10-fold Cross-validation Research Articles

The machine learning-based model for lateral lymph node metastasis of thyroid medullary carcinoma improved the prediction ability of occult metastasis.

For medullary thyroid carcinoma (MTC) with no positive findings in the lateral neck before surgery, whether prophylactic lateral neck dissection (LND) is needed remains controversial. A better way to predict occult metastasis in the lateral neck is needed. From January 2010 to January 2022, patients who were diagnosed with MTC and underwent primary surgery at our hospital were retrospectively reviewed. We collected the patients' baseline characteristics, surgical procedure, and rescored the ultrasound images of the primary lesions using American College of Radiology (ACR) Thyroid Imaging, Reporting and Data System (TI-RADS). Regularized logistic regression, 5-fold cross-validation and decision curve analysis was applied for lateral lymph node metastasis (LLNM) model's development and validation. Then, we tested the predictive ability of the LLNM model for occult LLNM in cN0-1a patients. A total of 218 patients were enrolled. Five baseline characteristics and two TI-RADS features were identified as high-risk factors for LLNM: gender, baseline calcitonin (Ctn), tumor size, multifocality, and central lymph node (CLN) status, as well as TI-RADS margin and level. A LLNM model was developed and showed a good discrimination with 5-fold cross-validation mean area under curve (AUC) = 0.92 ± 0.03 in the test dataset. Among cN0-1a patients, our LLNM model achieved an AUC of 0.91 (95% CI, 0.88-0.94) for predicting occult LLNM, which was significantly higher than the AUCs of baseline Ctn (0.83) and CLN status (0.64). We developed a LLNM prediction model for MTC using machine learning based on clinical baseline characteristics and TI-RADS. Our model can predict occult LLNM for cN0-1a patients more accurately, then benefit the decision of prophylactic LND.

Investigating the Precise Identification of Citrullination Sites with High-Performance Score Metrics using a Powerful Computation Predicting Tool.

To elucidate the detailed mechanisms of citrullination at the molecular level and design drugs applicable to major human diseases, predicting protein citrullination sites (PCSs) is essential. Using experimental approaches to predict PCSs is time-consuming and costly. However, there is a limited scope of the current PCS predictors. In particular, most predictors are commonly used for PCS prediction and have limited performance scores. This work aims to provide an improved sophisticated predictor of citrullination sites using a benchmark dataset in a machine learning platform. This study presents a reliable citrullination site predictor based on a benchmark dataset containing a 1:1 ratio of positive and negative samples. We classified citrullination sites using the Composition of the K-Spaced Amino Acid Pairs (CKSAAP) and Support Vector Machine (SVM). We developed PCS predictors using integrated machine-learning methods that produced the highest average scores. Using 10-fold cross-validation on test datasets, the True Positive Rate (TPR) was 98.34%, the True Negative Rate (TNR) was 99.44%, the accuracy was 98.89%, the Mathew Correlation Coefficient (MCC) was 98.21%, the Area Under the ROC Curve (AUC) was 0.999, and the partial Area Under the ROC Curve (pAUC) was 0.1968. According to overall performance, our developed predictor has a significantly higher implementation in comparison with the current tools on the same benchmark dataset. Moreover, it showed better performance metrics on both test and training datasets. Our developed predictor is promising and can be implemented as a complementary technique for identifying fast and precise citrullination sites.

Engineering G protein-coupled receptors for stabilization.

G protein-coupled receptors (GPCRs) are one of the most important families of targets for drug discovery. One of the limiting steps in the study of GPCRs has been their stability, with significant and time-consuming protein engineering often used to stabilize GPCRs for structural characterization and drug screening. Unfortunately, computational methods developed using globular soluble proteins have translated poorly to the rational engineering of GPCRs. To fill this gap, we propose GPCR-tm, a novel and personalized structurally driven web-based machine learning tool to study the impacts of mutations on GPCR stability. We show that GPCR-tm performs as well as or better than alternative methods, and that it can accurately rank the stability changes of a wide range of mutations occurring in various types of class A GPCRs. GPCR-tm achieved Pearson's correlation coefficients of 0.74 and 0.46 on 10-fold cross-validation and blind test sets, respectively. We observed that the (structural) graph-based signatures were the most important set of features for predicting destabilizing mutations, which points out that these signatures properly describe the changes in the environment where the mutations occur. More specifically, GPCR-tm was able to accurately rank mutations based on their effect on protein stability, guiding their rational stabilization. GPCR-tm is available through a user-friendly web server at https://biosig.lab.uq.edu.au/gpcr_tm/.

