Unsupervised Learning Research Articles

Enhancing Fraud Detection Systems through Advanced Data Engineering Techniques

Aims: Fraud remains a persistent issue in various industries, particularly in finance, e-commerce, and healthcare, where traditional rule-based systems have struggled to keep pace with the evolving complexity of fraudulent activities. This study aims to develop an enhanced fraud detection framework by addressing the limitations of traditional rule-based systems, particularly in industries where sophisticated fraud schemes prevail. Study Design: The research utilizes advanced data engineering techniques, including big data analytics, machine learning, and real-time processing, to improve the accuracy and efficiency of fraud detection systems. Place and Duration of Study: The study was conducted over two years across industries with high fraud susceptibility, including financial services, e-commerce platforms, and healthcare organizations. Methodology: The framework integrates various data sources, including transaction logs, user behavior, and external fraud indicators. These datasets were pre-processed through data cleaning, feature engineering, and integration. Supervised and unsupervised machine learning models, such as Random Forest and Gradient Boosting, were applied to detect fraud patterns. Real-time data processing enabled immediate detection and response. The system continuously learned from historical data, adapting to new fraud tactics and improving detection over time. Results: The proposed framework demonstrated a significant improvement in fraud detection accuracy, with machine learning models achieving over 90% accuracy rates. There was also a 30% reduction in false positives compared to traditional methods, and detection times were shortened by 40%, enabling faster identification and mitigation of emerging fraud schemes. Conclusion: This study concludes that integrating advanced data engineering techniques with machine learning significantly enhances fraud detection systems' accuracy, scalability, and adaptability. While promising, further improvements are needed, particularly in addressing the evolving nature of fraud schemes and ensuring the scalability of real-time data processing. These areas present opportunities for future research and development.

A signal-diffusion-based unsupervised contrastive representation learning for spatial transcriptomics analysis.

Spatial transcriptomics allows for the measurement of high-throughput gene expression data while preserving the spatial structure of tissues and histological images. Integrating gene expression, spatial information, and image data to learn discriminative low-dimensional representations is critical for dissecting tissue heterogeneity and analyzing biological functions. However, most existing methods have limitations in effectively utilizing spatial information and high-resolution histological images. We propose a signal-diffusion-based unsupervised contrast learning method (SDUCL) for learning low-dimensional latent embeddings of cells/spots. SDUCL integrates image features, spatial relationships and gene expression information. We designed a signal diffusion microenvironment discovery algorithm, which effectively captures and integrates interaction information within the cellular microenvironment by simulating the biological signal diffusion process. By maximizing the mutual information between the local representation and the microenvironment representation of cells/spots, SDUCL learns more discriminative representations. SDUCL was employed to analyze spatial transcriptomics datasets from multiple species, encompassing both normal and tumor tissues. SDUCL performed well in downstream tasks such as clustering, visualization, trajectory inference, and differential gene analysis, thereby enhancing our understanding of tissue structure and tumor microenvironments. https://github.com/WeiMin-Li-visual/SDUCL. Supplementary data are available at Bioinformatics online.

Prospective and External Validation of an Ensemble Learning Approach to Sensitively Detect Intravenous Fluid Contamination in Basic Metabolic Panels.

