Performance Of Prediction Models Research Articles

Machine learning driven benchtop Vis/NIR spectroscopy for online detection of hybrid citrus quality.

The aim of this study was to explore application of visible and near-infrared (Vis/NIR) spectroscopy combined with machine learning models for SSC and TA prediction of hybrid citrus. The Vis/NIR spectra of samples including navel-region, equator-region and multi-region combination spectra in navel-region and equator-region were collected using a benchtop equipment. The performance of SSC and TA prediction models with different region spectra, including partial least squares (PLS), random forest (RF), k-nearest neighbors (KNN), support vector machine (SVM) and multilayer feedforward neural network (MFNN), was assessed. The accuracy of SSC and TA prediction models with multi-region combination (raw) spectra was better compared to navel-region and equator-region, suggesting that multi-region combination spectra collection method was more suitable. Subsequently, the spectral pre-processing, including Savitzky-Golay smoothing (SGS), maximum normalization (MN), multiplicative scatter correction (MSC), linear baseline correction (LBC) and first derivative (1stD), were performed. The performance of SSC and TA prediction models with different pre-processing spectra was further compared. The PLS with SGS spectra (SGS-PLS) and MFNN with raw spectra (Raw-MFNN) exhibited superior validation effects for SSC and TA prediction, respectively. In a subsequent prediction in new samples, SGS-PLS achieved an RP2 of 0.875, an RMSEP of 0.572% and a MAEP of 0.469% for SSC prediction, and Raw-MFNN achieved an RP2 of 0.800, an RMSEP of 0.0322% and a MAEP of 0.0249% for TA prediction, indicating excellent generalization ability. These results indicate the great potential of benchtop Vis/NIR spectroscopy for online detection of hybrid citrus quality at mass-scale level.

Prognostic prediction for inflammatory breast cancer patients using random survival forest modeling.

Inflammatory breast cancer (IBC) is an aggressive and rare phenotype of breast cancer, which has a poor prognosis. Thus, it is necessary to establish a novel predictive model of high accuracy for the prognosis of IBC patients. Clinical information of 1,230 IBC patients from 2010 to 2020 was extracted from the Surveillance, Epidemiology and End Results (SEER) database. Cox analysis was applied to identify clinicopathological characteristics associated with the overall survival (OS) of IBC patients. Random survival forest (RSF) algorithm was adopted to construct an accurate prognostic prediction model for IBC patients. Kaplan-Meier analysis was performed for survival analyses. Race, N stage, M stage, molecular subtype, history of chemotherapy and surgery, and response to neoadjuvant therapy were identified as independent predictive factors for the OS of IBC patients. The top five significant variables included surgery, response to neoadjuvant therapy, chemotherapy, breast cancer molecular subtypes, and M stage. The C-index of RSF model was 0.7704 and the area under curve (AUC) values for 1, 3, 5 years in training and validation datasets were 0.879-0.955, suggesting the excellent predictive performance of RSF model. IBC patients were divided into high-risk group and low-risk group according the risk score of RSF model, and the OS of patients in the low-risk group was significantly longer than those in the high-risk group. In this study, we constructed a prognosis prediction model for IBC patients through RSF algorithm, which may potentially serve as a useful tool during clinical decision-making.

Predicting photovoltaic greenhouse irradiance at low-latitudes of plateau based on ultra-short-term time series

