NILM model for multi-appliance power disaggregation based on the combination of common features and individual features

Spatiotemporal prediction of aeropollen concentration using tree-based machine learning.

Integrating probabilistic trees and causal networks for clinical and epidemiological data.

Prevalence of dysmorphic sacral features in the general population.

Understanding causal mechanisms in deep learning is essential for clinical adoption, where interpretability and reliability are critical. Most existing AI systems act as black boxes, limiting transparency in medicine. We propose a causal inference framework integrated into neural networks to assess the influence of individual features on predictions. Using a prospective pediatric ophthalmology cohort of over 3000 children with longitudinal follow-up, our method estimates direct and indirect causal effects through intervention. Applied to myopia progression in children, our approach not only achieved good performance but also identified clinically plausible causal pathways. Refutation experiments with multiple falsification strategies confirm the robustness and reliability of causal effects. The approach is model-agnostic and suitable for digital health interventions requiring explainability. By incorporating unit-level causal reasoning into deep learning, this work advances transparent and reliable AI systems aligned with the goals of precision medicine and equitable healthcare.

Abstract Genomic testing is the standard for guiding adjuvant treatment for patients with early-stage HR+/HER2-IBC. Prognostic testing strategies using the biopsy specimen are needed to individualize patient management in the neoadjuvant and possibly adjuvant setting. The current objective was to clinically validate the PreciseBreast Biopsy test (PDxBRBx) which includes the patient’s age and morphologic features, derived from a standard H&E stained IBC biopsy slide, to predict recurrence risk earlier in the treatment planning and management process. Methods 1788 patients (pts) with IBC from the Mount Sinai Health System were identified (2004-2016) with median follow-up of 6 years (yrs). 60%-40% balanced training (surgical) and validation (biopsy) cohorts were generated. The PreciseBreast Biopsy test was performed using H&E stained slides that were imaged (40X) with a Philips Ultra-Fast scanner. Images were deconstructed to morphologic analytes using an AI-enabled platform to quantify tumor cell and tissue architectural features. Age is the only clinical factor available at the time of diagnostic biopsy and is algorithmically combined with 8 morphologic features (AI-grade model) that stratifies pts into low or high risk of recurrence. Risk stratification for invasive breast cancer free survival (IBCFS) was assessed by concordance index (C-index), area under the curve (AUC), Kaplan-Meier analysis, hazards ratios (HR), sensitivity (Se), specificity (Sp), negative predictive value (NPV), and positive predictive value (PPV). Matched excisional specimens from the biopsy test cohort were evaluated with the PreciseBreast excision test for risk score correlation using Cohen’s Kappa and Odds Ratio. Results In the training surgical cohort (n=1012): median age 59 yrs, 80% grade 2/3, 93% stage 1/2, 77% pN0, 23% pN1-3, ER+(87%)/PR+(81%)/HER2-(88%), 82 (8%) triple negative, 91 (9%) HER2+, with a 14% event rate and median 6-yrs of follow-up. PDxBRBx (Age + AI-grade) yielded a C-index of 0.72 (95% CI, 0.69-0.78) vs age C-index 0.61 (95% CI, 0.55 - 0.66) vs AI-grade 0.69 (95% CI, 0.64 - 0.73). Applying a risk score cut-off of 73 (scale 0-100) stratified patients as low- or high-risk, respectively, with a HR of 4.49 (95% CI 2.96 - 6.8, p<0.001) for predicting IBCFS, with Se 0.70, Sp 0.67, NPV 0.93, and PPV 0.25. In the biopsy validation cohort (n=776): median age 60 yrs, 43% Grade 2, ER+ (87%)/PR+(81%)/HER2-(88%), 56 (7%) triple negative, 97 (12.5%) HER2+ with a 14% event rate. PDxBRBx yielded a C-index of 0.73 (95% CI, 0.69-0.78) vs age 0.64 (95% CI, 0.57-0.70) vs AI-grade 0.69 (95% CI, 0.64-0.74). Patients stratified by a risk score of 73 had a HR of 4.49 (95% CI, 2.96-6.8, p<0.0001) for predicting IBCFS within 6 years, with Se 0.70, Sp 0.67, NPV 0.93, and PPV 0.25. The HR for AI-grade was 2.45 (95% CI 1-68-3.59, p<0.0001). All individual morphologic biopsy features including mitotic figure quantitation, nuclear pleomorphism, tumor-stromal ratio, lymphocytic content, and tumor architecture were statistically significant predictors of event risk (all p<0.05, most p<0.001). Comparison of biopsy (n=776) and matched excisional PreciseBreast risk scores showed substantial agreement, with Cohen’s κ = 0.57 (95% CI 0.51-0.62, p<0.0001) and an odds ratio of 31.4 (95% CI 18.8-55.1, p<0.0001). Conclusion We clinically validated a breast biopsy digital pathology-based AI prognostic test, PreciseBreast Biopsy, which successfully predicted IBCFS within 6 years. The test is designed to assist in the accurate characterization of clinical pathologic risk and patient management at the time of diagnosis. Additional studies in the neoadjuvant and possibly adjuvant settings will further refine the impact of these results on treatment selection. Citation Format: G. Fernandez, S. Vaisman, A. Sainath Madduri, R. Scott, M. Prastawa, X. Zhang, M. Donovan. Clinical validation of an Artificial Intelligence digital pathology-based prognostic test to predict risk of recurrence using biopsy specimens from patients with invasive breast cancer [abstract]. In: Proceedings of the San Antonio Breast Cancer Symposium 2025; 2025 Dec 9-12; San Antonio, TX. Philadelphia (PA): AACR; Clin Cancer Res 2026;32(4 Suppl):Abstract nr PS1-13-13.

