Neural microstates underlying categorical speech perception using Bayesian non-parametrics

In simulation-based surgical training, a key task is to rate the performance of the operator, which is done currently by senior surgeons. This is a costly practice and objectively quantifiable assessment metrics are often missing. Researchers have been working towards building automated systems to achieve computational understanding of surgical skills, largely through analysis of motion data captured by video or other sensors. In this paper, we extend the Hierarchical Dirichlet Process Hidden Markov Model (HDP-HMM) for this purpose. We start with detecting spatial temporal interest points from the video capturing the tool motion of an operator, and then generate visual from the descriptors of those interest points. For each frame, we construct a histogram with the associated interest points, i.e. the bag of words and then every video is represented by a sequence of those histograms. For sequences of each motion expertise level, we infer an HDP-HMM model. Finally, the classification of the motion expertise level for a testing sequence is based on choosing a model that maximizes the likelihood of the given sequence. Compared with the other action recognition algorithms, such as kernel SVM, our method leads to a better result. Further, the proposed approach also provides some important cues on the patterns of motion for each expertise level.

Nonparametric Bayesian models use a Bayesian framework to learn the model complexity automatically from the data and eliminate the need for a complex model selection process. The Hierarchical Dirichlet Process hidden Markov model (HDP-HMM) is the nonparametric Bayesian equivalent of an HMM. However, HDP-HMM is restricted to an ergodic topology and uses a Dirichlet Process Model (DPM) to achieve a mixture distribution-like model. For applications such as speech recognition, where we deal with ordered sequences, it is desirable to impose a left-to-right structure on the model to improve its ability to model the sequential nature of the speech signal. In this paper, we introduce three enhancements to HDP-HMM: (1) a left-to-right structure: needed for sequential decoding of speech, (2) non-emitting initial and final states: required for modeling finite length sequences, (3) HDP mixture emissions: allows sharing of data across states. The latter is particularly important for speech recognition because Gaussian mixture models have been very effective at modeling speaker variability. Further, due to the nature of language, some models occur infrequently and have a small number of data points associated with them, even for large corpora. Sharing allows these models to be estimated more accurately. We demonstrate that this new HDP-HMM model produces a 15% increase in likelihoods and a 15% relative reduction in error rate on a phoneme classification task based on the TIMIT Corpus.

Future advanced driver assistance systems (ADASs) should observe a driving behavior and detect contextual changing points of driving behaviors. In this paper, we propose a novel method for predicting the next contextual changing point of driving behavior on the basis of a Bayesian double articulation analyzer. To develop the method, we extended a previously proposed semiotic predictor using an unsupervised double articulation analyzer that can extract a two-layered hierarchical structure from driving-behavior data. We employ the hierarchical Dirichlet process hidden semi-Markov model [4] to model duration time of a segment of driving behavior explicitly instead of the sticky hierarchical Dirichlet process hidden Markov model (HDP-HMM) employed in the previous model [13]. Then, to recover the hierarchical structure of contextual driving behavior as a sequence of chunks, we use the Nested Pitman-Yor Language model [6], which can extract latent words from sequences of latent letters. On the basis of the extension, we develop a method for calculating posterior probability distribution of the next contextual changing point by marginalizing potentially possible results of the chunking method and potentially successive words theoretically. To evaluate the proposed method, we applied the method to synthetic data and driving behavior data that was recorded in a real environment. The results showed that the proposed method can predict the next contextual changing point more accurately and in a longer-term manner than the compared methods: linear regression and Recurrent Neural Networks, which were trained through a supervised learning scheme.