Advanced Ensemble Framework for Diabetes Outcome Forecasting

Abstract: Diabetes is one of the chronic diseases, which is increasing from year to year. Developing an automated system that can detect diabetes patients plays an important role in medical science. A stack classifier model is designed for the detection of diabetes by combining three base estimators such as Random Forest Classifier, LightGBM Classifier, and K-Nearest Neighbors Classifier, and Logistic Regression as meta-classifiers. The data preprocessing includes transforming categorical variables into numerical format. Each base learner is trained on the preprocessed data and predictions are made. In the meta learner stage, Logistic Regression is trained to make predictions based on the predictions of the base learners. The goal of the meta learner is to learn how to combine the predictions of the base learners to make a more accurate final prediction on whether the patient is diabetic or not. The use of a stacking classifier improves prediction accuracy compared to using a single classifier. The developed model gives an accuracy of 98%. The 5-fold cross validation is used to get a more robust estimation of generalization error. Thus, the developed model offers a means to enhance early detection and elevate the quality of care for diabetic patients.

Global Weight Optimization on Artificial Neural Network for Optimizing Crop Yield Prediction Using Genetic Algorithm

Abstract: Accurate crop yield prediction is a complex and intricate task that encompasses a multitude of factors and variables, rendering it arduous to establish a dependable mathematical model. Conventional machine learning (ML) models for crop yield prediction have surpassed their efficacy. This research endeavour sought to enhance crop yield prediction accuracy by hybridizing a neuro-genetic model utilizing statistical data amassed over a 35-year period from various pertinent agricultural departments in Tamil Nadu, India, including the Statistical, Agricultural, and Meteorological Departments. This research delved into analyzing and identifying the optimal weight configuration for the artificial neural network (ANN) to bolster accuracy with the assistance of genetic algorithms (GA). Seventy-five percent of the data was employed to train the model, while the remaining 25% was utilized for model testing. To gauge the performance of this research work, 5-fold cross-validation was implemented. RMSE, MAE, and Adj R2 were employed to evaluate the performance and contrast the performance of the neuro-genetic model with the conventional ANN. The neuro-genetic model exhibited superior accuracy compared to the conventional ANN.

DeepRoughNetID: A Robust Framework for Network Anomaly Intrusion Detection with High Detection Rates

Network security faces challenges, including reduced true positives, increased false positives, and inadequate anomaly detection efficacy. This paper introduces the DeepRoughNetID (DRNID) approach to address these issues and confront the increasing cyberattack threat in modern communication systems. DRNID presents an innovative framework for network intrusion detection, incorporating kNNImputer, GreedyRoughSelector, DeepVAEEnsembler, and IntrusionNet components. kNNImputer enables adaptable data preprocessing by leveraging instance-based learning, facilitating efficient handling of evolving datasets without necessitating full retraining. GreedyRoughSelector enhances classifier performance through systematic attribute selection, focusing on relevant features while eliminating redundancies. DeepVAEEnsembler leverages Variational Autoencoders (VAEs) to learn underlying data distributions, enabling robust anomaly detection. IntrusionNet utilizes VAEs to classify intrusions, providing a comprehensive solution to network security challenges. By integrating these components, DRNID offers a refined approach that promises significant advancements in network intrusion detection. The experimental results demonstrate the effectiveness of DRNID, with an impressive accuracy of 98.8% achieved through 10-fold cross-validation. DRNID exhibits superior performance compared to existing methods across various datasets, showcasing its robust anomaly detection capabilities. This methodology empowers organizations to proactively prevent security violations and safeguard sensitive data from malicious entities. Beyond its theoretical contributions, our research carries tangible implications for strengthening cybersecurity defenses across diverse sectors, including telecommunications, finance, and healthcare.

