Development and External Validation of [18F]FDG PET-CT-Derived Radiomic Models for Prediction of Abdominal Aortic Aneurysm Growth Rate DOI Creative Commons

Simran Singh Dhesi,

Pratik Adusumilli, Nishant Ravikumar

et al.

Algorithms, Journal Year: 2025, Volume and Issue: 18(2), P. 86 - 86

Published: Feb. 5, 2025

Objective (1): To develop and validate a machine learning (ML) model using radiomic features (RFs) extracted from [18F]FDG PET-CT to predict abdominal aortic aneurysm (AAA) growth rate. Methods (2): This retrospective study included 98 internal 55 external AAA patients undergoing PET-CT. RFs were manual segmentations of AAAs PyRadiomics. Recursive feature elimination (RFE) reduced for optimisation. A multi-layer perceptron (MLP) was developed prediction compared against Random Forest (RF), XGBoost, Support Vector Machine (SVM). Accuracy evaluated via cross-validation, with uncertainty quantified dropout (MLP), standard deviation 95% intervals (XGBoost). External validation used independent data two centres. Ground truth rates calculated serial ultrasound (US) measurements or CT volumes. Results (3): From 93 initial RFs, 29 remained after RFE. The MLP achieved an MAE ± SEM 1.35 3.2e−4 mm/year the full set 2.5e−4 yielded 1.8 8.9e−8 mm/year. RF, SVM models produced comparable accuracies internally (1.4–1.5 mm/year) but showed higher errors during (1.9–1.97 mm/year). demonstrated across all datasets. Conclusions (4): An leveraging radiomics accurately predicted generalised well data. In future, more sophisticated stratification could guide individualised patient care, facilitating risk-tailored management AAAs.

Language: Английский

TRIPOD+AI statement: updated guidance for reporting clinical prediction models that use regression or machine learning methods DOI Creative Commons
Gary S. Collins, Karel G.M. Moons, Paula Dhiman

et al.

BMJ, Journal Year: 2024, Volume and Issue: unknown, P. e078378 - e078378

Published: April 16, 2024

The TRIPOD (Transparent Reporting of a multivariable prediction model for Individual Prognosis Or Diagnosis) statement was published in 2015 to provide the minimum reporting recommendations studies developing or evaluating performance model. Methodological advances field have since included widespread use artificial intelligence (AI) powered by machine learning methods develop models. An update is thus needed. TRIPOD+AI provides harmonised guidance studies, irrespective whether regression modelling been used. new checklist supersedes checklist, which should no longer be This article describes development and presents expanded 27 item with more detailed explanation each recommendation, Abstracts checklist. aims promote complete, accurate, transparent that evaluate its performance. Complete will facilitate study appraisal, evaluation, implementation.

Language: Английский

Citations

375
Evaluation of clinical prediction models (part 2): how to undertake an external validation study DOI Creative Commons
Richard D. Riley, Lucinda Archer, Kym I E Snell

et al.

BMJ, Journal Year: 2024, Volume and Issue: unknown, P. e074820 - e074820

Published: Jan. 15, 2024

External validation studies are an important but often neglected part of prediction model research. In this article, the second in a series on evaluation, Riley and colleagues explain what external study entails describe key steps involved, from establishing high quality dataset to evaluating model's predictive performance clinical usefulness.

Language: Английский

Citations

94
Evaluation of clinical prediction models (part 3): calculating the sample size required for an external validation study DOI Creative Commons
Richard D. Riley, Kym I E Snell, Lucinda Archer

et al.

BMJ, Journal Year: 2024, Volume and Issue: unknown, P. e074821 - e074821

Published: Jan. 22, 2024

An external validation study evaluates the performance of a prediction model in new data, but many these studies are too small to provide reliable answers. In third article their series on evaluation, Riley and colleagues describe how calculate sample size required for studies, propose avoid rules thumb by tailoring calculations setting at hand.