Intravenous (IV) fluid contamination within clinical specimens causes an operational burden on the laboratory when detected, and potential patient harm when undetected. Even mild contamination is often sufficient to meaningfully alter results across multiple analytes. A recently reported unsupervised learning approach was more sensitive than routine workflows, but still lacked sensitivity to mild but significant contamination. Here, we leverage ensemble learning to more sensitively detect contaminated results using an approach which is explainable and generalizable across institutions. An ensemble-based machine learning pipeline of general and fluid-specific models was trained on real-world and simulated contamination and internally and externally validated. Benchmarks for performance assessment were derived from in silico simulations, in vitro experiments, and expert review. Fluid-specific regression models estimated contamination severity. SHapley Additive exPlanation (SHAP) values were calculated to explain specimen-level predictions, and algorithmic fairness was evaluated by comparing flag rates across demographic and clinical subgroups. The sensitivities, specificities, and Matthews correlation coefficients were 0.858, 0.993, and 0.747 for the internal validation set, and 1.00, 0.980, and 0.387 for the external set. SHAP values provided plausible explanations for dextrose- and ketoacidosis-related hyperglycemia. Flag rates from the pipeline were higher than the current workflow, with improved detection of contamination events expected to exceed allowable limits for measurement error and reference change values. An accurate, generalizable, and explainable ensemble-based machine learning pipeline was developed and validated for sensitively detecting IV fluid contamination. Implementing this pipeline would help identify errors that are poorly detected by current clinical workflows and a previously described unsupervised machine learning-based method.

Assessing and predicting the dynamics of land use/land cover in northern Bangladesh using cellular Automata-Markov chain model

ABSTRACT Rapid urbanization and land use changes significantly impact environmental sustainability and resource management, particularly in developing regions. Therefore, this study examines the spatiotemporal dynamics of land use and land cover (LULC) in Rangpur, Bangladesh, from 1991 to 2021 and projects future trends to 2041. Using supervised and unsupervised classification techniques, along with cellular automata and Markov-chain models, we assessed historical LULC changes and predicted future scenarios. Results show a 38.86% increase in built-up areas (BAs) and a 49.86% decrease in vegetation land (VL) during the study period, with classification accuracy above 87%. Projections indicate a further loss of over 210 km² of VL and an increase of more than 123 km² in urban areas by 2041. Notably, urban expansion is linked to the development of road networks, with significant growth from 115.06 km2 in 2021 to 124.33 km2 by 2041 within a 15-kilometer radius around the city center. These findings offer crucial insights for urban planning, emphasizing the need for sustainable strategies to manage urban expansion and protect environmental and socio-economic resilience in Rangpur.

Mapping global public perspectives on mRNA vaccines and therapeutics.

The development and rollout of mRNA vaccines during COVID-19 marked a significant advancement in vaccinology, yet public hesitation to vaccination was prevalent, indicating the potential risk that future mRNA-based medical innovations will fail to be adopted. Utilizing a combined approach of large language models with manual validation and unsupervised machine learning, we conducted a social listening analysis to assess attitudes towards mRNA vaccines and therapeutics on Twitter from June 2022 to May 2023, contrasting online perspectives with data from the Vaccine Adverse Event Reporting System. Our findings reveal widespread negative sentiment and a global lack of confidence in the safety, effectiveness, and trustworthiness of mRNA vaccines and therapeutics, with frequent discussions of severe vaccine side effects, rumors, and misinformation. This underscores the need for targeted communication strategies to foster acceptance of medical treatments and strengthen public trust in order to enhance societal resilience to future health challenges.

Prognostic Significance and Associations of Neural Network-Derived Electrocardiographic Features.

Subtle, prognostically important ECG features may not be apparent to physicians. In the course of supervised machine learning, thousands of ECG features are identified. These are not limited to conventional ECG parameters and morphology. We aimed to investigate whether neural network-derived ECG features could be used to predict future cardiovascular disease and mortality and have phenotypic and genotypic associations. We extracted 5120 neural network-derived ECG features from an artificial intelligence-enabled ECG model trained for 6 simple diagnoses and applied unsupervised machine learning to identify 3 phenogroups. Using the identified phenogroups, we externally validated our findings in 5 diverse cohorts from the United States, Brazil, and the United Kingdom. Data were collected between 2000 and 2023. In total, 1 808 584 patients were included in this study. In the derivation cohort, the 3 phenogroups had significantly different mortality profiles. After adjusting for known covariates, phenogroup B had a 20% increase in long-term mortality compared with phenogroup A (hazard ratio, 1.20 [95% CI, 1.17-1.23]; P<0.0001; phenogroup A mortality, 2.2%; phenogroup B mortality, 6.1%). In univariate analyses, we found phenogroup B had a significantly greater risk of mortality in all cohorts (log-rank P<0.01 in all 5 cohorts). Phenome-wide association study showed phenogroup B had a higher rate of future atrial fibrillation (odds ratio, 2.89; P<0.00001), ventricular tachycardia (odds ratio, 2.00; P<0.00001), ischemic heart disease (odds ratio, 1.44; P<0.00001), and cardiomyopathy (odds ratio, 2.04; P<0.00001). A single-trait genome-wide association study yielded 4 loci. SCN10A, SCN5A, and CAV1 have roles in cardiac conduction and arrhythmia. ARHGAP24 does not have a clear cardiac role and may be a novel target. Neural network-derived ECG features can be used to predict all-cause mortality and future cardiovascular diseases. We have identified biologically plausible and novel phenotypic and genotypic associations that describe mechanisms for the increased risk identified.