Accurate and reliable ultra-short-term prediction of solar irradiance in photovoltaic (PV) greenhouses at low-latitude plateau is essential to precisely control electricity consumption of greenhouse equipment and ensure high quality crop yields. However, the irradiance in the low-latitude plateau has problems such as poor data quality, limited short-term prediction accuracy, and insufficient ability to capture nonlinear characteristics. Therefore, in order to achieve efficient utilization of photovoltaic resources, this study proposed a new hybrid integrated model TTAO-CNN-BiGRU-Attention framework to predict ultra-short-term photovoltaic greenhouse irradiance in the region. Monthly and seasonal characteristics of irradiance in low-latitude plateau areas were analyzed by statistical methods. The performance of the proposed model was verified using 9 different models for 5 different data volumes and 4 different seasons. Comprehensive analysis results show that Total radiant instantaneous(TRI) demonstrates a seasonal trend, generally low in spring, high in summer and autumn, relatively stable in autumn and winter. The monthly trend initially increases and then decreases, reaching the highest value of the year in September. The scheme proposed in this paper makes full use of the advantages of CNN, BiGRU, Attention and TTAO, greatly improving the comprehensive prediction ability of the model. In predicting different data amounts, 1 year prediction performance was the best, with RMSE, MAE, MAPE and R2 reaching 70.61 W/m2, 31 W/m2, 9.3% and 95.84%, respectively. With regard to different seasons, autumn prediction performance was the best, with RMSE, MAE, MAPE and R2 reaching 66.27 W/m2, 31.02 W/m2, 8.37% and 95.87%, respectively. The TRI prediction curve of the proposed model was closer to the actual value than other comparison models. The study found that the TTAO-CNN-BiGRU-Attention model is more accurate and stable than many traditional models in predicting ultra-short-term TRI in low-latitude plateau photovoltaic greenhouses, which can provide a reference for the comprehensive performance of PV greenhouse irradiance prediction models and precise regulation of energy supply in the future.

The role of artificial intelligence in colorectal cancer and polyp detection: A systematic review.

47 Background: Colorectal cancer (CRC) remains a leading cause of cancer-related mortality worldwide. Early detection and accurate diagnosis are essential for improving survival rates and optimizing therapeutic strategies. Recently, artificial intelligence (AI) technologies, such as machine learning algorithms, convolutional neural networks (CNNs), and computer-assisted diagnostic (CAD) systems, have significantly enhanced traditional diagnostic tools like colonoscopy and histopathology. This systematic review evaluates the current role of AI in CRC management, focusing on diagnostic improvements, predictive modeling, and outcome prediction. Methods: A comprehensive literature review was conducted using PubMed, Google Scholar, and MEDLINE using MeSH term to identify studies utilizing AI in CRC diagnosis and management, with a focus on diagnostic imaging, histopathological analysis, and predictive modeling. Studies were evaluated for the accuracy of AI-driven diagnostic systems and the predictive performance of AI models in clinical outcomes. Results: Out of 147 studies identified, 37 met the eligibility criteria for this review. AI tools such as CNNs, CAD systems, and EndoBRAIN showed high accuracy in CRC detection and polyp classification. Kudo et al. (2020) reported 98% CRC detection accuracy using EndoBRAIN. Blanes-Vidal et al. (2019) achieved 96.4% accuracy in polyp detection via DCNN. Additionally, Wang et al. (2019, 2020) demonstrated improved Adenoma Detection Rates (ADR) of 34.1% and 29.1%, respectively, using AI over traditional methods. For outcome prediction, AI models, including CNNs, predicted patient prognosis with reasonable accuracy, as highlighted by Skrede et al. (2020) with 76% accuracy in prognosis prediction. Conclusions: The systematic review and analysis of AI-assisted diagnostic modalities in colorectal cancer and polyp detection reveal promising results, with most models demonstrating high sensitivity, specificity, and accuracy. CNN-based models and CAD systems, in particular, show strong potential for improving detection rates and clinical outcomes in CRC screening. The studies included in this review demonstrate that AI can enhance diagnostic precision, particularly in identifying polyps and predicting clinical outcomes, with high validation rates. However, more external validation and long-term clinical studies are needed to fully establish the robustness of these AI tools in routine clinical practice.

ConvXGB: A novel deep learning model to predict recurrence risk of early-stage cervical cancer following surgery using multiparametric MRI images.

Accurate estimation of recurrence risk for cervical cancer plays a pivot role in making individualized treatment plans. We aimed to develop and externally validate an end-to-end deep learning model for predicting recurrence risk in cervical cancer patients following surgery by using multiparametric MRI images. The clinicopathologic data and multiparametric MRI images of 406 cervical cancer patients from three institutions were collected. We designed a novel deep learning model called "ConvXGB" for predicting recurrence risk by combining the convolutional neural network (CNN) and eXtreme Gradient Boost (XGBoost). The predictive performance of the ConvXGB model was evaluated using time-dependent area under curve (AUC), compared with the deep learning radio-clinical model, clinical model, conventional radiomics nomogram and an existing histology-specific tool. The potential of the ConvXGB model in predicting the recurrence-free survival (RFS) and overall survival (OS) was assessed. The ConvXGB model outperformed other models in predicting recurrence risk, with AUCs for 1 and 3 year-RFS of 0.872(95% CI, 0.857-0.906) and 0.882(95% CI, 0.860-0.904) respectively in the test cohort. This model showed better discrimination, calibration and clinical utility. Grad-CAM analysis was adopted to help clinicians better understand the predictive results. Moreover, Kaplan-Meier survival analysis revealed that patients who were stratified into high-risk group by the ConvXGB model were significantly susceptible to higher cumulative recurrence risk rates and worse outcome. The ConvXGB model allowed for predicting postoperative recurrence risk in cervical cancer patients and for stratifying the risk of RFS and OS.