Prediction model for asymptomatic carotid atherosclerosis using retinal microvascular intelligent analysis: A retrospective study

ImportanceEarly detection and timely diagnosis of asymptomatic carotid atherosclerosis significantly assist in the prevention of ischemic stroke for them. ObjectiveThis observational study aimed to develop and validate a novel prediction model to assist in the early diagnosis of carotid atherosclerosis based on new characteristic variables screened by retinal microvascular intelligence analysis. Main Outcome(s) and Method (s)The least absolute shrinkage and selection operator (LASSO) combined with 10-fold cross-validation were screened for characteristic variables, and nomograms were plotted to demonstrate the prediction model. Receiver operating characteristic (ROC) curves and area under the curve (AUC), calibration plots and brier score (BS), and decision curve analysis (DCA) were used to evaluate the risk model's discrimination, calibration, and clinical applicability. ResultsAge, gender, diabetes mellitus (DM), drinking history, vascular branching angle, mean vascular diameter within 0.5–1.0 papillary diameter (PD), curvature tortuosity arteriole in the inferior region of the optic disc, and vascular density in the nasal region of the optic disc were identified as characteristic variables for carotid atherosclerosis with retinal microvascular intelligence analysis. The predictive nomogram model presented good discrimination with AUCs of 0.790 (0.774–0.806), and the calibration curve displayed high consistency between predicted and actual probability. The DCA demonstrated that this nomogram model led to net benefits in a threshold probability range of 20 %–94 % and could be adapted for clinical decision-making. The results of the 100-bootstrap resampling strategy for internal validation also show that the risk model is well discriminated with an AUC of 0.789 and excellent calibration. External validation showed good discrimination with AUCs of 0.703 (0.627 - 0.779) and good calibration, the risk threshold is 10 %–92 % in terms of DCA. Conclusions and RelevanceThe novel prediction model based on retinal microvascular intelligence analysis constructed in this study could be effective prognoses for predicting the risk of asymptomatic carotid atherosclerosis in a Chinese screening population.

A Study on the Screening of Children at Risk for Developmental Disabilities Using Facial Landmarks Derived From a Mobile-Based Application

Objective Early detection and intervention of developmental disabilities (DDs) are critical to improving the long-term outcomes of afflicted children. In this study, our objective was to utilize facial landmark features from mobile application to distinguish between children with DDs and typically developing (TD) children.Methods The present study recruited 89 children, including 33 diagnosed with DD, and 56 TD children. The aim was to examine the effectiveness of a deep learning classification model using facial video collected from children through mobile-based application. The study participants underwent comprehensive developmental assessments, which included the child completion of the Korean Psychoeducational Profile-Revised and caregiver completing the Korean versions of Vineland Adaptive Behavior Scale, Korean version of the Childhood Autism Rating Scale, Social Responsiveness Scale, and Child Behavior Checklist. We extracted facial landmarks from recorded videos using mobile application and performed DDs classification using long short-term memory with stratified 5-fold cross-validation.Results The classification model shows an average accuracy of 0.88 (range: 0.78–1.00), an average precision of 0.91 (range: 0.75–1.00), and an average F1-score of 0.80 (range: 0.60–1.00). Upon interpreting prediction results using SHapley Additive exPlanations (SHAP), we verified that the most crucial variable was the nodding head angle variable, with a median SHAP score of 2.6. All the top 10 contributing variables exhibited significant differences in distribution between children with DD and TD (p<0.05).Conclusion The results of this study provide evidence that facial landmarks, utilizing readily available mobile-based video data, can be used to detect DD at an early stage.

Heart rate is an important predictor of pacing induced cardiomyopathy in a machine learning analysis