Sepsis-Induced coagulopathy (SIC) is not only a common complication in the development process of sepsis but also related to poor prognosis of sepsis. We aimed to establish a machine learning (ML) model to predict the 28-day mortality risk of patients with SIC. We collected data for model training from the Medical Information Mart for Intensive Care IV Database version 2.2 to establish the model. We extracted patient data from the First Affiliated Hospital of Wenzhou Medical University for the model's external validation. We used Least Absolute Shrinkage and Selection Operator (LASSO) regression and logistic regression analysis to identify predictive factors for a 28-day mortality risk. Then, we built prognostic prediction models for SIC patients using multiple ML classification models. We evaluated predictive performance using Receiver Operating Characteristic (ROC) curves, calibration curves, and Decision Curve Analysis (DCA). We used Shapley Additive Explanations (SHAP) to interpret the models. We selected seventeen variables for model development, and the XGBoost model performed the best. The area under the curve (AUC) (95% CI) of the test set reached 0.840 (0.810-0.870), with an accuracy of 0.807, sensitivity of 0.836, and specificity of 0.798. The model also demonstrated excellent predictive performance in external validation, with an AUC (95% CI) of 0.864 (0.794-0.934). We constructed an XGBoost model and provided model interpretability using the SHAP. This model provides a basis for assessing the 28-day mortality risk of patients with SIC, aiding in clinical decision support and the formulation of personalized treatment strategies.

The evolution of communities in dynamic (time-varying) network data is a prominent topic of interest. A popular approach to understanding these dynamic networks is to embed the dyadic relations into a latent metric space. While methods for clustering with this approach exist for dynamic networks, they all assume a static community structure. This paper presents a Bayesian nonparametric model for dynamic networks that can model networks with evolving community structures. Our model extends existing latent space approaches by explicitly modeling the additions, deletions, splits, and mergers of groups with a hierarchical Dirichlet process hidden Markov model. Our proposed approach, the hierarchical Dirichlet process latent position cluster model (HDP-LPCM), incorporates transitivity, models both individual and group level aspects of the data, and avoids the computationally expensive selection of the number of groups required by most popular methods. We provide a Markov chain Monte Carlo estimation algorithm and demonstrate its ability to detect evolving community structure in a network of military alliances during the Cold War and a narrative network constructed from the Game of Thrones television series.

Understanding of human dynamics has drawn attention to various areas. The wide spread of positioning technologies, such as GPS facilitates location information to be obtained with high spatio-temporal resolution as well as at low costs. By collecting individual location information in real time, monitoring of human dynamics has recently become possible and is expected to the area of dynamic traffic control. In this monitoring, detecting anomalous states of human dynamics become important. This research aims to define an anomaly detection problem of the human dynamics monitoring with time-series gridded population data and develop an anomaly detection method for this problem. According to the result of a review we have conducted, we discussed the characteristics of the anomaly detection in human dynamics monitoring and categorized our problem to a semi-supervised anomaly detection problem that detects contextual anomalies behind time-series data. We developed an anomaly detection method based on a sticky hierarchical Dirichlet process hidden Markov model, which is able to estimate the number of latent states according to the input data. Results of the experiment with synthetic data showed that our proposed method has good fundamental performance with respect to the detection rate. Through the experiments with real gridded population data, anomalies were detected when and where an actual social event had occurred.

Developing an automated vehicle, that can handle complicated driving scenarios and appropriately interact with other road users, requires the ability to semantically learn and understand driving environment, oftentimes, based on analyzing massive amounts of naturalistic driving data. An important paradigm that allows automated vehicles to both learn from human drivers and gain insights is understanding the principal compositions of the entire traffic, termed as traffic primitives. However, the exploding data growth presents a great challenge in extracting primitives from high-dimensional time-series traffic data with various types of road users engaged. Therefore, automatically extracting primitives is becoming one of the cost-efficient ways to help autonomous vehicles understand and predict the complex traffic scenarios. In addition, the extracted primitives from raw data should 1) be appropriate for automated driving applications and also 2) be easily used to generate new traffic scenarios. However, existing literature does not provide a method to automatically learn these primitives from large-scale traffic data. The contribution of this paper has two manifolds. The first one is that we proposed a new framework to generate new traffic scenarios from a handful of limited traffic data. The second one is that we introduce a nonparametric Bayesian learning method -- a sticky hierarchical Dirichlet process hidden Markov model -- to automatically extract primitives from multidimensional traffic data without prior knowledge of the primitive settings. The developed method is then validated using one day of naturalistic driving data. Experiment results show that the nonparametric Bayesian learning method is able to extract primitives from traffic scenarios where both the binary and continuous events coexist.