Language: Английский

Citations

55
Developing clinical prediction models: a step-by-step guide DOI Creative Commons
Orestis Efthimiou, Michael Seo, Konstantina Chalkou

et al.

BMJ, Journal Year: 2024, Volume and Issue: unknown, P. e078276 - e078276

Published: Sept. 3, 2024

Predicting future outcomes of patients is essential to clinical practice, with many prediction models published each year. Empirical evidence suggests that studies often have severe methodological limitations, which undermine their usefulness. This article presents a step-by-step guide help researchers develop and evaluate model. The covers best practices in defining the aim users, selecting data sources, addressing missing data, exploring alternative modelling options, assessing model performance. steps are illustrated using an example from relapsing-remitting multiple sclerosis. Comprehensive R code also provided.

Language: Английский

Citations

47
Uncertainty of risk estimates from clinical prediction models: rationale, challenges, and approaches DOI Creative Commons
Richard D. Riley, Gary S. Collins, Laura Kirton

et al.

BMJ, Journal Year: 2025, Volume and Issue: unknown, P. e080749 - e080749

Published: Feb. 13, 2025

Language: Английский

Citations

7
External validation of AI-based scoring systems in the ICU: a systematic review and meta-analysis DOI Creative Commons
Patrick Rockenschaub, Ela M. Akay, Benjamin Gregory Carlisle

et al.

BMC Medical Informatics and Decision Making, Journal Year: 2025, Volume and Issue: 25(1)

Published: Jan. 6, 2025

Abstract Background Machine learning (ML) is increasingly used to predict clinical deterioration in intensive care unit (ICU) patients through scoring systems. Although promising, such algorithms often overfit their training cohort and perform worse at new hospitals. Thus, external validation a critical – but frequently overlooked step establish the reliability of predicted risk scores translate them into practice. We systematically reviewed how regularly ML-based performed performance changed data. Methods searched MEDLINE, Web Science, arXiv for studies using ML ICU from routine included primary research published English before December 2023. summarised many were externally validated, assessing differences over time, by outcome, data source. For validated studies, we evaluated change area under receiver operating characteristic (AUROC) attributable linear mixed-effects models. Results 572 which 84 (14.7%) increasing 23.9% Validated made disproportionate use open-source data, with two well-known US datasets (MIMIC eICU) accounting 83.3% studies. On average, AUROC was reduced -0.037 (95% CI -0.052 -0.027) more than 0.05 reduction 49.5% Discussion External validation, although increasing, remains uncommon. Performance generally lower questioning some recently proposed scores. Interpretation results challenged an overreliance on same few datasets, implicit case mix, exclusive AUROC.

Language: Английский

Citations

4
Explainable artificial intelligence to identify follicles that optimize clinical outcomes during assisted conception DOI Creative Commons
Simon Hanassab, Scott M. Nelson,

Artur Akbarov

et al.

Nature Communications, Journal Year: 2025, Volume and Issue: 16(1)

Published: Jan. 8, 2025

Abstract Infertility affects one-in-six couples, often necessitating in vitro fertilization treatment (IVF). IVF generates complex data, which can challenge the utilization of full richness data during decision-making, leading to reliance on simple ‘rules-of-thumb’. Machine learning techniques are well-suited analyzing provide data-driven recommendations improve decision-making. In this multi-center study ( n = 19,082 treatment-naive female patients), including 11 European centers, we harnessed explainable artificial intelligence identify follicle sizes that contribute most relevant downstream clinical outcomes. We found intermediately-sized follicles were important number mature oocytes subsequently retrieved. Maximizing proportion by end ovarian stimulation was associated with improved live birth rates. Our suggests larger mean sizes, especially those >18 mm, premature progesterone elevation and a negative impact rates fresh embryo transfer. These highlight potential computer technologies aid personalization optimize outcomes pending future prospective validation.

Language: Английский

Citations

4
PROBAST+AI: an updated quality, risk of bias, and applicability assessment tool for prediction models using regression or artificial intelligence methods DOI Creative Commons
Karel G.M. Moons, Johanna AAG Damen, T. K. Kaul

et al.