Abstract Background 10-20% of patients who have right ventricular pacing develop pacing induced cardiomyopathy (PICM). Home monitoring (HM) could identify patients developing PICM in real-time and facilitate echocardiographic confirmation, medication optimisation and if necessary cardiac resynchronisation therapy (CRT). Purpose To identify predictors of PICM using machine learning models. Methods Consecutive cases of patients who developed PICM (12) after dual chamber pacing were identified from our implant database and matched 4:1 with propensity matched controls who underwent pacing but did not develop PICM (48) (Table 1.) HM data from the 6 months prior to echocardiographic diagnosis of PICM was analysed. PICM was defined as (1) a deterioration in left ventricular (LV) ejection fraction (EF) of >10%, (2) to an EF of <45%, (3) with right ventricular (RV) pacing >20% and (4) the exclusion of alternative causes. Three types of ML models were generated (1) random forest, (2) extreme gradient boosted (XGBoost) (3) K-nearest neighbours algorithm (KNN), with 10-fold cross-validation. Results HM data were available for 12 PICM cases (75 ± 8 years, 72% male). PICM occurred at 2.0 ± 1.2 years. There was no difference between cases and controls in atrial pacing (8±15% vs 18±22% p=0.14), ventricular pacing (81±29% vs 85±24% p=0.62) or base rate (53±5bpm vs 53±5bpm p=0.80). Of the 34 RM features analysed, mean ventricular heart rate was identified as the most important predictor of PICM; area under curve with XG Boost 0.95, KNN 0.95, and random forest 0.86. Other RM features failed to add any more predictive power to the models. Mean ventricular rate was significantly higher in cases than controls (80±6 bpm vs 69±9 bpm p<0.001). Sensitivity analysis was performed to assess whether episodes of atrial fibrillation (AF) were responsible for the higher ventricular rates. After days of AF days were excluded, mean ventricular rate was still significantly higher in cases than controls (79± 6vs 69±9bpm p<0.001) (Figure 1). 10/12 patients had a subsequent CRT upgrade. No consistent pattern emerged in heart rate changes in the first 3 months after CRT upgrade. Conclusion In a group of patients with normal left ventricular function at implant, remotely monitored heart rate can predict early-onset PICM. Further work is required to understand when these heart rate differences emerge, and whether they improve with positive LV remodelling.Table 1.Figure 1.

An integrated model combining BERT and tree-augmented naive Bayes for analyzing risk factors of construction accident

PurposeLearning from past construction accident reports is critical to reducing their occurrence. Digital technology provides feasibility for extracting risk factors from unstructured reports, but there are few related studies, and there is a limitation that textual contextual information cannot be considered during extraction, which tends to miss some important factors. Meanwhile, further analysis, assessment and control for the extracted factors are lacking. This paper aims to explore an integrated model that combines the advantages of multiple digital technologies to effectively solve the above problems.Design/methodology/approachA total of 1000 construction accident reports from Chinese government websites were used as the dataset of this paper. After text pre-processing, the risk factors related to accident causes were extracted using KeyBERT, and the accident texts were encoded into structured data. Tree-augmented naive (TAN) Bayes was used to learn the data and construct a visualized risk analysis network for construction accidents.Findings The use of KeyBERT successfully considered the textual contextual information, prompting the extracted risk factors to be more complete. The integrated TAN successfully further explored construction risk factors from multiple perspectives, including the identification of key risk factors, the coupling analysis of risk factors and the troubleshooting method of accident risk source. The area under curve (AUC) value of the model reaches up to 0.938 after 10-fold cross-validation, indicating good performance.Originality/value This paper presents a new machine-assisted integrated model for accident report mining and risk factor analysis, and the research findings can provide theoretical and practical support for accident safety management.

Predicting atrial fibrillation ablation success: a novel approach using pre-procedure sinus rhythm ecg and deep learning

Abstract Introduction Pulmonary vein isolation (PVI) has been the corner stone of the treatment of patients with atrial fibrillation (AF), yet its efficacy varies notably between patient groups. Specifically, it shows reduced efficacy in individuals with persistent AF and those with atrial fibrosis. Recent advancements suggest that additional substrate modification could enhance outcomes in specific patient groups. Therefore, predicting the outcomes of lone PVI could help in stratifying patients who would benefit most from this additional substrate modification. Addressing this gap, our study explores the prognostication of PVI outcomes using pre-procedural 12-lead ECG recordings taken in Sinus rhythm, employing advanced deep learning techniques. Methods This retrospective study analyzed 209 patients with paroxysmal and persistent AF who underwent PVI. These patients were followed up for 12 months after the blanking period. Pre-procedural sinus rhythm 12-lead ECGs were processed using an innovative deep learning architecture, integrating Long Short-Term Memory and Convolutional Neural Networks with a cross-attention module. This multimodal deep learning model was trained using ECG and clinical covariates like age, sex, and BMI to predict binary outcomes – AF type (Paroxysmal or Persistent) and AF recurrence. The model's efficacy was evaluated through a 5-fold stratified cross-validation to ensure accuracy and reliability. Model was also tested using only ECG data. Results The multimodal deep learning model performed significantly better as compared to the model trained using ECG data only. The multimodal deep learning model achieved an accuracy of 75.56% and a sensitivity of 82% in prognosticating ablation outcomes. Additionally, the model demonstrated a significant capability in classifying AF into paroxysmal or persistent types with an accuracy of 86.67% and a recall of 89%. Conclusion Our deep learning approach in analyzing sinus rhythm ECG recordings shows promising results in predicting PVI outcomes and identifying AF types. This approach could offer a more accessible and cost-effective tool to stratify AF patients undergoing PVI, identifying those less likely to benefit from standard procedures. Such insights can assist physicians and patients in determining ablation candidacy and selecting the most appropriate ablation strategy and technique.Proposed StrategyDeep learning model approach