Remaining Useful Life Prediction based on Multisource Domain Transfer and Unsupervised Alignment

Transfer learning （TL） enhances remaining useful life (RUL) predictions by addressing data scarcity and operational challenges. Nonetheless, when a significant disparity in degradation data distribution exists between source and target domains, single-source domain TL may lead to misleading or negative transfer. Multisource domain TL partially mitigates these issues but fails to account for substantial discrepancies in feature-label correlations, impairing RUL prediction accuracy. To cope with this problem, we propose a multisource domain unsupervised adaptive learning method powered by a temporal convolutional network. Using a multilinear conditioning strategy, we combine degradation data and subregion labels to construct input characteristics for the domain discriminator. Additionally, we design a feature extractor that produces label-related features invariant across domains, thereby enhancing prediction precision. We evaluate our method using the publicly available C-MAPSS degradation dataset, demonstrating its effectiveness through a case study and ablation experiments.

EDMF: A New Benchmark for Multi-Focus Images with the Challenge of Exposure Difference

The goal of the multi-focus image fusion (MFIF) task is to merge images with different focus areas into a single clear image. In real world scenarios, in addition to varying focus attributes, there are also exposure differences between multi-source images, which is an important but often overlooked issue. To address this drawback and improve the development of the MFIF task, a new image fusion dataset is introduced called EDMF. Compared with the existing public MFIF datasets, it contains more images with exposure differences, which is more challenging and has a numerical advantage. Specifically, EDMF contains 1000 pairs of color images captured in real-world scenes, with some pairs exhibiting significant exposure difference. These images are captured using smartphones, encompassing diverse scenes and lighting conditions. Additionally, in this paper, a baseline method is also proposed, which is an improved version of memory unit-based unsupervised learning. By incorporating multiple adaptive memory units and spatial frequency information, the network is guided to focus on learning features from in-focus areas. This approach enables the network to effectively learn focus features during training, resulting in clear fused images that align with human visual perception. Experimental results demonstrate the effectiveness of the proposed method in handling exposure difference, achieving excellent fusion results in various complex scenes.

Pathomic and bioinformatics analysis of clinical-pathological and genomic factors for pancreatic cancer prognosis.

Pancreatic cancer exhibits a high degree of malignancy with a poor prognosis, lacking effective prognostic targets. Utilizing histopathological methodologies, this study endeavors to predict the expression of pathological features in pancreatic ductal adenocarcinoma (PAAD) and investigate their underlying molecular mechanisms. Pathological images, transcriptomic, and clinical data from TCGA-PAAD were collected for survival analysis. Image segmentation using unsupervised machine learning was employed to extract features, perform clustering, and establish models. The prognostic value of pathological features and associated clinical risk factors were evaluated; the correlation between pathological features and molecular mechanisms, gene mutations, and immune infiltration was analyzed. By clustering 45 effective pathological features, we divided PAAD patients into two groups: cluster 1 and cluster 2. Significant associations with poor prognosis were found for cluster 2 in both the training group (n = 113) and validation group (n = 75) (p = 0.006), with pathological stages II-IV identified as potential synergistic risk factors (HR = 2.421, 95% CI = 1.263-4.639, p = 0.008). Subsequently, through multi-omics correlation analysis, we further revealed a close association between cluster 2 and the oxidative phosphorylation mechanism. Within the cluster 2 group, 28 oxidative phosphorylation genes exhibited reduced expression, CDKN2A gene mutations were upregulated, and there was significant downregulation of Tregs infiltration and related immune gene expression. The pathomic model constructed using machine learning serves as a valuable prognostic target for PAAD. The histopathological features cluster 2 are closely associated with the downregulation of oxidative phosphorylation levels and Tregs immune infiltration.