Motivation: When analysing gene expression time series data, an often overlooked but crucial aspect of the model is that the regulatory network structure may change over time. Although some approaches have addressed this problem previously in the literature, many are not well suited to the sequential nature of the data.Results: Here, we present a method that allows us to infer regulatory network structures that may vary between time points, using a set of hidden states that describe the network structure at a given time point. To model the distribution of the hidden states, we have applied the Hierarchical Dirichlet Process Hidden Markov Model, a non-parametric extension of the traditional Hidden Markov Model, which does not require us to fix the number of hidden states in advance. We apply our method to existing microarray expression data as well as demonstrating is efficacy on simulated test data.Contact: thomas.thorne@imperial.ac.uk

This paper presents a comprehensive machine learning (ML) framework for predicting liquefaction risk, a crucial aspect of seismic hazard assessment. A benchmark geotechnical dataset with multi-dimensional input features was used to evaluate several ML classifiers, followed by hyperparameter optimization through stratified 5-fold cross-validation. Optimized models were combined into a soft Voting Ensemble to enhance stability and accuracy of liquefaction potential prediction. The proposed ensemble model achieved a mean accuracy of 90.12% and a recall of 97.23%, outperforming individual models in most folds. The ensemble’s effectiveness was further evidenced by its precision-recall (PR) and receiver operating characteristic (ROC) curves, with areas under the curve (AUC) of 0.962 and 0.931, respectively—closely matching those of the Gradient Boosting classifier, indicating comparable discriminatory performance. Additionally, SHapley Additive exPlanations (SHAP) analysis was conducted on the ensemble model to assess contributions of each geotechnical inputs to the predictions, revealing that normalized shear wave velocity (VS1) as the most influential variable in liquefaction prediction. The proposed framework demonstrates a robust, interpretable, and performance-consistent approach for liquefaction risk assessment.

BackgroundDynamic 18F-fluorodeoxyglucose (18F-FDG) PET/CT imaging has been shown to provide additional information for diagnosing lung cancer. The aim of this study was to investigate whether metabolic and flow features directly extracted from time activity curves (TACs) help differentiate between benign and malignant conditions of lung lesions.MethodsTACs at the primary lesion were extracted from each dynamic 18F-FDG PET/CT scan. The TAC signal was then decomposed into metabolism and blood flow components through kinetic modeling. Dynamic features including area under the curve (AUC), time-to-peak, and slopes were then extracted from each component. The extracted features from 187 patients (mean age, 60.41 ± 11.01 years; 117 males) were used to train a classification model based on bagging, a machine-learning method built with decision trees. The performance of the trained model on differentiating benign and malignant was tested using receiver operating characteristic analysis with cross-validation. External testing was then performed for an independent dataset that consisted of 42 dynamic scans. For the results, SHapley Additive exPlanations (SHAP) were used to assess the relative importance of the contributed features for individuals. Waterfall charts were also plotted, together with assessment of Cohen’s effect size to demonstrate the superiority of the proposed model over SUVmax and the net FDG influx rate Ki.ResultsThe combination of the multiple dynamic features was able to separate benign and malignant lesions. For cross-validation, the trained model had an AUC of 0.89, sensitivity of 0.80, and specificity of 0.88, which was significantly higher than that of either SUVmax (AUC = 0.79, DeLong p < 0.001) or Ki (AUC = 0.76, DeLong test p < 0.001). For the testing dataset, the model had an AUC of 0.86, which again was better than either SUVmax (AUC of 0.72) or Ki (AUC of 0.71). The most important features that contributed to the diagnosis identified by SHAP included the slope and maximum metabolism TAC at the lesion, the AUC, and the peak time of blood TAC at the lesion. The waterfall chart illustrated that the model had significantly different prediction scores between the benign and malignant groups (p < 0.001) with a Cohen’s effect size of 1.71, which was higher than that of the values for SUV and Ki (Cohen’s effect size 0.96 and 0.81, respectively).ConclusionAn explainable machine learning model that combines dynamic FDG metabolic and flow features can predict benign or malignant lung cancer patients more accurately than conventional parameters such as SUVmax or net influx rate Ki.Supplementary InformationThe online version contains supplementary material available at 10.1007/s00259-025-07231-0.

