most common reaction, which impacts individuals' well-being and the prosperity of organizations. To minimize such negative consequences, researchers have sought to identify boundary conditions that could potentially attenuate the occurrence of customer-directed sabotage. In this study, we explore potential attenuation effects of emotional stability and attentiveness on the customer injustice-sabotage linkage. The results showed emotional stability and attentiveness moderate the relationship between customer injustice and customer-directed sabotage. Specifically, the representatives with higher (vs. lower) emotional stability or higher (vs. lower) attentiveness are less likely to engage in customer-directed sabotage when they experience customer injustice. Moreover, there is a three-way interaction among daily customer injustice, emotional stability, and attentiveness that predicts daily customer-directed sabotage. Theoretical and practical contributions, limitations, and directions for future development are also discussed.

Reducing False-Positive Biopsies using Deep Neural Networks that Utilize both Local and Global Image Context of Screening Mammograms.

Breast cancer is the most common cancer in women, and hundreds of thousands of unnecessary biopsies are done around the world at a tremendous cost. It is crucial to reduce the rate of biopsies that turn out to be benign tissue. In this study, we build deep neural networks (DNNs) to classify biopsied lesions as being either malignant or benign, with the goal of using these networks as second readers serving radiologists to further reduce the number of false-positive findings. We enhance the performance of DNNs that are trained to learn from small image patches by integrating global context provided in the form of saliency maps learned from the entire image into their reasoning, similar to how radiologists consider global context when evaluating areas of interest. Our experiments are conducted on a dataset of 229,426 screening mammography examinations from 141,473 patients. We achieve an AUC of 0.8 on a test set consisting of 464 benign and 136 malignant lesions.

Artificial intelligence system reduces false-positive findings in the interpretation of breast ultrasound exams.

Though consistently shown to detect mammographically occult cancers, breast ultrasound has been noted to have high false-positive rates. In this work, we present an AI system that achieves radiologist-level accuracy in identifying breast cancer in ultrasound images. Developed on 288,767 exams, consisting of 5,442,907 B-mode and Color Doppler images, the AI achieves an area under the receiver operating characteristic curve (AUROC) of 0.976 on a test set consisting of 44,755 exams. In a retrospective reader study, the AI achieves a higher AUROC than the average of ten board-certified breast radiologists (AUROC: 0.962 AI, 0.924 ± 0.02 radiologists). With the help of the AI, radiologists decrease their false positive rates by 37.3% and reduce requested biopsies by 27.8%, while maintaining the same level of sensitivity. This highlights the potential of AI in improving the accuracy, consistency, and efficiency of breast ultrasound diagnosis.

An artificial intelligence system for predicting the deterioration of COVID-19 patients in the emergency department.

During the coronavirus disease 2019 (COVID-19) pandemic, rapid and accurate triage of patients at the emergency department is critical to inform decision-making. We propose a data-driven approach for automatic prediction of deterioration risk using a deep neural network that learns from chest X-ray images and a gradient boosting model that learns from routine clinical variables. Our AI prognosis system, trained using data from 3661 patients, achieves an area under the receiver operating characteristic curve (AUC) of 0.786 (95% CI: 0.745-0.830) when predicting deterioration within 96 hours. The deep neural network extracts informative areas of chest X-ray images to assist clinicians in interpreting the predictions and performs comparably to two radiologists in a reader study. In order to verify performance in a real clinical setting, we silently deployed a preliminary version of the deep neural network at New York University Langone Health during the first wave of the pandemic, which produced accurate predictions in real-time. In summary, our findings demonstrate the potential of the proposed system for assisting front-line physicians in the triage of COVID-19 patients.

An interpretable classifier for high-resolution breast cancer screening images utilizing weakly supervised localization.

Medical images differ from natural images in significantly higher resolutions and smaller regions of interest. Because of these differences, neural network architectures that work well for natural images might not be applicable to medical image analysis. In this work, we propose a novel neural network model to address these unique properties of medical images. This model first uses a low-capacity, yet memory-efficient, network on the whole image to identify the most informative regions. It then applies another higher-capacity network to collect details from chosen regions. Finally, it employs a fusion module that aggregates global and local information to make a prediction. While existing methods often require lesion segmentation during training, our model is trained with only image-level labels and can generate pixel-level saliency maps indicating possible malignant findings. We apply the model to screening mammography interpretation: predicting the presence or absence of benign and malignant lesions. On the NYU Breast Cancer Screening Dataset, our model outperforms (AUC = 0.93) ResNet-34 and Faster R-CNN in classifying breasts with malignant findings. On the CBIS-DDSM dataset, our model achieves performance (AUC = 0.858) on par with state-of-the-art approaches. Compared to ResNet-34, our model is 4.1x faster for inference while using 78.4% less GPU memory. Furthermore, we demonstrate, in a reader study, that our model surpasses radiologist-level AUC by a margin of 0.11.

Detecting Conditional Dependence Using Flexible Bayesian Latent Class Analysis.

A fundamental assumption underlying latent class analysis (LCA) is that class indicators are conditionally independent of each other, given latent class membership. Bayesian LCA enables researchers to detect and accommodate violations of this assumption by estimating any number of correlations among indicators with proper prior distributions. However, little is known about how the choice of prior may affect the performance of Bayesian LCA. This article presents a Monte Carlo simulation study that investigates (1) the utility of priors in a range of prior variances (i.e., strongly non-informative to strongly informative priors) in terms of Type I error and power for detecting conditional dependence and (2) the influence of imposing approximate independence on model fit of Bayesian LCA. Simulation results favored the use of a weakly informative prior with large variance-model fit (posterior predictive p-value) was always satisfactory when the class indicators were either independent or dependent. Based on the current findings and the additional literature, this article offers methodological guidelines and suggestions for applied researchers.