Definitions and Characteristics of Patient Digital Twins Being Developed for Clinical Use: Scoping Review.

The concept of digital twins, widely adopted in industry, is entering health care. However, there is a lack of consensus on what constitutes the digital twin of a patient. The objective of this scoping review was to analyze definitions and characteristics of patient digital twins being developed for clinical use, as reported in the scientific literature. We searched PubMed, Scopus, Embase, IEEE, and Google Scholar for studies claiming digital twin development or evaluation until August 2023. Data on definitions, characteristics, and development phase were extracted. Unsupervised classification of claimed digital twins was performed. We identified 86 papers representing 80 unique claimed digital twins, with 98% (78/80) in preclinical phases. Among the 55 papers defining "digital twin," 76% (42/55) described a digital replica, 42% (23/55) mentioned real-time updates, 24% (13/55) emphasized patient specificity, and 15% (8/55) included 2-way communication. Among claimed digital twins, 60% (48/80) represented specific organs (primarily heart: 15/48, 31%; bones or joints: 10/48, 21%; lung: 6/48, 12%; and arteries: 5/48, 10%); 14% (11/80) embodied biological systems such as the immune system; and 26% (21/80) corresponded to other products (prediction models, etc). The patient data used to develop and run the claimed digital twins encompassed medical imaging examinations (35/80, 44% of publications), clinical notes (15/80, 19% of publications), laboratory test results (13/80, 16% of publications), wearable device data (12/80, 15% of publications), and other modalities (32/80, 40% of publications). Regarding data flow between patients and their virtual counterparts, 16% (13/80) claimed that digital twins involved no flow from patient to digital twin, 73% (58/80) used 1-way flow from patient to digital twin, and 11% (9/80) enabled 2-way data flow between patient and digital twin. Based on these characteristics, unsupervised classification revealed 3 clusters: simulation patient digital twins in 54% (43/80) of publications, monitoring patient digital twins in 28% (22/80) of publications, and research-oriented models unlinked to specific patients in 19% (15/80) of publications. Simulation patient digital twins used computational modeling for personalized predictions and therapy evaluations, mostly for one-time assessments, and monitoring digital twins harnessed aggregated patient data for continuous risk or outcome forecasting and care optimization. We propose defining a patient digital twin as "a viewable digital replica of a patient, organ, or biological system that contains multidimensional, patient-specific information and informs decisions" and to distinguish simulation and monitoring digital twins. These proposed definitions and subtypes offer a framework to guide research into realizing the potential of these personalized, integrative technologies to advance clinical care.

The diffusion of industrial robots in Europe: regional or country effect?

Abstract The paper investigates whether the penetration of advanced manufacturing technologies can be better explained at the regional or national level. If regional effects prevail, policy actions would focus on local investments, while if country effects make regional covariates redundant, they should be redirected to more structural reform of the national systems of innovation. In this respect, the contribution is 2-fold. First, data on acquisitions of industrial robots in the five largest European economies are rescaled at regional levels to draw a clear picture of winners and losers in the robotics race after the 2008 financial crisis. Second, we explain differential of growth rates in robot adoption with (1) traditional measures of industrial variety, (2) an unsupervised machine learning approach classifying a region’s industry profile (3) usual determinants of innovation and, thereafter test the robustness of the results when country effects are added. As the main result, we highlight a process of regional convergence in which country-fixed effects hold greater explanatory power, although related variety and the number of skilled people are statistically significant regional explanatory factors. We do not discover a specific industry mix associated with the rise of adoption, but we highlight the one associated with its decline.

