Embedded Files
Predicting Daily Cardiovascular Emergencies Using Weather and Air Quality Data: A 23-Year Machine-Learning Analysis in Taiwan (2026)
Predicting Daily Cardiovascular Emergencies Using Weather and Air Quality Data: A 23-Year Machine-Learning Analysis in Taiwan (2026)
Short-term variability in weather and air quality is known to influence cardiovascular emergencies, yet its day-to-day predictive value at the population level remains insufficiently understood. Using 23 years of nationwide data from Taiwan (2000–2022), we evaluated how weather and air quality conditions shape daily cardiovascular disease (CVD) emergency visits across geographic regions. We first applied unsupervised learning methods, including UMAP and K-means clustering, to 184 environmental features to identify data-driven environmental regimes and examine the distribution of high-risk CVD days. We then trained eight supervised learning models to predict daily CVD emergency visits and used SHAP values to interpret key predictors. Unsupervised analyses revealed consistent seasonal and pollution-related patterns. High-risk days tended to cluster during cool conditions accompanied by elevated air pollution, with temperatures higher than in winter but substantially greater pollution levels. This pattern was particularly evident in northern Taiwan and among populations aged 65 years and older. Air-pollution variables produced more clearly defined high-risk clusters than meteorological variables alone, indicating a stronger pollution-related contribution to acute CVD risk. In the supervised framework, tree-based ensemble models (Random Forest, LightGBM, XGBoost) achieved the best performance, with R2 values up to 0.67 and mean absolute percentage errors of 7%–8%. Predictability was highest for elderly populations and northern Taiwan. SHAP analyses identified NOx-related metrics as dominant predictors of CVD emergency visits. These results demonstrate that high-resolution environmental data, combined with machine-learning methods, can effectively predict CVD emergencies, delineate high-risk environmental regimes, and support region-specific early-warning systems centered on air-pollution monitoring, particularly NOx.
Ensemble-MIL: deep learning-based ensemble framework for biomarker prediction from histopathological images in colorectal cancer (2026)
Ensemble-MIL: deep learning-based ensemble framework for biomarker prediction from histopathological images in colorectal cancer (2026)
Recent studies have explored histopathological whole slide images (WSIs) for predicting colorectal cancer (CRC) biomarkers, aiming to create cost-effective and efficient diagnostic tools. However, achieving strong predictive performance and generalizability across datasets remains a challenge. Here, we introduce a deep learning-based ensemble framework, Ensemble-MIL, designed to robustly predict key CRC biomarkers, including BRAF V600E, KRAS mutations, and MSI-H status, with improved cross-dataset performance. We employed two independent CRC datasets: TCGA-COAD for model training and internal evaluation, and CPTAC-COAD as an external test set to assess generalizability. All WSIs were preprocessed and divided into small image patches. A tumor detection model was applied to identify tumor regions, and patch-level; features were extracted via SimCLR, a contrastive learning method. These features were utilized to train three multiple instance learning (MIL) models: Att-MIL, Tran-MIL, and GNN-MIL. The models were then integrated into the final Ensemble-MIL framework. In internal testing with TCGA-COAD, the proposed method achieved area under the curve (AUC) scores of 0.90, 0.87, and 0.64 for MSI-H, BRAF, and KRAS, respectively. In external testing on CPTAC-COAD, it achieved AUCs of 0.78, 0.76, and 0.61 for MSI-H, BRAF, and KRAS, respectively, outperforming previous results. This framework offers a scalable and effective solution for image-based biomarker screening and demonstrates strong potential for clinical application, particularly in resource-limited settings. The code is available at https://github.com/chenh2lab/Ensemble-MIL.
Hybrid Attention-Enhanced Regularized Convolutional Autoencoder for Robust EEG Feature Extraction in Seizure Prediction (2025)
Hybrid Attention-Enhanced Regularized Convolutional Autoencoder for Robust EEG Feature Extraction in Seizure Prediction (2025)
Electroencephalography (EEG)-based seizure prediction has garnered significant attention in epilepsy management, with deep learning methods enhancing prediction performance. However, persistent challenges include insufficient robust low-dimensional representations and poor generalization across datasets. Consequently, this study proposes a novel hybrid attention-enhanced regularized convolutional autoencoder (HA-RCAE) feature extraction method that generates low-dimensional features, effectively preserving critical biological information while mitigating technical batch effects in EEG signals. The HA-RCAE is augmented with regularization techniques and a novel hybrid attention mechanism for enhanced feature extraction. Coupled with classic machine learning classifiers, the proposed method outperforms advanced techniques on two epilepsy EEG datasets. On the Kaggle dataset, the model achieves area under the curve (AUC), sensitivity (SS), specificity (SP), and false positive rate (FPR) values of 90.6%, 89.1%, 89.8%, and 0.186/h, respectively; on the CHB-MIT dataset, it achieves AUC, SS, SP, and FPR values of 92.7%, 85.4%, 91.7%, and 0.144/h, respectively. The proposed method exhibits excellent generalizability, characterized by minimal performance discrepancies across datasets. Furthermore, its applicability extends to EEG anomaly detection, demonstrating superior performance and generalizability across clinical tasks. This comprehensive approach offers a robust solution for seizure prediction and other clinical applications, distinguished by strong generalization and negligible batch effects, rendering it promising for clinical implementation.