Screening prediction models using artificial intelligence for moderate-to-severe obstructive sleep apnea in patients with acute ischemic stroke.

Obstructive sleep apnea (OSA) is common after stroke. Still, routine screening of OSA with polysomnography (PSG) is often unfeasible in clinical practice, primarily because of how limited resources are and the physical condition of patients. In this study, we used several artificial intelligence techniques to predict moderate-to-severe OSA and identify its features in patients with acute ischemic stroke. A total of 146 patients with acute ischemic stroke underwent PSG screening for OSA. Their baseline demographic characteristics, including age, sex, body mass index (BMI), Epworth Sleepiness Scale (ESS) score, and stroke risk factors, were recorded. Logistic regression analysis was conducted to identify significant features associated with moderate-to-severe OSA in patients with stroke. These significant features were used with six machine learning and ensemble learning algorithms, namely decision tree, support vector machine, random forest, extreme gradient boosting (XGBoost), adaptive boosting (AdaBoost), and gradient boosting, to compare the performance of several predictive models. Multivariate logistic regression analysis revealed that age, sex, BMI, neck circumference, and ESS score were significantly associated with the presence of moderate-to-severe OSA. According to the machine learning and ensemble learning results, the XGBoost model achieved the highest performance, with an area under the receiver operating characteristic curve of 0.89 and an accuracy and F1 score of 0.80. This study identified key factors contributing to moderate-to-severe OSA in patients with ischemic stroke. The XGBoost model exhibited high predictive performance, indicating it has potential as a supporting tool for decision-making in health-care settings.

A hybrid deep learning model based on signal decomposition and dynamic feature selection for forecasting the influent parameters of wastewater treatment plants.

Accurate prediction of influent parameters such as chemical oxygen demand (COD) and biochemical oxygen demand over five days (BOD5) is crucial for optimizing wastewater treatment processes, enhancing efficiency, and reducing costs. Traditional prediction methods struggle to capture the dynamic variations of influent parameters. Mechanistic biochemical models are unable to predict these parameters, and conventional machine learning methods show limited accuracy in forecasting key water quality indicators such as COD and BOD5. This study proposes a hybrid model that combines signal decomposition and deep learning to improve the accuracy of COD and BOD5 predictions. Additionally, a new dynamic feature selection (DFS) mechanism is introduced to optimize feature selection in real-time, reducing model redundancy and enhancing prediction stability. The model achieved R2 values of 0.88 and 0.96 for COD, and 0.75 and 0.93 for BOD5 across two wastewater treatment plants. RMSE and MAE values were significantly reduced, with decreases of 14.93% and 12.55% for COD at WWTP No. 5, and 20.89% and 20.40% for COD at WWTP No. 7. For BOD5, RMSE and MAE decreased by 3.56% and 5.28% at WWTP No. 5, and by 10.06% and 10.20% at WWTP No. 7. These results highlight the effectiveness of the proposed model and DFS mechanism in improving prediction accuracy and model performance. This approach provides valuable insights for wastewater treatment optimization and broader time series forecasting applications.

Extensive benchmarking of a method that estimates external model performance from limited statistical characteristics

Predictive model performance may deteriorate when applied to data sources that were not used for training, thus, external validation is a key step in successful model deployment. As access to patient-level external data sources is typically limited, we recently proposed a method that estimates external model performance using only external summary statistics. Here, we benchmark the proposed method on multiple tasks using five large heterogeneous US data sources, where each, in turn, plays the role of an internal source and the remaining—external. Results showed accurate estimations for all metrics: 95th error percentiles for the area under the receiver operating characteristics (discrimination), calibration-in-the-large (calibration), Brier and scaled Brier scores (overall accuracy) of 0.03, 0.08, 0.0002, and 0.07, respectively. These results demonstrate the feasibility of estimating the transportability of prediction models using an internal cohort and external statistics. It may become an important accelerator of model deployment.