Analyzing Daytime/Nighttime Pedestrian Crash Patterns in Michigan Using Unsupervised Machine Learning Techniques and their Potential as a Decision-Making Tool

Background Data mining applications are becoming increasingly common across various fields. Numerous data mining methodologies have been utilized in pedestrian crash data analysis. However, association and correspondence analyses have yet to be extensively employed in pedestrian safety literature to support decision-making. Material and Methods Road lighting significantly affects pedestrian crashes, highlighting the importance of examining pedestrian crash patterns during daytime and nighttime. This study analyzed pedestrian fatal and injury crashes in Michigan from 2011 to 2021 under varying road lighting conditions. The study identified pedestrian crash patterns using unsupervised machine-learning techniques based on several multidimensional factors. The Association Rules Learning (ARL) technique was used to determine general associations of patterns leading to pedestrian crashes. At the same time, Multiple Correspondence Analysis (MCA) helped understand crash attribute patterns in two notable scenarios: elderly pedestrian crashes during daytime lighting and high-speed midblock crashes during nighttime. These methods revealed key patterns associated with contributing factors to pedestrian crashes in different lighting conditions. Results In the analysis of cloud combinations involving elderly pedestrians during daylight, it was found that: 1) elderly pedestrians were significantly involved in severe injury crashes at two-lane midblock locations with a speed limit between 45 and 65 mph; 2) improper actions by elderly drivers were significantly associated with elderly pedestrians on wide roads (&gt;5 lanes); and 3) the complex interaction between city streets with 3-5 lanes was significantly correlated with drivers 'failed to yield' actions. Additionally, in the analysis of cloud combinations involving nighttime pedestrian crashes at high-speed midblock locations, it was found that 1) a specific age group of alcohol- or drug-involved pedestrians (19-30) was significantly associated with young drivers and improper driver actions; 2) during the winter season, young pedestrians were significantly involved in moderate injury crashes at undivided midblock locations; and 3) elderly drivers and 'failed to yield' actions were highly correlated with 3-5 lane roadways. Discussion The research findings are expected to raise awareness of Michigan pedestrian crash patterns under daytime and nighttime lighting conditions. They will also recommend safety countermeasures for practitioners to enhance pedestrian safety in walkable cities. The proposed methodologies can uncover relationships that need to be well-documented in the existing literature on pedestrian safety. Additionally, these methods can identify valuable relationships without restricting variables to being either dependent or independent, and they can reveal relationships that might be challenging to detect otherwise. The results will also provide decision rules and visualizations that are easy to understand. Conclusion The results showed that the proposed techniques can analyze pedestrian crash data in a way that aligns with understanding pedestrian crash characteristics. Traffic safety administrators could benefit from this methodology as a decision-support tool.

Target-aware cross-modality unsupervised domain adaptation for vestibular schwannoma and cochlea segmentation.

There is growing interest in research on segmentation for the vestibular schwannoma (VS) and cochlea using high-resolution T2 (hrT2) imaging over contrast-enhanced T1 (ceT1) imaging due to the contrast agent side effects. However, the hrT2 imaging remains a problem of insufficient annotated data, which is fatal for building more robust segmentation models. To address the issue, recent studies have adopted unsupervised domain adaptation approaches that translate ceT1 images to hrT2 images. However, previous studies did not consider the size and visual characteristics of the target objects, such as VS and cochlea, during image translation. Specifically, those works simply performed normalization on the entire image without considering its significant impact on the quality of the translated images. These approaches tend to erase the small target objects, making it difficult to preserve the structure of these objects when generating pseudo-target images. Furthermore, they may also struggle to accurately reflect the unique style of the target objects within the images. Therefore, we propose a target-aware unsupervised domain adaptation framework, designed for translating target objects, each tailored to their unique visual characteristics and size using target-aware normalization. We demonstrate the superiority of the proposed framework on a publicly available challenge dataset. Codes are available at https://github.com/Bokyeong-Kang/TANQ .