To examine potential differences in electroencephalogram (EEG) dynamic functional connectivity between patients with major depressive disorder (MDD) and healthy controls (HC), and thereby enhance the effectiveness of depression identification. This study presents a novel approach that combines EEG microstate analysis with functional connectivity networks. Resting-state 19-channel EEG data were obtained from 36 participants (17 healthy controls and 19 patients with depression). Through microstate analysis, significant inter-group differences were observed in the average durations of microstates A and C. Subsequently, EEG segments corresponding to microstate classes A and C were extracted. Following the surface Laplacian transformation, the phase locking value (PLV) was applied to construct functional connectivity networks, and their topological characteristics were extracted. Based on the analysis of network indicators (node degree, clustering coefficient, local efficiency, and global efficiency), global and nodal features showing significant group differences were screened and fused with equal weighting. The classification performance of the fused features and individual features was then assessed using three models: Support Vector Machine (SVM), Backpropagation Neural Network (BP), and K-Nearest Neighbors (KNN). The findings indicate that network features derived from microstate C exhibited higher discriminative ability. Across all classification models, node degree features consistently outperformed other individual topological attributes in recognition accuracy, with the KNN model achieving the highest average accuracy of 96.48%. Furthermore, the fused feature set, incorporating more comprehensive EEG information, showed improved classification performance across all models, exceeding the results obtained using any single feature. The average accuracy reached 97.35% under different model configurations. Dynamic analysis of brain networks can effectively distinguish patients with depression from healthy controls. This study not only provides a basis for exploring dynamic activities of brain regions associated with depression, but also offers potential objective physiological indicators for disease diagnosis.

Hot subdwarfs (hot sds) are compact evolved stars near the extreme horizontal branch, and they are key to understanding stellar evolution and the UV excess in galaxies. In this work, we extend our previous analysis of Gaia XP spectra of hot sd stars to a much larger sample, enabling a comprehensive study of their physical and binary properties. Our goal is to identify patterns in Gaia XP spectra, investigate binarity, and assess the influence of parameters such as temperature, helium abundance, and variability. We analysed ∼20 , 000 hot sd candidates selected from the literature, combining Gaia XP data with published parameters. We applied the uniform manifold approximation and projection technique to the XP coefficients, which represent the Gaia XP spectra in a compact feature-based form, to construct a similarity map. We then used self-organising maps and convolutional neural networks (CNNs) to classify spectra as binaries or singles and as cool/He-poor or hot/He-rich. The spectra were normalised using asymmetric least squares baseline fitting to emphasise individual spectral features. We found that the BP--RP colour dominates the similarity map, with additional influence from temperature, helium abundance, and variability. Most binaries, which were identified via the Virtual Observatory SED analyser, cluster in two filaments linked to main sequence companions. The CNN classification suggests a strong correlation between variability and binarity, with binary fractions exceeding 60% for active hot sds. The Gaia XP spectra combined with dimensionality reduction and machine learning effectively revealed patterns in hot sd properties. Our findings indicate that binarity and environmental density strongly shape the evolutionary paths of hot subdwarfs. We identified possible contamination by main sequence and cataclysmic variable stars in our base sample.