Machine Learning to Predict Drug-Induced Liver Injury and Its Validation on Failed Drug Candidates in Development

Drug-induced liver injury (DILI) poses a significant challenge for the pharmaceutical industry and regulatory bodies. Despite extensive toxicological research aimed at mitigating DILI risk, the effectiveness of these techniques in predicting DILI in humans remains limited. Consequently, researchers have explored novel approaches and procedures to enhance the accuracy of DILI risk prediction for drug candidates under development. In this study, we leveraged a large human dataset to develop machine learning models for assessing DILI risk. The performance of these prediction models was rigorously evaluated using a 10-fold cross-validation approach and an external test set. Notably, the random forest (RF) and multilayer perceptron (MLP) models emerged as the most effective in predicting DILI. During cross-validation, RF achieved an average prediction accuracy of 0.631, while MLP achieved the highest Matthews Correlation Coefficient (MCC) of 0.245. To validate the models externally, we applied them to a set of drug candidates that had failed in clinical development due to hepatotoxicity. Both RF and MLP accurately predicted the toxic drug candidates in this external validation. Our findings suggest that in silico machine learning approaches hold promise for identifying DILI liabilities associated with drug candidates during development.

#958 Development and Validation of a Score Estimating the Probability of obtaining a genetic diagnosis in adults with CKD of uncertain origin

Abstract Background and Aims We aim to derive and validate a score estimating the probability of identifying a variant explicative of the renal phenotype using Exome Sequencing (ES) in adults with CKD of uncertain origin. Method Participants: prospective cohort study including all consecutive index patients with a CKD from uncertain origin in three French nephrology units (metropolitan: Sorbonne University Hospitals, Paris, and Conception Hospital, Marseille; overseas: La Réunion University Hospital, Réunion island) who underwent ES between October 11th, 2017 and May 31st, 2023. Outcome measure: identification of a causal variant using ES data, according to the American College of Medical Genetics and Genomics diagnostic criteria. Candiate variables and feature engineering: raw data regarding patients’ phenotypes was prospectively entered in a collection form at the time of ES by the prescribing physician. Structured data (yes/no questions and lists) was mapped to Human Phenotype Ontology (HPO) terms. Unstructured data (free text) was manually translated into HPO terms by the same investigator (nephrologist), which was blinded to ES results. First, second and third-order ancestor HPO terms were included in the analysis. Score's derivation: an optimal weighted average of multiple models (“ensemble model”) was specified, following guidelines provided in a recently published article (PMID:36905602). Score's validation: internal validation using 10-fold cross validation, followed by internal-external validation (data from 2/3 centers was used to develop a model which was then tested using data from the remaining center; this process was repeated in each center). Results We included 2,490 patients, 560/2,490 (22.5%) of whom had a causal variant identified by ES. We collected 1,028 distinct HPO terms describing the patients’ phenotypes. The most common term was “Hypertension”, occurring in 2,079/2,490 (83.5%) patients. In internal validation, the score showed accurate calibration and discrimination (area under the receiver operating characteristics curve (AUC): 0.71, 95% CI 0.68 to 0.73; index of precision accuracy (IPA): 0.11, 95% CI 0.09 to 0.13). Performances were moderately diminished in internal-external validation but remained satisfactory (Tenon: AUC 0.64, IPA 0.04; Marseille: AUC 0.67, IPA 0.04; La Réunion: 0.68, IPA 0.09). Conclusion We derived, internally and internally-externally validated a clinical score which accurately predicts the probability of obtaining a genetic diagnosis using Exome Sequencing. Feature engineering could further increase the score's performances. Automatic computation by clinicians using an interactive tool is achievable and could be of clinical utility, especially to guide resource allocation in case of a low pretest probability.