Pre-Impact Fall Detection for E-Scooter Riding Using an IMU: Threshold-Based, Supervised, and Unsupervised Approaches

Pre-impact fall detection during e-scooter riding is essential for rider safety. Both threshold-based and deep learning algorithms (supervised and unsupervised models) were developed in this study. Twenty participants performed normal driving maneuvers such as straight driving, speed bumps, clockwise roundabouts, and counterclockwise roundabouts, along with falls (abnormal driving maneuvers). A 6-axis IMU sensor (Xsens DOT, The Netherlands) was positioned at the T7 location to record data at 60 Hz. The approaches included threshold-based, supervised learning, and unsupervised learning models The threshold-based approach yielded an accuracy of 98.86% with an F1 score of 0.99, while the supervised model had a slightly lower performance, reaching 86.29% accuracy and an F1 score of 0.56. The unsupervised knowledge distillation model achieved 98.86% accuracy, an F1 score of 0.99, and a memory size of only 46 kB. All models demonstrated lead times of more than 250 ms, sufficient for airbag deployment.

Inter-chromosomal contacts demarcate genome topology along a spatial gradient.

Non-homologous chromosomal contacts (NHCCs) between different chromosomes participate considerably in gene and genome regulation. Due to analytical challenges, NHCCs are currently considered as singular, stochastic events, and their extent and fundamental principles across cell types remain controversial. We develop a supervised and unsupervised learning algorithm, termed Signature, to call NHCCs in Hi-C datasets to advance our understanding of genome topology. Signature reveals 40,282 NHCCs and their properties across 62 Hi-C datasets of 53 diploid human cell types. Genomic regions of NHCCs are gene-dense, highly expressed, and harbor genes for cell-specific and sex-specific functions. Extensive inter-telomeric and inter-centromeric clustering occurs across cell types [Rabl's configuration] and 61 NHCCs are consistently found at the nuclear speckles. These constitutive 'anchor loci' facilitate an axis of genome activity whilst cell-type-specific NHCCs act in discrete hubs. Our results suggest that non-random chromosome positioning is supported by constitutive NHCCs that shape genome topology along an off-centered spatial gradient of genome activity.

Alternative Learning Paradigms for Image Quality Transfer

Image Quality Transfer (IQT) aims to enhance the contrast and resolution of low-quality medical images, e.g. obtained from low-power devices, with rich information learned from higher quality images. In contrast to existing IQT methods in the literature which adopt supervised learning frameworks, in this work, we propose two novel formulations of the IQT problem. The first approach uses an unsupervised learning framework, whereas the second is a combination of both supervised and unsupervised learning. The unsupervised learning approach considers a sparse representation (SRep) and dictionary learning model, which we call IQT-SRep, whereas the combination of supervised and unsupervised learning ap- proach is based on deep dictionary learning (DDL), which we call IQT-DDL. The IQT-SRep approach trains two dictionaries using a sparse representation model using pairs of low- and high-quality volumes. Subsequently, the sparse representation of a low-quality block, in terms of the low-quality dictionary, can be directly used to recover the corresponding high-quality block using the high-quality dictionary. On the other hand, the IQT-DDL ap- proach explicitly learns a high-resolution dictionary to upscale the input volume, while the entire network, including high dictionary generator, is simultaneously optimised to take full advantage of deep learning methods. The two models are evaluated using a low-field mag- netic resonance imaging (MRI) application aiming to recover high-quality images akin to those obtained from high-field scanners. Experiments comparing the proposed approaches against state-of-the-art supervised deep learning IQT method (IQT-DL) identify that the two novel formulations of the IQT problem can avoid bias associated with supervised meth- ods when tested using out-of-distribution data that differs from the distribution of the data the model was trained on. This highlights the potential benefit of these novel paradigms for IQT.