BMJ, Journal Year: 2025, Volume and Issue: unknown, P. e082505 - e082505

Published: March 24, 2025

The Prediction model Risk Of Bias ASsessment Tool (PROBAST) is used to assess the quality, risk of bias, and applicability prediction models or algorithms model/algorithm studies. Since PROBAST's introduction in 2019, much progress has been made methodology for modelling use artificial intelligence, including machine learning, techniques. An update PROBAST-2019 thus needed. This article describes development PROBAST+AI. PROBAST+AI consists two distinctive parts: evaluation. For development, users quality using 16 targeted signalling questions. evaluation, bias 18 Both parts contain four domains: participants data sources, predictors, outcome, analysis. Applicability rated outcome domains. may replace original PROBAST tool allows all key stakeholders (eg, developers, AI companies, researchers, editors, reviewers, healthcare professionals, guideline policy organisations) examine any type sector, irrespective whether regression techniques are used.

Language: Английский

Citations

4
Temporal validation of machine learning models for pre-eclampsia prediction using routinely collected maternal characteristics: A validation study DOI Creative Commons
Sofonyas Abebaw Tiruneh, Daniel L. Rolnik, Helena Teede

et al.

Computers in Biology and Medicine, Journal Year: 2025, Volume and Issue: 191, P. 110183 - 110183

Published: April 14, 2025

Pre-eclampsia (PE) contributes to more than one-fourth of all maternal deaths and half a million newborn worldwide every year. Early screening interventions can reduce PE incidence related complications. We aim 1) temporally validate three existing models (two machine learning (ML) one logistic regression) developed in the same region 2) compare performances validated ML with regression model prediction. This work addresses gap literature by undertaking comprehensive evaluation risk prediction models, which is an important step advancing this field. obtained dataset routinely collected antenatal data from maternity hospitals South-East Melbourne, Australia, extracted between July 2021 December 2022. models: extreme gradient boosting (XGBoost, 'model 1'), random forest ('model 2') 3'). Area under receiver-operating characteristic (ROC) curve (AUC) was evaluated discrimination performance, calibration assessed. The AUCs were compared using 'bootstrapping' test. temporal consisted 12,549 singleton pregnancies, 431 (3.43 %, 95 % confidence interval (CI) 3.13-3.77) PE. characteristics similar original development dataset. XGBoost 1' 3' exhibited performance AUC 0.75 (95 CI 0.73-0.78) 0.76 0.74-0.78), respectively. 2' showed 0.71 0.69-0.74). Model 3 perfect slope 1.02 0.92-1.12). Models 1 2 1.15 1.03-1.28) 0.62 0.54-0.70), Compared stable whereas significantly lower performance. better clinical net benefits 22 threshold probabilities default strategies. During validation performance; however, both did not outperform model. To facilitate insights into interpretability deployment, could be integrated routine practice as first-step two-stage approach identify higher-risk woman for further second stage accurate

Language: Английский

Citations

3
Risk estimation for the primary prevention of cardiovascular disease: considerations for appropriate risk prediction model selection DOI Creative Commons
Kim Robin van Daalen, Dudan Zhang, Stephen Kaptoge

et al.

The Lancet Global Health, Journal Year: 2024, Volume and Issue: 12(8), P. e1343 - e1358

Published: July 17, 2024

Cardiovascular diseases remain the number one cause of death globally. disease risk scores are an integral tool in primary prevention, being used to identify individuals at highest and guide assignment preventive interventions. Available differ substantially terms population sample data sources for their derivation and, consequently, absolute risks they assign individuals. Differences cardiovascular epidemiology between populations contributing development scores, target which applied, can result overestimation or underestimation individuals, poorly informed clinical decisions. Given wide plethora available, identification appropriate score a be challenging. This Review provides up-to-date overview guideline-recommended from global, regional, national contexts, evaluates comparative characteristics qualities, guidance on selection score.

Language: Английский

Citations

15