Epistatic Features and Machine Learning Improve Alzheimer's Disease Risk Prediction Over Polygenic Risk Scores.

Polygenic risk scores (PRS) are linear combinations of genetic markers weighted by effect size that are commonly used to predict disease risk. For complex heritable diseases such as late-onset Alzheimer's disease (LOAD), PRS models fail to capture much of the heritability. Additionally, PRS models are highly dependent on the population structure of the data on which effect sizes are assessed and have poor generalizability to new data. The goal of this study is to construct a paragenic risk score that, in addition to single genetic marker data used in PRS, incorporates epistatic interaction features and machine learning methods to predict risk for LOAD. We construct a new state-of-the-art genetic model for risk of Alzheimer's disease. Our approach innovates over PRS models in two ways: First, by directly incorporating epistatic interactions between SNP loci using an evolutionary algorithm guided by shared pathway information; and second, by estimating risk via an ensemble of non-linear machine learning models rather than a single linear model. We compare the paragenic model to several PRS models from the literature trained on the same dataset. The paragenic model is significantly more accurate than the PRS models under 10-fold cross-validation, obtaining an AUC of 83% and near-clinically significant matched sensitivity/specificity of 75%. It remains significantly more accurate when evaluated on an independent holdout dataset and maintains accuracy within APOE genotype strata. Paragenic models show potential for improving disease risk prediction for complex heritable diseases such as LOAD over PRS models.

#2271 Cardiovascular event prediction in kidney transplant recipients: a new risk score derivation and internal validation

Abstract Background and Aims Kidney transplant recipients (KTRs) have a cardiovascular (CV) risk 3 to 5-fold higher than that of the general population.CV disease represents the main cause of death in KTRs, both due to the presence of traditional risk factors and factors strictly related to transplant. The main CV risk calculators proved to be poorly predictive in the KTR population. We, therefore, aimed to compare the performance of CV risk scores already in use and to develop and validate a CV risk score in our KTRs population. Method Our analysis included 371 adult KTRs in follow-up at the outpatient clinic from 1 January 2015 to 31 December 2016. The composite outcome was the occurrence of MACE. We compared the performance of three main CV risk calculators (Framingham Risk Score-FRS-, INDANA score and PORT score). Thereafter, we built a predictive model selecting study variables by backward stepwise Cox regression. The risk score for each variable was weighted according to the HR of the final multivariable model. Internal and external validation of the prediction score and its discriminative capacity was assessed via area under the curve (AUC) curve, and the calibration via the Hosmer–Lemeshow test. Results After a mean follow-up of 68 months, CV events occurred in 71 (19%) KTRs. FRS, INDANA and PORT were demonstrated to have low predictive power in our population (Fig. 1). The accuracy did not improve after adjustment for immunological and transplant-related variables. Therefore, we derived a model including significative variables at multivariate analysis: age, dialysis vintage, systolic blood pressure, eGFR and 24-hour proteinuria (Akaike information criteria 713). The model discrimination was good (Harrel's c: 0.73). According to the hazard ratio, 3 points have been attributed to age higher than 60 years and 24 h proteinuria higher than 1 g/d, 2 points to dialysis vintage longer than 5 years, systolic blood pressure higher than 140 mmHg and eGFR less than 30 ml/min/1.73 m2. The new score demonstrated good discrimination (AUC 0.70, Fig. 2) and calibration (Hosmer-Lemeshow test 11.34, P = 0.12). The new score was internally validated by 10-fold cross-validation (mean AUC 0.70, 95% CI 0.60-0.77). For each point of the new score, the risk of the event increased by 40% and a score higher than three was associated with a 4-fold increased risk of composite endpoint. Conclusion We confirmed the poor predictive power of the known CV risk assessment tools in our real-world KTRs population. Moreover, their discriminatory power was not improved by adjustment for the occurrence of biopsy-proven rejection, HLA compatibility and donor-specific antibody. A new score for CV event prediction including traditional risk factors, such as age and systolic blood pressure, and variables related to renal function and the effects of previous chronic kidney disease, was developed and internally validated.