Abstract 4121812: Longitudinal Trajectory-Based Classification of Heart Failure with preserved Ejection Fraction: A Machine Learning Approach

Background: Heart failure (HF) with preserved ejection fraction (HFpEF) is a complex syndrome characterized by heterogeneous pathophysiology and dynamic changes in cardiac structure and function. Current research predominantly focuses on left ventricular ejection fraction (LVEF) or left ventricular structures with predefined categories. This study hypothesized that applying unsupervised methodologies to longitudinal echocardiography will reveal distinct trajectory patterns in cardiac structure and function related to clinical outcomes in HFpEF patients. Methods: This retrospective cohort study, conducted at UC Davis Medical Center (2014 to 2022), consisted of adult patients with HFpEF. The inclusion criteria included ICD HF codes, a left ventricular ejection fraction (LVEF) ≥50%, and a Brain natriuretic peptide (BNP) ≥100 pg/ml or N-terminal-pro BNP (NT-proBNP) ≥225 pg/ml. Longitudinal data were analyzed using unsupervised machine learning, including UMAP for dimensionality reduction and Gaussian Mixture Models for clustering. Primary and secondary outcomes included all-cause mortality and changes in BNP levels, respectively. Results: The study included 1,626 patients. The cohort had a median age of 69 years, was 57.5% female, and 51.8% white. Three distinct echocardiographic trajectory patterns emerged (p&lt;.05) among the HFpEF cohort. Trajectory #1 (n=568) exhibited a high-stable pattern with the highest mortality rates, Trajectory #2 (n=512) showed a high-decreasing pattern with high mortality rates, and Trajectory #3 (n=546) had a low-increasing pattern with the lowest mortality rates. Trajectories varied significantly in baseline characteristics and longitudinal echocardiographic changes, with Trajectory #3 having a lower baseline left ventricular mass index (p&lt;.001) and a higher mean arterial pressure ( p =.001) compared to the other trajectories. Kaplan-Meier survival analysis indicated significant differences in survival among the trajectories ( p =.004). Conclusion: The study utilized unsupervised analysis of longitudinal echocardiographic data to identify three distinct HFpEF trajectories, each associated with unique clinical features and outcomes. These findings underscore the heterogeneity of HFpEF and highlight the potential trajectory-based patient stratification to improve targeted intervention strategies. Future validation in larger, multicenter cohorts is warranted to enhance HFpEF management and patient prognosis.

An adaptive decision support system for outpatient appointment scheduling with heterogeneous service times.

Appointment scheduling (AS) plays a crucial role in outpatient clinic management. Traditional methods involve patient grouping using pre-defined rules and scheduling based on these groups. However, pre-defined rules may not adequately capture the heterogeneity in patients' service times (i.e., consultation duration). Advanced machine learning (ML) methods can address individual-level heterogeneity but pose challenges for practical scheduling. To strike a balance, we propose a data-driven AS decision support system, Cluster-Predict-Schedule (CPS), integrating both supervised and unsupervised ML for efficient patient grouping and scheduling. The novelty of CPS lies in its adaptability to service time heterogeneity through a data-driven approach, determining patient groups based on data rather than pre-defined rules. Additionally, CPS includes a generic and efficient algorithm for generating appointment templates adaptable to any number of patient groups. Our system's efficacy is demonstrated using a real-world dataset. Evaluated by the weighted sum of patient wait times, physician idle time, and overtime, CPS achieves up to 15.0% cost reduction compared to the FCFA (first-call, first-appointment) scheme and over 4.7% savings against the common New/Return classification with traditional sequencing candidate (TSC) rules. In addition, CPS enhances outpatient operational efficiency without compromising fairness.