The added value of ECG abnormalities in predicting incident cardiovascular disease risk for people with type 2 diabetes: The Hoorn Diabetes Care System cohort.

To investigate if adding ECG abnormalities as a predictor improves the performance of incident CVD-risk prediction models for people with type 2 diabetes (T2D). We evaluated the four major prediction models that are recommended by the guidelines of the American College of Cardiology/American Heart Association and European Society of Cardiology, in 11,224 people with T2D without CVD (coronary heart disease, heart failure, stroke, thrombosis) from the Hoorn Diabetes Care System cohort (1998-2018). Baseline measurements included CVD-risk factors and ECG recordings coded according to the Minnesota Classification as no, minor or major abnormalities. After confirming good reference model fit, model performance was assessed before and after addition of ECG abnormalities and compared using c-statistics, net reclassification improvement (NRI) and integrated discrimination improvement (IDI). C-statistics (95%CI) of reference models (ASCVD, AD-ON, ADVANCE and SCORE2-Diabetes) were 0.67 (0.65-0.70), 0.73 (0.71-0.76), 0.71 (0.68-0.74) and 0.67 (0.65-0.69), respectively. Adding ECG abnormalities as predictor improved c-statistics with +0.02 (0.01-0.03), +0.01 (0.00-0.01), +0.02 (0.01-0.03), and +0.02 (0.01-0.02), respectively. Reclassification indicators also showed improvement: categorical NRI (+6%, +3%, +8%, and +5%), continuous NRI (95%CI) 0.25 (0.08-0.37), 0.32 (0.23-0.42), 0.54 (0.34-0.69) and 0.28 (0.09-0.33)), and IDI (95%CI) 0.005 (0.001-0.010), 0.002 (-0.001-0.007), 0.006 (0.001-0.007) and 0.004 (0.000-0.006)). Sensitivity analyses yielded similar results. The addition of ECG abnormalities to incident CVD-risk prediction models moderately but consistently improves the ability of models to correctly classify people with T2D in the appropriate CVD-risk category with up to 8%, which is approximately equivalent to many established predictors and (bio)markers.

Prediction of hip fracture by high-resolution peripheral quantitative computed tomography in older Swedish women.

The socioeconomic burden of hip fractures, the most severe osteoporotic fracture outcome, is increasing and the current clinical risk assessment lacks sensitivity. This study aimed to develop a method for improved prediction of hip fracture by incorporating measurements of bone microstructure and composition derived from high-resolution peripheral quantitative computed tomography (HR-pQCT). In a prospective cohort study of 3028 community-dwelling women aged 75 to 80, all participants answered questionnaires and underwent baseline examinations of anthropometrics and bone by dual x-ray absorptiometry (DXA) and HR-pQCT. Medical records, a regional x-ray archive, and registers were used to identify incident fractures and death. Prediction models for hip, major osteoporotic fracture (MOF), and any fracture were developed using Cox proportional hazards regression and machine learning algorithms (neural network, random forest, ensemble, and XGBoost). In the 2856 (94.3%) women with complete HR-pQCT data at 2 tibia sites (distal and ultra-distal), the median follow-up period was 8.0years, and 217 hip, 746 MOF, and 1008 any type of incident fracture occurred. In Cox regression models adjusted for age, BMI, clinical risk factors (CRF), and femoral neck bone mineral density (FN BMD) the strongest predictors of hip fracture were tibia total volumetric BMD and cortical thickness. The performance of the Cox regression-based prediction models for hip fracture was significantly improved by HR-pQCT (time-dependent AUC; area under receiver operating characteristic curve at 5years of follow-up 0.75 [0.64-0.85]), compared to a reference model including CRFs and FN BMD (AUC = 0.71 [0.58-0.81], p<.001) and a FRAX risk score model (AUC = 0.70 [0.60-0.80], p<.001). The Cox regression model for hip fracture had a significantly higher accuracy than the neural network-based model, the best-performing machine learning algorithm, at clinically relevant sensitivity levels. We conclude that the addition of HR-pQCT parameters improves the prediction of hip fractures in a cohort of older Swedish women.

