Predictive value of machine learning for in-hospital mortality risk in acute myocardial infarction: A systematic review and meta-analysis

International Journal of Medical Informatics, Год журнала: 2025, Номер 198, С. 105875 - 105875

Опубликована: Март 8, 2025

Язык: Английский

---

A Systematic Review of Artificial Intelligence and Machine Learning Applications to Inflammatory Bowel Disease, with Practical Guidelines for Interpretation

Inflammatory Bowel Diseases, Год журнала: 2022, Номер 28(10), С. 1573 - 1583

Опубликована: Июнь 14, 2022

Abstract Background Inflammatory bowel disease (IBD) is a gastrointestinal chronic with an unpredictable course. Computational methods such as machine learning (ML) have the potential to stratify IBD patients for provision of individualized care. The use ML was surveyed, additional focus on how field has changed over time. Methods On May 6, 2021, systematic review conducted through search MEDLINE and Embase databases, structure (“machine learning” OR “artificial intelligence”) AND (“Crohn* Disease” “Ulcerative Colitis” “Inflammatory Bowel Disease”). Exclusion criteria included studies not written in English, no human patient data, publication before 2001, that were peer reviewed, nonautoimmune comorbidity research, record types primary research. Results Seventy-eight (of 409) records met inclusion criteria. Random forest most prevalent, there increase neural networks, mainly applied imaging data sets. main applications clinical tasks diagnosis (18 78), course (22 severity (16 78). median sample size 263. Clinical microbiome-related sets popular. Five percent used external set after training testing model validation. Discussion Availability longitudinal deep phenotyping could lead better modeling. Machine pipelines consider imbalanced feature selection only will generate more generalizable models. models are increasingly being complex specific phenotypes, indicating progress towards personalized medicine IBD.

Язык: Английский

---

Short-term atrial fibrillation detection using electrocardiograms: A comparison of machine learning approaches

International Journal of Medical Informatics, Год журнала: 2022, Номер 163, С. 104790 - 104790

Опубликована: Май 7, 2022

Atrial fibrillation (AF) is one of the most prevalent cardiac arrhythmias, which challenges healthcare systems globally.Timely detection AF can potentially reduce mortality and morbidity rates as well alleviate economic burden caused by this.Digital solutions are shown to enhance diagnosis abnormalities.By latest advancements in field medical informatics tele-health monitoring, huge amount electro-physiological signals, such electrocardiograms (ECG), be easily collected.One common ways for physicians/cardiologists analyse these signals through visual inspection.However, it not always easy cases cumbersome big amounts ECG data.Therefore, great interest develop models that capable analyzing data help physicians making better decisions.This paper proposes compares well-known machine learning (ML) algorithms diagnose short episodes AF. This also paves way real-time clinical settings.Different ML Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), Stacking Classifier (SC), Extreme Gradient Boosting (XGBoost), Adaptive (AdaBoost) were applied detect These trained using extracted statistical features from signals.The proposed on a dataset with 23 records length approximately 10 h each leave group out cross validation (LOGO-CV) technique achieved best sensitivity (Se), specificity (Sp), positive predictive value (PPV), false rate (FPR), F1-score 85.67%, 81.25%, 90.85%, 18.75% 88.18%, respectively, classify normal sinus rhythms (NSR) segments 20 heartbeats.Additionally, examined three unseen datasets, namely Long Term dataset, MIT-BIH Arrhythmia Normal Sinus Rhythm assess their robustness generalization.The obtained results show high performance flexibility some compared other algorithms. In general, empirical confirm ensemble methods, AdaBoost, generalized perform than approaches.

Язык: Английский

---

Data-centric artificial intelligence in oncology: a systematic review assessing data quality in machine learning models for head and neck cancer

Journal Of Big Data, Год журнала: 2023, Номер 10(1)

Опубликована: Март 4, 2023

Abstract Machine learning models have been increasingly considered to model head and neck cancer outcomes for improved screening, diagnosis, treatment, prognostication of the disease. As concept data-centric artificial intelligence is still incipient in healthcare systems, little known about data quality proposed clinical utility. This important as it supports generalizability standardization. Therefore, this study overviews structured unstructured used machine construction cancer. Relevant studies reporting on use based custom datasets between January 2016 June 2022 were sourced from PubMed, EMBASE, Scopus, Web Science electronic databases. Prediction Risk Bias Assessment (PROBAST) tool was assess individual before comprehensive parameters assessed according type dataset construction. A total 159 included review; 106 utilized while 53 datasets. Data assessments deliberately performed 14.2% 11.3% Class imbalance fairness most common limitations both types outlier detection lack representative outcome classes respectively. Furthermore, review found that class reduced discriminatory performance higher image resolution good overlap resulted better using during internal validation. Overall, infrequently ML irrespective or To improve generalizability, discussed should be introduced achieve intelligent systems management.

Язык: Английский

---

Prediction of emergency department revisits among child and youth mental health outpatients using deep learning techniques

BMC Medical Informatics and Decision Making, Год журнала: 2024, Номер 24(1)

Опубликована: Фев. 8, 2024

Abstract Background The proportion of Canadian youth seeking mental health support from an emergency department (ED) has risen in recent years. As EDs typically address urgent crises, revisiting ED may represent unmet needs. Accurate revisit prediction could aid early intervention and ensure efficient healthcare resource allocation. We examine the potential increased accuracy performance graph neural network (GNN) machine learning models compared to recurrent (RNN), baseline conventional regression for predicting electronic record (EHR) data. Methods This study used EHR data children aged 4–17 services at McMaster Children’s Hospital’s Child Youth Mental Health Program outpatient service develop evaluate GNN RNN predict whether a child/youth with visit had within 30 days. were developed against models. Model GNN, RNN, XGBoost, decision tree logistic was evaluated using F1 scores. Results model outperformed by F1-score increase 0.0511 best performing 0.0470. Precision, recall, receiver operating characteristic (ROC) curves, positive negative predictive values showed that performed best, similarly XGBoost model. Performance increases most noticeable recall value than precision value. Conclusions demonstrates improved utility revisits among youth, although not be sufficient clinical implementation. Given improvements value, should further explored algorithms can inform decision-making ways facilitate targeted interventions, optimize allocation, improve outcomes youth.

Язык: Английский

---