Chronic Hepatitis B (CHB) is a leading cause of liver fibrosis. Accurate and non-invasive diagnosis of liver fibrosis in CHB patients is of critical clinical importance. This study aimed to develop and validate machine learning (ML)-based models for predicting significant liver fibrosis in CHB patients. This retrospective cohort study included 328 CHB patients (225 with non-significant liver fibrosis and 103 with significant liver fibrosis) from 2017 to 2022. Four ML models were constructed based on four selected features identified through the least absolute shrinkage and selection operator (LASSO) regression. Model performance was assessed using the receiver operating characteristic (ROC) curve, and the area under the curve (AUC), accuracy, sensitivity, specificity, and SHapley Additive exPlanations (SHAP) analysis. The random forest (RF) model demonstrated the highest predictive performance, with an AUC of 0.874 (95% CI: 0.813-0.934) in the training set and 0.863 (95% CI: 0.772-0.955) in the test set, outperforming extreme gradient boosting (XGBoost), logistic regression (LR), and support vector machine (SVM). Compared with the traditional fibrosis indices such as aspartate aminotransferase to platelet ratio index (APRI) (AUC = 0.585) and fibrosis-4 (FIB-4) (AUC = 0.633), the RF model (AUC = 0.863) demonstrated significantly higher predictive accuracy. SHAP analysis identified platelet count (PLT) as the most influential predictor in the RF model. The ML-based RF model offers a highly accurate, non-invasive interpretable tool for predicting significant liver fibrosis in patients with CHB, offering potential for clinical application in routine fibrosis risk assessment.

Background To assess the feasibility of a machine learning (ML) approach using radiomics features of perfusion defects on rest myocardial perfusion imaging (MPI) to detect the presence of hibernating myocardium. Methodology Data of patients who underwent 99mTc-sestamibi MPI and 18F-FDG PET/CT for myocardial viability assessment were retrieved. Rest MPI data were processed on ECToolbox, and polar maps were saved using theNFile PMap tool. The reference standard for defining hibernating myocardium was the presence of mismatched perfusion-metabolism defect with impaired myocardial contractility at rest. Perfusion defects on the polar maps were delineated with regions of interest (ROIs) after spatial resampling and intensity discretization. Replicable random sampling allocated 80% (257) of the perfusion defects of the patients from January 2017 to September 2022 to the training set and the remaining 20% (64) to the validation set. An independent dataset of perfusion defects from 29 consecutive patients from October 2022 to January 2023 was used as the testing set for model evaluation. One hundred ten first and second-order texture features were extracted for each ROI. After feature normalization and imputation, 14 best-ranked features were selected using a multistep feature selection process including the Logistic Regression and Fast Correlation-Based Filter. Thirteen supervised ML algorithms were trained with stratified five-fold cross-validation on the training set and validated on the validation set. The ML algorithms with a Log Loss of <0.688 and <0.672 in the cross-validation and validation steps were evaluated on the testing set. Performance matrices of the algorithms assessed included area under the curve (AUC), classification accuracy (CA), F1 score, precision, recall, and specificity. To provide transparency and interpretability, SHapley Additive exPlanations (SHAP) values were assessed and depicted as beeswarm plots. Results Two hundred thirty-nine patients (214 males; mean age 56 ± 11 years) were enrolled in the study. There were 371 perfusion defects (321 in the training and validation sets; 50 in the testing set). Based on the reference standard, 168 perfusion defects had hibernating myocardium (139 in the training and validation sets; 29 in the testing set). On cross-validation, six ML algorithms with Log Loss <0.688 had AUC >0.800. On validation, 10 ML algorithms had a Log Loss value <0.672, among which six had AUC >0.800. On model evaluation of the selected models on the unseen testing set, nine ML models had AUC >0.800 with Gradient Boosting Random Forest (xgboost) [GB RF (xgboost)] achieving the highest AUC of 0.860 and could detect the presence of hibernating myocardium in 21/29 (72.4%) perfusion defects with a precision of 87.5% (21/24), specificity 85.7% (18/21), CA 78.0% (39/50) and F1 Score 0.792. Four models depicted a clear pattern of model interpretability based on the beeswarm SHAP plots. These were GB RF (xgboost), GB (scikit-learn), GB (xgboost), and Random Forest. Conclusion Our study demonstrates the potential of ML in detecting hibernating myocardium using radiomics features extracted from perfusion defects on rest MPI images. This proof-of-concept underscores the notion that radiomics features capture nuanced information beyond what is perceptible to the human eye, offering promising avenues for improved myocardial viability assessment.