The Harms of Class Imbalance Corrections for Machine Learning Based Prediction Models: A Simulation Study.

Risk prediction models are increasingly used in healthcare to aid in clinical decision-making. In most clinical contexts, model calibration (i.e., assessing the reliability of risk estimates) is critical. Data available for model development are often not perfectly balanced with the modeled outcome (i.e., individuals with vs. without the event of interest are not equally prevalentin the data). It is common for researchers to correct for class imbalance, yet, the effect of such imbalance corrections on the calibration of machine learning models is largely unknown. We studied the effect of imbalance corrections on model calibration for a variety of machine learning algorithms. Using extensive Monte Carlo simulations we compared the out-of-sample predictive performance of models developed with an imbalance correction to those developed without a correction for class imbalance across different data-generating scenarios (varying sample size, the number of predictors, and event fraction). Our findings were illustrated in a case study using MIMIC-III data. In all simulation scenarios, prediction models developed without a correction for class imbalance consistently had equal or better calibration performance than prediction models developed with a correction for class imbalance. The miscalibration introduced by correcting for class imbalance was characterized by an over-estimation of risk and was not always able to be corrected with re-calibration. Correcting for class imbalance is not always necessary and may even be harmful to clinical prediction models which aim to produce reliable risk estimates on an individual basis.

Evaluating stress relaxation and creep-fatigue interactions in Alloy 617

ABSTRACT This paper investigates the creep-fatigue interactions of Alloy 617 for its exceptional high-temperature mechanical and physical resistance. We focus on modelling the stress relaxation curves under varying total strain ranges, with the primary goal of developing a comprehensive model to verify relaxed stress over extended hold times, up to 5,000 hours. The predictive performance of models fitted to short-term data is compared with long-term data, particularly in the context of the linear damage summation (LDS) model. Additionally, we analyse the fractured surfaces in relation to the creep-fatigue damage mechanism. Our findings identify three leading models – Jeong’s formula, the Power-law fit, and Peleg’s model – as effective representations of the stress relaxation behaviour in Alloy 617. The Peleg’s model is utilised for parameter extrapolation and offers significant predictive accuracy. We also propose an improved LDS diagram for Alloy 617, integrating our findings to enhance the understanding of creep-fatigue interactions and damage prediction accuracy.

Association between triglyceride glucose index and adverse cardiovascular prognosis in patients with atrial fibrillation without diabetes: a retrospective cohort study

BackgroundAtrial fibrillation (AF) is the most prevalent arrhythmia encountered in clinical practice. Triglyceride glucose index (Tyg), a convenient evaluation variable for insulin resistance, has shown associations with adverse cardiovascular outcomes. However, studies on the Tyg index’s predictive value for adverse prognosis in patients with AF without diabetes are lacking.MethodsThis retrospective study utilized electronic medical records to collect data on patients with AF hospitalized at West China Hospital from January to June 2020. Participants were categorized into three groups based on their Tyg index levels. The primary outcome, major adverse cardiovascular events, included cardiac death, stroke, and myocardial infarction. Kaplan–Meier curve, Cox proportional hazards regression model, and restricted cubic spline were employed to explore the relationship between the Tyg index and outcomes. The predictive performance of the CHA2DS2-VASc model was evaluated after incorporating the Tyg index.ResultsThe study comprised 864 participants (mean age 67.69 years, 55.32% male, 57.52% paroxysmal AF). Patients with high Tyg index had a significantly higher risk of developing major adverse cardiovascular events (MACE) (P < 0.001, hazard ratio: 2.05, 95% confidence interval:1.65–2.56). The MACE risk in the middle Tyg group was similar to that in the low Tyg group (P = 0.1) during the 48-month follow-up period. However, focusing on the last 24 months revealed a higher MACE risk (P = 0.015) in the middle Tyg group. The restricted cubic spline analysis revealed an S-shaped correlation between Tyg and MACE. The CHA2DS2-VASc model combined with the Tyg index showed improved predictive performance and net benefit.ConclusionsA high Tyg index is associated with poorer prognosis in patients with AF without diabetes. Integrating the Tyg index into the CHA2DS2-VASc model may enhance its predictive performance, offering clinical utility.