An auditory brain-computer interface based on selective attention to multiple tone streams.

In this study, we attempted to improve brain-computer interface (BCI) systems by means of auditory stream segregation in which alternately presented tones are perceived as sequences of various different tones (streams). A 3-class BCI using three tone sequences, which were perceived as three different tone streams, was investigated and evaluated. Each presented musical tone was generated by a software synthesizer. Eleven subjects took part in the experiment. Stimuli were presented to each user's right ear. Subjects were requested to attend to one of three streams and to count the number of target stimuli in the attended stream. In addition, 64-channel electroencephalogram (EEG) and two-channel electrooculogram (EOG) signals were recorded from participants with a sampling frequency of 1000 Hz. The measured EEG data were classified based on Riemannian geometry to detect the object of the subject's selective attention. P300 activity was elicited by the target stimuli in the segregated tone streams. In five out of eleven subjects, P300 activity was elicited only by the target stimuli included in the attended stream. In a 10-fold cross validation test, a classification accuracy over 80% for five subjects and over 75% for nine subjects was achieved. For subjects whose accuracy was lower than 75%, either the P300 was also elicited for nonattended streams or the amplitude of P300 was small. It was concluded that the number of selected BCI systems based on auditory stream segregation can be increased to three classes, and these classes can be detected by a single ear without the aid of any visual modality.

#2714 An AI based tool for hydration management in dialysis patients.

Abstract Background and Aims Hemodialysis patients often experience over-hydration (OH), a critical factor derived from excess extracellular water and associated with cardiovascular complications and overall morbidity. Accurate estimation of pre-dialysis OH is essential for optimizing fluid management and improving patient outcomes. Traditional methods for assessing OH, such as the Body Composition Monitor (BCM) are, in general, time-consuming for clinical staff and costly for healthcare providers. In this study, we present a machine learning model combining data automatically recorded by dialysis machine sensors and medical information captured in electronic chart records to estimate pre-dialysis OH. The aim of this study is to evaluate the possibility of a tool that predicts OH from patient's data without, or less frequently using, a bioimpedance device for OH measurements. Method We collected a comprehensive dataset from a cohort of CKD patients undergoing hemodialysis, including demographic information, clinical parameters, and hydration status parameters. The value of OH was assessed using BCM device. A feature selection approach, guided by a balance between prediction accuracy and model complexity, was used to identify the optimal set of input variables. All variables utilized in the model are either regularly gathered in clinical practice or automatically recorded from the dialysis machine. We tried different combinations of features using an XGBoost regressor algorithm. We focused the study on the comparison of the prediction of OH with the input variable of the average BCM OH value in the past year (Model 1) and without this variable (Model 2). Mean absolute error (MAE), Root mean square error (RMSE), R2, calibration and residuals plots were used to assess the models’ accuracy. The measurement unit is in litres. The models were validated with nested 5-fold cross-validation, dividing the data repeatedly into disjunct train, validation and test folds and 5 outer loops, totalling 25 folds. Results The dataset contains the observations from 37,011 patients for a total of 934,386 BCM assessments collected in the EuCliD database by Fresenius Medical Care clinics in Cech Republic, Italy, Portugal, Slovakia, Spain, and the United Kingdom between January 2016 to September 2023. The average value of OH is 2.06 (std 1.88). Model 1 generalizes on unseen data samples with an MAE of 0.74, RMSE of 1.02 and R2 of 0.57 after evaluating the performance on the test folds. Model 2 MAE is 0.93, RMSE is 1.25 and R2 is 0.36. The degradation of the accuracy between the two models indicates the importance of patients’ past BCM information for the prediction of the next OH. This degradation is also clear in the calibration and residuals plots (Fig. 1). Nevertheless, model 2 is able to predict OH with an acceptable level of accuracy. Conclusion Fluid management is a critical task for physicians and nurses. We present an AI-based tool that can be utilized to support clinicians in their daily decisions about fluid management, with and without the use of a device to measure OH. The results, as expected, show an important degradation of the accuracy between model 1 and 2, however, model 2 has sufficient overall accuracy to consider the approach valuable, meriting further exploration. In general, this approach can be used to reduce the number of direct OH assessments.