ABSTRACT Deepfake videos, employing artificial intelligence techniques to create highly realistic but fabricated content, have emerged as a major concern for the society. Proposed paper develops deep learning model called, EnsembleResnet, that ensembles multiple feature descriptors, and then judicially select optimal features for deepfake video detection. Initially, individual features such as DFT (Discrete Fourier Transform), DCT (Discrete Cosine Transform), SIFT (Scale-Invariant Feature Transform), and Gabor Filter are trained on ResNet-18 model to classify real and fake videos. Afterward, ensemble learning of optimal feature descriptors is trained on the same model to improve detection accuracy. Experimental evaluations on diverse datasets: FF++ and Celeb-DF show potency of ensemble learning in identifying real and fake videos. Experimental results show an accuracy with paired feature descriptors (SIFT + Gabor + DCT) on datasets FF++ (raw data, high quality, low quality) is (99.80%, 99.25%, 99.79%), and on Celeb-DF dataset is 98.01%, respectively.

This study evaluated the adaptation of relapse prevention treatment, originally for substance use, to address problematic gaming in children and adolescents. The goal was to understand the feasibility of this treatment in child and adolescent mental health services from the clinicians’ perspective. Six clinicians, four women and two men aged 37–47 with 4–20 years of work experience, were interviewed post-treatment. Key findings included the importance of individual patient and clinician features, successes and barriers in treatment, and ideas for future development. Cognitive maturity and neurodevelopmental disorders, such as ADHD and Autism, significantly influenced treatment efficacy. Parental involvement and addressing the challenges of online sessions were deemed crucial. Tailoring treatment to individual needs, allowing time for follow-up, setting realistic expectations for change, and parental involvement were highlighted for positive outcomes. The study contributes to the development of knowledge regarding treatments for young people’s behavioral and health problems, such as gaming problems.

MR findings of dysembryoplastic neuroepithelial tumors and low-grade astrocytomas.

The aim of the study was to identify the peculiarities of discourse predicativity and the role of punctuation in sentence-expressions using works of different national and genre traditions. The theoretical analysis of predicativity as a key textual category and the comparative-analytical method were applied for the study of literary works. The research material included texts reflecting a variety of genres and cultural contexts, which made it possible to identify general and specific patterns. The results of the study showed that predicativity was the main property of the text, providing its logical completeness and coherence, forming a link between sentence elements and context and maintaining thematic and semantic coherence. It was found that punctuation performed an important function in text structuring, contributing to the allocation of key semantic accents, organising rhythm, and enhancing emotional expressiveness. The comparative analysis of the works revealed that in all the texts, predicativity contributed to the creation of a coherent discourse, but its use varied depending on the national and genre context. In some cases, the emphasis was on emotional intensity and expressiveness, while in others it was on logical simplicity and clarity of the narrative. Punctuation, as a tool enhancing predicativity, also manifested itself in different ways: in some works, it emphasised drama and symbolism, while in others it facilitated the perception of the text and highlighted logical connections. The generalisation of the results confirmed the universality of predicativity and punctuation as means of textual organisation, reflecting both individual features of the author's intention and cultural and stylistic traditions. These aspects were recognised as key in creating a coherent, logically structured, and expressive text, opening up new opportunities for studies across different genres and cultural contexts.

ABSTRACT An agent’s epistemic prospects depend on a combination of that agent’s individual characteristics and features of that agent’s epistemic environment. Such factors cannot always be cleanly separated. Often, individual characteristics impact agents’ epistemic prospects by shaping the epistemic environments in which individuals find themselves. In particular, features of individuals often repel or attract certain sorts of information, a phenomenon I label epistemic magnetism. I argue that epistemic magnetism is a ubiquitous and underrecognized phenomenon that sometimes promotes and sometimes frustrates the achievement of positive epistemic outcomes. Then, I consider a series of simple proposals concerning what distinguishes between beneficial and harmful forms of epistemic magnetism. I argue that these proposals cannot capture the impacts of epistemic magnetism. Instead, I offer a series of principles that serve to roughly characterize the consequences of this phenomenon. I conclude with some remarks on why epistemologists have thus far tended to overlook epistemic magnetism.