Performance of the FORD Versus Other Available Models for the Noninvasive Prediction of Pulmonary Hypertension in Patients with Interstitial Lung Disease

PurposePulmonary hypertension (PH) is associated with morbidity and mortality in patients with interstitial lung disease (ILD). Several prediction models have been proposed to predict PH in ILD patients. We sought to discern how previously described prediction models perform in predicting PH in patients with ILD.MethodsPatients with ILD who completed a baseline right heart catheterization, from Inova Fairfax Hospital, Northwestern Memorial Hospital, and Asan Medical Center in Korea were enrolled. The performance of various prediction models (FORD model, the FORD calculator, the PH‐ILD Detection tool, and the mean pulmonary artery pressure prediction model) were assessed using receiver operating characteristic (ROC) curves and area under the receiver operating characteristic curve (AUROC). There were four definitions of pulmonary hypertension against which the models were evaluated.ResultsThere were a total of 192 patients with ILD, of whom 32.8% (n = 63/192) met the modified 5th world symposium on PH definition of precapillary PH. Among the models assessed, the FORD calculator had an AUROC (0.733) that was marginally highest. Subgroup analysis revealed that the FORD index had the highest AUROC (0.817) in patients with idiopathic pulmonary fibrosis, while the FORD calculator had the highest AUROC (0.751) in patients with non-IPF ILD.ConclusionThe FORD model can be used to predict group 3 PH in both IPF patients and non-IPF ILD patients. It could serve as a tool for ILD patient selection for right heart catheterization as well as an enrichment tool for clinical trials targeting the pulmonary vasculature.

Managing Pregnancies of Unknown Location With the M4 Prediction Model or the NICE Algorithm: A Randomised Controlled Trial With Cross-Sectional Diagnostic Accuracy Data.

To determine the diagnostic performance and clinical utility of the M4 prediction model and the NICE algorithm managing women with pregnancy of unknown location (PUL). The study has a superiority design regarding specificity for non-ectopic pregnancy for M4, given that the primary outcome of sensitivity for ectopic pregnancy (EP) is non-inferior in comparison with the NICE algorithm. Emergency gynaecology units in Sweden. 595 women with PUL. Participants were randomised (1:1) to M4 or the NICE algorithm after two serum human chorionic (hCG) levels and were categorised as high or low risk of having an EP. The diagnostic performance was evaluated on cross-sectional data and utility by parallel groups. The proportion of EP categorised as high risk (sensitivity) and non-ectopic pregnancies categorised as low risk (specificity). Clinical outcomes were assessed. The sensitivity for EP was 79% (115 of 146) for M4 versus 85% (124 of 146) for the NICE algorithm, p = 0.1496 and the specificity for non-ectopic pregnancies was 67% (300 of 449) for M4 and 74% (334 of 449) for the NICE algorithm, p = 0.0003. Clinical outcomes were similar between groups. The sensitivity for EP by M4 was non-inferior to NICE, but specificity was better for the NICE algorithm. No between group differences were observed for clinical outcomes. NCT03461835, https://www. gov.

Clinical Applications of Artificial Intelligence (AI) in Human Cancer: Is It Time to Update the Diagnostic and Predictive Models in Managing Hepatocellular Carcinoma (HCC)?