Abstract 4141460: Exploring Novel Data-Driven Clustering Methods for Uncovering Patterns in Longitudinal Neonatal Postoperative Temperature Measurements

Introduction: Neonatal hypothermia (&lt; 36.5 o C) after CPB is considered benign despite lack of evidence on its prognostic significance. Question: Are group-based trajectory modeling (GBTM), k-means, and self-organizing map (SOM) clustering approaches clinically useful for evaluating their associations with important outcomes? Aims: Identify distinct postoperative temperature trajectories in neonates after CPB using novel machine learning (ML) clustering methods, corroborate findings, and evaluate prognostic value on outcomes. Methods: Secondary cohort analysis of prospectively collected data. Accessed data from a single pediatric referral center's CardioAccess registry consistent of postoperative neonates with congenital heart defects (CHD). 48-hourly postoperative temperature measurements were extracted from patient records. GBTM, SOM, and k-means clustering identified cluster membership and model fit was optimized for 3 clusters based on existing knowledge. The primary outcome was a complication composite of severe bleeding, illness severity, cardiac arrest, arrhythmia, delayed first successful extubation, prolonged ICU LOS, very poor weight gain, and 30-day mortality. 3 data driven techniques were compared and then associated with outcomes using adjusted multivariable binary logistic regression (LR). Results: 450 neonates ≥ 34 weeks gestation undergoing CPB between 2015 and 2019 met inclusion criteria. GBTM, SOM, and k-means clustering identified temperature group assignment for 3 clusters: 1) persistent hypothermia (n=38, 9% vs. n=49, 11% vs. n=40, 9%), 2) resolving hypothermia (n=233, 51% vs. n=227, 50% vs. n=147, 33%), and 3) normothermia (n=179, 40% vs. n=174, 39% vs. n=263, 58%). Concordance of techniques showed strong agreement between GBTM and SOM (kappa= 0.92). The kappa test comparing GBTM and k-means assignment was 0.41, indicating a weak to moderate agreement. After adjustment in the multivariable binary LR, persistently hypothermic neonates compared to normothermic neonates were associated with higher odds of the complication composite outcome in the GBTM (OR: 2.8 [1.0-7.3], p 0.041) and SOM (OR: 2.3 [1.0-5.4], p 0.045) models, but not the k-means model (OR: 1.4 [0.7-3.1], p 0.38). Conclusion: Exploring concordance between different unsupervised ML techniques, shows that neonatal temperature after CPB follows distinct postoperative trajectories, and those exhibiting a persistent hypothermia trend are at higher risk of adverse outcomes.

The Use of Artificial Intelligence to Analyze the Exposome in the Development of Chronic Diseases: A Review of the Current Literature

The “Exposome” is a concept that indicates the set of exposures to which a human is subjected during their lifetime. These factors influence the health state of individuals and can drive the development of Noncommunicable Diseases (NCDs). Artificial Intelligence (AI) allows one to analyze large amounts of data in a short time. As such, several authors have used AI to study the relationship between exposome and chronic diseases. Under such premises, this study reviews the use of AI in analyzing the exposome to understand its role in the development of chronic diseases, focusing on how AI can identify patterns in exposure-related data and support prevention strategies. To achieve this, we carried out a search on multiple databases, including PubMed, ScienceDirect, and SCOPUS, from 1 January 2019 to 31 May 2023, using the MeSH terms (exposome) and (‘Artificial Intelligence’ OR ‘Machine Learning’ OR ‘Deep Learning’) to identify relevant studies on this topic. After completing the identification, screening, and eligibility assessment, a total of 18 studies were included in this literature review. According to the search, most authors used supervised or unsupervised machine learning models to study multiple exposure factors’ role in the risk of developing cardiovascular, metabolic, and chronic respiratory diseases. In some more recent studies, authors also used deep learning. Furthermore, the exposome analysis is useful to study the risk of developing neuropsychiatric disorders or evaluating pregnancy outcomes and child growth. Understanding the role of the exposome is pivotal to overcome the classic concept of a single exposure/disease. The application of AI allows one to analyze multiple environmental risks and their combined effects on health conditions. In the future, AI could be helpful in the prevention of chronic diseases, providing new diagnostic, therapeutic, and follow-up strategies.