Circular RNA detection identifies circPSEN1 alterations in brain specific to autosomal dominant Alzheimer's disease (2022)
Circular RNA detection identifies circPSEN1 alterations in brain specific to autosomal dominant Alzheimer's disease (2022)
Background
Autosomal-dominant Alzheimer's disease (ADAD) is caused by pathogenic mutations in APP, PSEN1, and PSEN2, which usually lead to an early age at onset (< 65). Circular RNAs are a family of non-coding RNAs highly expressed in the nervous system and especially in synapses. We aimed to investigate differences in brain gene expression of linear and circular transcripts from the three ADAD genes in controls, sporadic AD, and ADAD.
Methods
We obtained and sequenced RNA from brain cortex using standard protocols. Linear counts were obtained using the TOPMed pipeline; circular counts, using python package DCC. After stringent quality control (QC), we obtained the counts for PSEN1, PSEN2 and APP genes. Only circPSEN1 passed QC. We used DESeq2 to compare the counts across groups, correcting for biological and technical variables. Finally, we performed in-silico functional analyses using the Circular RNA interactome website and DIANA mirPath software.
Results
Our results show significant differences in gene counts of circPSEN1 in ADAD individuals, when compared to sporadic AD and controls (ADAD = 21, AD = 253, Controls = 23—ADADvsCO: log2FC = 0.794, p = 1.63 × 10–04, ADADvsAD: log2FC = 0.602, p = 8.22 × 10–04). The high gene counts are contributed by two circPSEN1 species (hsa_circ_0008521 and hsa_circ_0003848). No significant differences were observed in linear PSEN1 gene expression between cases and controls, indicating that this finding is specific to the circular forms. In addition, the high circPSEN1 levels do not seem to be specific to PSEN1 mutation carriers; the counts are also elevated in APP and PSEN2 mutation carriers. In-silico functional analyses suggest that circPSEN1 is involved in several pathways such as axon guidance (p = 3.39 × 10–07), hippo signaling pathway (p = 7.38 × 10–07), lysine degradation (p = 2.48 × 10–05) or Wnt signaling pathway (p = 5.58 × 10–04) among other KEGG pathways. Additionally, circPSEN1 counts were able to discriminate ADAD from sporadic AD and controls with an AUC above 0.70.
Conclusions
Our findings show the differential expression of circPSEN1 is increased in ADAD. Given the biological function previously ascribed to circular RNAs and the results of our in-silico analyses, we hypothesize that this finding might be related to neuroinflammatory events that lead or that are caused by the accumulation of amyloid-beta.
[Paper in Acta neuropathologica communications]
Online Prediction of Lead Seizures from iEEG Data (2021)
Online Prediction of Lead Seizures from iEEG Data (2021)
We describe a novel system for online prediction of lead seizures from long-term intracranial electroencephalogram (iEEG) recordings for canines with naturally occurring epilepsy. This study adopts new specification of lead seizures, reflecting strong clustering of seizures in observed data. This clustering results in fewer lead seizures (~7 lead seizures per dog), and hence new challenges for online seizure prediction, that are addressed in the proposed system. In particular, the machine learning part of the system is implemented using the group learning method suitable for modeling sparse and noisy seizure data. In addition, several modifications for the proposed system are introduced to cope with the non-stationarity of a noisy iEEG signal. They include: (1) periodic retraining of the SVM classifier using most recent training data; (2) removing samples with noisy labels from training data; and (3) introducing a new adaptive post-processing technique for combining many predictions made for 20 s windows into a single prediction for a 4 h segment. Application of the proposed system requires only two lead seizures for training the initial model, and results in high prediction performance for all four dogs (with mean 0.84 sensitivity, 0.27 time-in-warning, and 0.78 false-positive rate per day). The proposed system achieves accurate prediction of lead seizures during long-term test periods, 3–16 lead seizures during a 169–364 day test period, whereas earlier studies did not differentiate between lead vs. non-lead seizures and used much shorter test periods (~few days long).
[Paper in Brain Science]
Performance metrics for online seizure prediction (2020)
Performance metrics for online seizure prediction (2020)
Many recent studies on online seizure prediction from iEEG signal describe various prediction algorithms and their prediction performance. In contrast, this paper focuses on proper specification of system parameters, such as prediction period, prediction horizon and data-driven characterization of lead seizures. Whereas prediction performance clearly depends on these system parameters many researchers simply set the values of these parameters in an ad hoc manner. Our paper investigates the effect of these system parameters on online prediction performance, using both synthetic and real-life data sets. Therefore, meaningful comparison of methods/algorithms (for online seizure prediction) should consider proper specification of system parameters.
[Paper in Neural Networks]
Group Learning for High-Dimensional Sparse Data (2019)
Group Learning for High-Dimensional Sparse Data (2019)
We describe new methodology for supervised learning with sparse data, i.e., when the number of input features is (much) larger than the number of training samples (n). Under the proposed approach, all available (d) input features are split into several (t) subsets, effectively resulting in a larger number (t*n) of labeled training samples in lower-dimensional input space (of dimensionality d/t). This (modified) training data is then used to estimate a classifier for making predictions in lower-dimensional space. In this paper, standard SVM is used for training a classifier. During testing (prediction), a group of t predictions made by SVM classifier needs to be combined via intelligent post-processing rules, in order to make a prediction for a test input (in the original d-dimensional space). The novelty of our approach is in the design and empirical validation of these post-processing rules under Group Learning setting. We demonstrate that such post-processing rules effectively reflect general (common-sense) a priori knowledge (about application data). Specifically, we propose two different post-processing schemes and demonstrate their effectiveness for two real-life application domains, i.e., handwritten digit recognition and seizure prediction from iEEG signal. These empirical results show superior performance of the Group Learning approach for sparse data, under both balanced and unbalanced classification settings.
[Paper in 2019 IJCNN]
Page updated
Google Sites
Report abuse