In recent years, novel findings have progressively and promisingly supported the potential role of Artificial intelligence (AI) in transforming the management of various neoplasms, including hepatocellular carcinoma (HCC). HCC represents the most common primary liver cancer. Alarmingly, the HCC incidence is dramatically increasing worldwide due to the simultaneous “pandemic” spreading of metabolic dysfunction-associated steatotic liver disease (MASLD). MASLD currently constitutes the leading cause of chronic hepatic damage (steatosis and steatohepatitis), fibrosis, and liver cirrhosis, configuring a scenario where an HCC onset has been reported even in the early disease stage. On the other hand, HCC represents a serious plague, significantly burdening the outcomes of chronic hepatitis B (HBV) and hepatitis C (HCV) virus-infected patients. Despite the recent progress in the management of this cancer, the overall prognosis for advanced-stage HCC patients continues to be poor, suggesting the absolute need to develop personalized healthcare strategies further. In this “cold war”, machine learning techniques and neural networks are emerging as weapons, able to identify the patterns and biomarkers that would have normally escaped human observation. Using advanced algorithms, AI can analyze large volumes of clinical data and medical images (including routinely obtained ultrasound data) with an elevated accuracy, facilitating early diagnosis, improving the performance of predictive models, and supporting the multidisciplinary (oncologist, gastroenterologist, surgeon, radiologist) team in opting for the best “tailored” individual treatment. Additionally, AI can significantly contribute to enhancing the effectiveness of metabolomics–radiomics-based models, promoting the identification of specific HCC-pathogenetic molecules as new targets for realizing novel therapeutic regimens. In the era of precision medicine, integrating AI into routine clinical practice appears as a promising frontier, opening new avenues for liver cancer research and treatment.

P0024 Circadian regulator-mediated molecular subtypes depict the features of elderly ulcerative colitis

Abstract Background The prevalence of UC in older individuals (i.e., aged &amp;gt; 60 years) is increasing globally because of an ageing population and a subset of population with elderly-onset (i.e., onset aged ≥ 60 years) UC, which causes high burden on healthcare. Circadian rhythms are 24-h oscillations that control various biological processes in the living system and have been recognized to be associated with the development of autoimmune diseases and aging. However, the implication of circadian rhythm in elderly UC remains largely unknown. Methods We downloaded datasets from the Gene Expression Omnibus (GEO) database. The differential expression of circadian regulators between early adult (15-25 years old) and elderly (≥ 60 years old) UC patients was identified. We constructed and compared random forest (RF) and support vector machine (SVM) models. Feature genes were selected to develop a nomogram model with the superior model. The receiver operator characteristic (ROC) curves were used to validate the performances of predictive models and signature genes. The infiltration of 22 kinds of immune cells in elderly UC was analyzed by ssGSEA, and the relationship between feature circadian genes and the proportion of immune infiltration was studied. We identified circadian subtypes based on significantly differentially expressed circadian regulatory genes. Comprehensive evaluation of the two circadian patterns was performed, including immune characteristics and potential functions. Results The expression levels of circadian regulatory genes were extracted and a total of 44 differentially expressed circadian regulators were identified between early adult and elderly UC patients. Five feature circadian regulatory genes, AGT, EGF, NR3C2, PPY, and PYY were determined (p &amp;lt; 0.05) to establish a nomogram model that can predict the incidence of elderly UC with the RF model. The signature circadian regulators were found to be correlated with activated memory CD4+ T cells, gamma delta T cells, and macrophages. We identified two circadian subtypes based on the five significant circadian regulatory genes. Cluster A had a higher immune infiltration of activated CD4+T cells, CD8+ T cells and B cells than cluster B (p &amp;lt; 0.01). The signature circadian genes can also be used to predict the risk of elderly UC and have good diagnostic efficacy. Conclusion The circadian regulatory genes play non-negligible roles in elderly UC. A nomogram model developed by five feature circadian regulatory genes could be used to predict the incidence of elderly UC. Two circadian patterns were identified and evaluated comprehensively, which may provide insights into the classification of elderly UC patients and guide treatment.

Plasma SVEP1 Levels Predict Cardiovascular Events in Hypertrophic Cardiomyopathy Beyond Conventional Clinical Risk Models Including NT-proBNP.