ABSTRACTBackgroundAcute myocardial infarction (AMI) remains a leading global cause of mortality. This study explores predictors of in‐hospital mortality among AMI patients using advanced machine learning (ML) techniques.MethodsData from 7422 AMI patients treated with percutaneous coronary intervention (PCI) at Tehran Heart Center (2015–2021) were analyzed. Fifty‐eight clinical, demographic, and laboratory variables were evaluated. Seven ML algorithms, including Random Forest (RF), logistic regression with LASSO, and XGBoost, were implemented. The data set was divided into training (70%) and testing (30%) subsets, with fivefold cross‐validation. The class imbalance was addressed using the synthetic minority oversampling technique (SMOTE). Model predictions were interpreted using SHapley Additive exPlanations (SHAP).ResultsIn‐hospital mortality occurred in 129 patients (1.74%). RF achieved the highest predictive performance, with an area under the curve (AUC) of 0.924 (95% CI 0.893–0.954), followed by XGBoost (AUC 0.905) and logistic regression with LASSO (AUC 0.893). Sensitivity analysis in STEMI patients confirmed RF's robust performance (AUC 0.900). SHAP analysis identified key predictors, including lower left ventricular ejection fraction (LVEF; 33.24% vs. 43.46% in survivors, p < 0.001), higher fasting blood glucose (190.38 vs. 132.29 mg/dL, p < 0.001), elevated serum creatinine, advanced age (70.92 vs. 61.88 years, p < 0.001), and lower LDL‐C levels. Conversely, BMI showed no significant association (p = 0.456).ConclusionML algorithms, particularly RF, effectively predict in‐hospital mortality in AMI patients, highlighting critical predictors such as LVEF and biochemical markers. These insights offer valuable tools for enhancing clinical decision‐making and improving patient outcomes.

The aim of the study was to inspect the acoustic properties and sleep characteristics of a preapneic snoring sound. The feasibility of forecasting upcoming respiratory events by snoring sound was also investigated. Participants with habitual snoring or a heavy breathing sound during sleep were recruited consecutively. Polysomnography was conducted, and snoring-related breathing sound was recorded simultaneously. Acoustic features and sleep features were extracted from 30-second samples, and a machine learning algorithm was used to establish 2 prediction models. A total of 74 eligible participants were included. Model 1, tested by 5-fold cross-validation, achieved an accuracy of 0.92 and an area under the curve of 0.94 for respiratory event prediction. Model 2, with acoustic features and sleep information tested by Leave-One-Out cross-validation, had an accuracy of 0.78 and an area under the curve of 0.80. Sleep position was found to be the most important among all sleep features contributing to the performance of the 2 models. Preapneic sound presented unique acoustic characteristics, and snoring-related breathing sound could be deployed as a real-time apneic event predictor. The models, combined with sleep information, serve as a promising tool for an early warning system to forecast apneic events. Wang B, Yi X, Gao J, etal. Real-time prediction of upcoming respiratory events via machine learning using snoring sound signal. J Clin Sleep Med. 2021;17(9):1777-1784.

Machine learning for classifying chronic ankle instability based on ankle strength, range of motion, postural control and anatomical deformities in delivery service workers with a history of lateral ankle sprains.