Prediction of main engine power of oil tankers using artificial intelligence algorithms

In the preliminary ship design, the accurate determination of a vessel’s main engine power is one of the most critical aspects next to service speed, main particulars, and cargo capacity. However, this task can be quite intricate due to its reliance on an extremely great number of influencing factors. In the research that is presented in this paper dataset of 357 oil tankers was gathered and developed to research the idea in which genetic programming is applied to the mentioned dataset to obtain mathematical equations (MEs) that can estimate the ship’s main engine power with high accuracy. The highest estimation accuracy of MEs is achieved by tuning the GP hyperparameter values through the random hyperparameter search (RHS) method. The initial dataset was divided into train and test datasets in a 70:30 ratio. The train dataset was used to train GP in a 5-fold cross-validation process and after the process was done the obtained MEs were evaluated on the test dataset. To evaluate the GP training testing process several evaluation metrics were used i.e., coefficients of determination (R2), mean absolute error (MAE), root mean square error (RMSE), and length of obtained MEs. The conducted investigation showed that GP generated MEs that can estimate ship main engine power with high accuracy.

Prediction model of early recurrence of multimodal hepatocellular carcinoma with tensor fusion.

Clinical decision-making in oncology involves multimodal data, encompassing histopathological, radiological, and clinical factors. Several computer-aided multimodal decision-making systems have emerged in recent years to predict the recurrence of hepatocellular carcinoma (HCC) after hepatectomy, but they tend to employ simplistic feature-level concatenation, resulting in redundancy and hampering overall performance. More notably, these models often lack an effective integration with clinical relevance. Particularly, they still face major challenges of integrating data from diverse scales and dimensions, and introducing a liver background, which are clinically significant but previously overlooked aspects.

 In addressing these challenges, we provide new insight in two areas. Firstly, we introduce the tensor fusion method into the model, which demonstrates unique advantages in handling the fusion of multi-scale and multi-dimensional data, thus potentially enhancing the model's performance. Secondly, to our best knowledge, it's a precedent to take the impact of the liver background into account. We innovatively incorporate the impact of the liver background into the feature extraction process by using a deep learning segmentation-based algorithm. This inclusion makes the model closer to real-world clinical scenarios, as the liver background may contain vital information related to postoperative recurrence.
 
 We collected radiomics (MRI) and histopathological images of 176 cases diagnosed by experienced clinicians from two independent centers. Our proposed network went through training and 5-fold cross-validation on the dataset of 176 cases and was subsequently validated on an independent external dataset of 40 cases. Finally, our proposed network exhibited excellent performance in predicting the postoperative early recurrence of HCC with an AUC of 0.883. These results suggest significant progress in addressing the challenges related to multimodal data fusion which provides potential value for more accurate predictions of clinical outcomes.

A clinical nomogram for predicting occult lymph node metastasis in patients with non-small cell lung cancer≤2 cm.

Sublobar resection has been shown to be feasible for non-small cell lung cancers less than 2 cm in size based on several prospective studies. However, the prognosis of clinical N0 patients who experience an N-stage upgrade after surgery (known as occult lymph node metastasis) may be worse. The ability of predict occult lymph node metastasis in patients eligible for sublobar resection remains a controversial issue. Patients with non-small cell lung cancer ≤2 cm in diameter and containing a solid component who underwent surgical treatment at the Affiliated Hospital of Qingdao University were retrospectively enrolled, and 1:1 case matching was performed. The risk factors were identified through logistic regression analyses and theoretical criteria, followed by the development of a nomogram that was evaluated using 200 iterations of 10-fold cross-validation. After case matching, 130 pairs of patients were selected for modelling. According to the multivariable logistic regression analysis, the carcinoembryonic antigen level, consolidation tumour ratio, mean computed tomography number and tumour margin were included in the nomogram. The cross-validated average area under the receiver operating characteristic curve was found to be 0.86. Furthermore, calibration curve and decision curve analyses demonstrated the excellent predictive accuracy and clinical utility of the nomogram respectively. By utilizing accessible characteristics, we developed a nomogram that predicts the probability of occult lymph node metastasis in patients with NSCLC ≤2 cm with a solid component. Risk stratification with this nomogram could aid in surgical method decision-making.