Hypertrophic cardiomyopathy is the most common genetic cardiomyopathy and causes major adverse cardiovascular events (MACE). SVEP1 (Sushi, von Willebrand factor type A, epidermal growth factor, and pentraxin domain containing 1) is a large extracellular matrix protein that is detectable in the plasma. However, it is unknown whether adding plasma SVEP1 levels to clinical predictors including NT-proBNP (N-terminal pro-B-type natriuretic peptide) improves the prognostication in patients with hypertrophic cardiomyopathy. We performed a multicenter prospective cohort study of 610 patients with hypertrophic cardiomyopathy. The outcome was MACE defined as heart failure hospitalization or cardiac death. In 4 groups stratified by the median levels of SVEP1 and NT-proBNP, we compared the risk of MACE using the Cox proportional hazards model adjusting for 15 clinical predictors. We also developed a Lasso-regularized Cox proportional hazards model to predict time to first MACE by adding SVEP1 to the 15 clinical predictors with or without NT-proBNP and compared the predictive performance based on C statistics using 10-fold cross-validation. Even in the low NT-proBNP groups, the high SVEP1 group had higher risks of MACE compared with the low SVEP1 group (adjusted hazard ratio, 4.52 [95% CI, 1.05-19.4]; P=0.042). In predicting time to first MACE, the addition of SVEP1 improved the C statistics of the clinical plus NT-proBNP model (0.87 [0.83-0.91] versus 0.82 [0.78-0.86]; P=0.01). The clinical plus SVEP1 model also outperformed the clinical plus NT-proBNP model (0.86 [0.82-0.91] versus 0.82 [0.78-0.86]; P=0.04). SVEP1 improved the predictive performance of conventional models, including known clinical parameters with or without NT-proBNP, to predict future MACE in patients with hypertrophic cardiomyopathy.

Comparing innovative AI algorithms to assess echocardiographic videos for clinical modelling.

We conduct a comparative study that employs the use of multiple dynamic deep learning algorithms to develop predictive models with video-based echocardiographic images using sample size determination as a key variable to assess optimal performance metrics. Our study compares performance of 3D convolutional neural networks, video vision transformers, and hybrid convolutional neural networks and Long Short-Term Memory models within both supervised and semi-supervised domains using variable sample sizes. For supervised learning, the ResNet3D model achieved the lowest Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) across all training set sizes (200, 400, and 800-video datasets), with the best performance observed on the 800-video training set (MAE: 7.409, RMSE: 10.216). In the semi-supervised setting, both ResNet3D and ResNet+LSTM models benefited from the inclusion of unlabelled data, particularly at larger dataset sizes. Because semi-supervised models use both labelled and unlabelled data sets, our findings are significant in showing that performance of certain predictive models using mixtures of unlabelled and labelled data is comparable to those using only labelled data with similar sample sizes, thus obviating the need for large sample sizes of labelled data.

Mitigating bias in AI mortality predictions for minority populations: a transfer learning approach

BackgroundThe COVID-19 pandemic has highlighted the crucial role of artificial intelligence (AI) in predicting mortality and guiding healthcare decisions. However, AI models may perpetuate or exacerbate existing health disparities due to demographic biases, particularly affecting racial and ethnic minorities. The objective of this study is to investigate the demographic biases in AI models predicting COVID-19 mortality and to assess the effectiveness of transfer learning in improving model fairness across diverse demographic groups.MethodsThis retrospective cohort study used a population-based dataset of COVID-19 cases from the Centers for Disease Control and Prevention (CDC), spanning the years 2020–2024. The study analyzed AI model performance across different racial and ethnic groups and employed transfer learning techniques to improve model fairness by adapting pre-trained models to the specific demographic and clinical characteristics of the population.ResultsDecision Tree (DT) and Random Forest (RF) models consistently showed improvements in accuracy, precision, and ROC-AUC scores for Non-Hispanic Black, Hispanic/Latino, and Asian populations. The most significant precision improvement was observed in the DT model for Hispanic/Latino individuals, which increased from 0.3805 to 0.5265. The precision for Asians or Pacific Islanders in the DT model increased from 0.4727 to 0.6071, and for Non-Hispanic Blacks, it rose from 0.5492 to 0.6657. Gradient Boosting Machines (GBM) produced mixed results, showing accuracy and precision improvements for Non-Hispanic Black and Asian groups, but declines for the Hispanic/Latino and American Indian groups, with the most significant decline in precision, which dropped from 0.4612 to 0.2406 in the American Indian group. Logistic Regression (LR) demonstrated minimal changes across all metrics and groups. For the Non-Hispanic American Indian group, most models showed limited benefits, with several performance metrics either remaining stable or declining.ConclusionsThis study demonstrates the potential of AI in predicting COVID-19 mortality while also underscoring the critical need to address demographic biases. The application of transfer learning significantly improved the predictive performance of models across various racial and ethnic groups, suggesting these techniques are effective in mitigating biases and promoting fairness in AI models.