In adolescents, identifying objective biomarkers for treatment response is crucial for the development of effective interventions. Voice-based biomarkers have recently shown potential to capture treatment-related changes in Major Depressive Disorder (MDD). While prior studies have been cross-sectional experiments with single speech sample, this study addresses a critical gap by evaluating intra-patient changes in speech over treatment period, providing insight into how these voice biomarkers evolve within individuals. We collected pre- and post-treatment voice samples from 48 adolescent MDD patients. We hypothesized that deep learning models could detect clinically meaningful changes in depressive states during treatment. Therefore, we compared machine learning and deep learning models for depressive classification. Additionally, we introduced the Dual Voice-based Depressive State Analysis (DVDSA) method to categorize intra-patient depressive state changes as recovery, worsening, or unchanged, highlighting the deep learning models' ability to detect these variations. Among the acoustic features, only the fundamental frequency exhibits significant changes between pre- and post-treatment states after Holm-Bonferroni correction. Machine learning models demonstrate limited performance in distinguishing treatment states, with the best F1-score reaching 65.83%. In contrast, deep learning model, particularly WavLM, achieves remarkably higher performance in binary classification, with an F1-score of 78.05%. The WavLM maintains robust performance, when applied to the DVDSA method, achieves an F1-score of 70.58%. These findings suggest that machine learning models and individual acoustic features may not sufficiently capture treatment-related changes in MDD patients. This study underscores the value of deep learning models using the DVDSA method, addressing the limitations of pre- and post-treatment classification and highlighting their potential to advance personalized treatment strategies for adolescent MDD.

A previous study reported that among patients with complete Kawasaki Disease (KD), those exhibiting all six principal clinical features were more likely to develop coronary artery (CA) sequelae than those exhibiting only five features. We aimed to determine which specific features are associated with CA sequelae. This retrospective cohort study analyzed 14,732 patients diagnosed with complete KD across Japan from January 2019 to March 2020. Separate multivariable conditional logistic regression analyses were performed to evaluate relative risk for CA sequelae in patients with all six principal clinical features, compared individually to those lacking each specific feature. 7234 (49.1%) exhibited all six principal clinical features, while 7498 (50.9%) presented with five features. CA sequelae occurred in 2.1% of those with six features versus 1.7% with five. Multivariable conditional logistic regression analysis determined that patients with conjunctival injection were significantly more likely to develop CA sequelae compared with those lacking it (adjusted odds ratio [95% confidence interval], 3.6 [1.3-10.1]). Among patients with complete KD, the absence of conjunctival injection-a relatively rare presentation-was associated with a lower cumulative incidence of CA sequelae. This finding may help identify distinct low-risk phenotypes of KD and support risk stratification. This study emphasizes the importance of feature-specific risk for coronary artery (CA) sequelae among patients with complete Kawasaki Disease (KD). We found that among patients with complete KD, those with conjunctival injection were more likely to develop CA sequelae than were those lacking it. The absence of conjunctival injection-a relatively rare presentation in KD-is associated with a markedly lower cumulative incidence of CA sequelae. This finding may help identify a distinct low-risk phenotype of KD and aid risk stratification.

Wafer defect recognition is crucial for semiconductor manufacturing. During wafer fabrication, mixed-type defects with various morphologies, along with random noise, introduce data uncertainty into the wafer defect recognition process. However, most existing recognition methods fail to account for this uncertainty, leading to potentially unreliable results. In this paper, we propose a trustworthy mixed-type wafer defect recognition method (TMWDM) based on evidence theory, which models and leverages the evidential support of individual features during training to enhance recognition robustness. TMWDM consists of two main components: a directional evidential feature discrimination network and an evidential feature differentiated learning strategy. The former converts extracted features into evidence representations, enabling the evaluation of the directional alignment and strength of each feature’s evidential support under varying input and noise conditions. The latter employs an uncertainty-sensitive loss function that incorporates a penalty term measured using evidential features to optimise model learning under data uncertainty. Extensive experiments demonstrate that TMWDM consistently achieves over 98.80% in different recognition metrics across all mixed-type wafer defect scenarios, outperforming state-of-the-art methods. It also generalises well to WM-811 K and shows significant gains in ablation study.

Automated lung texture analysis for assessing interstitial lung disease in systemic sclerosis: Diagnostic accuracy in photon-counting-detector and conventional energy-integrating-detector CT.

An international, expert-based, multispecialty Delphi consensus document on stroke risk stratification and the optimal management of patients with asymptomatic and symptomatic carotid stenosis.

Multiview-cooperated graph neural network enables novel multi-omics cancer subtype classification.

Neural correlates of conversational turn-taking perception in individuals at clinical high-risk for psychosis.