Machine Learning–Based Prediction of Suicidal Thinking in Adolescents by Derivation and Validation in 3 Independent Worldwide Cohorts: Algorithm Development and Validation Study (Preprint) DOI Creative Commons
Hyejun Kim, Yejun Son, Hojae Lee

et al.

Published: Dec. 29, 2023

BACKGROUND Suicide is the second-leading cause of death among adolescents and associated with clusters suicides. Despite numerous studies on this preventable death, focus has primarily been single nations traditional statistical methods. OBJECTIVE This study aims to develop a predictive model for adolescent suicidal thinking using multinational data sets machine learning (ML). METHODS We used from Korea Youth Risk Behavior Web-based Survey 566,875 aged between 13 18 years conducted external validation 103,874 Norway’s University National General 19,574 adolescents. Several tree-based ML models were developed, feature importance Shapley additive explanations values analyzed identify risk factors thinking. RESULTS When trained South 95% CI, XGBoost reported an area under receiver operating characteristic (AUROC) curve 90.06% (95% CI 89.97-90.16), displaying superior performance compared other models. For United States Norway, achieved AUROCs 83.09% 81.27%, respectively. Across all sets, consistently outperformed highest AUROC score, was selected as optimal model. In terms predictors thinking, feelings sadness despair most influential, accounting 57.4% impact, followed by stress status at 19.8%. age (5.7%), household income (4%), academic achievement (3.4%), sex (2.1%), others, which contributed less than 2% each. CONCLUSIONS integrating diverse 3 countries address suicide. The findings highlight important role emotional health indicators in predicting Specifically, identified significant predictors, stressful conditions age. These emphasize critical need early diagnosis prevention mental issues during adolescence.

Language: Английский

Quantification of identifying cognitive impairment using olfactory-stimulated functional near-infrared spectroscopy with machine learning: a post hoc analysis of a diagnostic trial and validation of an external additional trial DOI Creative Commons
Jae-Won Kim, Hayeon Lee, Jinseok Lee

et al.

Alzheimer s Research & Therapy, Journal Year: 2023, Volume and Issue: 15(1)

Published: July 22, 2023

Abstract Background We aimed to quantify the identification of mild cognitive impairment and/or Alzheimer’s disease using olfactory-stimulated functional near-infrared spectroscopy machine learning through a post hoc analysis previous diagnostic trial and an external additional trial. Methods conducted two independent, patient-level, single-group, interventional trials (original trials) involving elderly volunteers (aged > 60 years) with suspected declining function. All were assessed by measuring oxygenation difference in orbitofrontal cortex open-label approach, medical interview, amyloid positron emission tomography, brain magnetic resonance imaging, Mini-Mental State Examination, Seoul Neuropsychological Screening Battery. Results In total, 97 trial) 36 (additional decline function met eligibility criteria. The statistical model reported classification accuracies 87.3% patients internal validation but 63.9% trial). algorithm achieved 92.5% accuracy data 82.5% data. For diagnosis impairment, performed better than methods (86.0% versus 85.2%) (85.4% 68.8%). Interpretation independent trials, models differences superior diagnosing compared classic models. Our results suggest that is stable across different patient groups increases generalization reproducibility. Trial registration Clinical Research Information Service (CRiS) Republic Korea; CRIS numbers, KCT0006197 KCT0007589. Graphical

Language: Английский

Citations

25
Machine Learning–Based Prediction of Suicidal Thinking in Adolescents by Derivation and Validation in 3 Independent Worldwide Cohorts: Algorithm Development and Validation Study DOI Creative Commons
Hyejun Kim, Yejun Son, Hojae Lee

et al.

Journal of Medical Internet Research, Journal Year: 2024, Volume and Issue: 26, P. e55913 - e55913

Published: May 17, 2024

Background Suicide is the second-leading cause of death among adolescents and associated with clusters suicides. Despite numerous studies on this preventable death, focus has primarily been single nations traditional statistical methods. Objective This study aims to develop a predictive model for adolescent suicidal thinking using multinational data sets machine learning (ML). Methods We used from Korea Youth Risk Behavior Web-based Survey 566,875 aged between 13 18 years conducted external validation 103,874 Norway’s University National General 19,574 adolescents. Several tree-based ML models were developed, feature importance Shapley additive explanations values analyzed identify risk factors thinking. Results When trained South 95% CI, XGBoost reported an area under receiver operating characteristic (AUROC) curve 90.06% (95% CI 89.97-90.16), displaying superior performance compared other models. For United States Norway, achieved AUROCs 83.09% 81.27%, respectively. Across all sets, consistently outperformed highest AUROC score, was selected as optimal model. In terms predictors thinking, feelings sadness despair most influential, accounting 57.4% impact, followed by stress status at 19.8%. age (5.7%), household income (4%), academic achievement (3.4%), sex (2.1%), others, which contributed less than 2% each. Conclusions integrating diverse 3 countries address suicide. The findings highlight important role emotional health indicators in predicting Specifically, identified significant predictors, stressful conditions age. These emphasize critical need early diagnosis prevention mental issues during adolescence.

Language: Английский

Citations

10
Artificial Intelligence–Driven Respiratory Distress Syndrome Prediction for Very Low Birth Weight Infants: Korean Multicenter Prospective Cohort Study DOI Creative Commons
Woocheol Jang, Yong‐Sung Choi, Ji Yoo Kim

et al.

Journal of Medical Internet Research, Journal Year: 2023, Volume and Issue: 25, P. e47612 - e47612

Published: June 14, 2023

Respiratory distress syndrome (RDS) is a disease that commonly affects premature infants whose lungs are not fully developed. RDS results from lack of surfactant in the lungs. The more infant is, greater likelihood having RDS. However, even though all have RDS, preemptive treatment with artificial pulmonary administered most cases.We aimed to develop an intelligence model predict avoid unnecessary treatment.In this study, 13,087 very low birth weight who were newborns weighing less than 1500 grams assessed 76 hospitals Korean Neonatal Network. To infants, we used basic information, maternity history, pregnancy/birth process, family resuscitation procedure, and test at such as blood gas analysis Apgar score. prediction performances 7 different machine learning models compared, 5-layer deep neural network was proposed order enhance performance selected features. An ensemble approach combining multiple 5-fold cross-validation subsequently developed.Our consisting top 20 features provided high sensitivity (83.03%), specificity (87.50%), accuracy (84.07%), balanced (85.26%), area under curve (0.9187). Based on developed, public web application enables easy access for deployed.Our may be useful preparations neonatal resuscitation, particularly cases involving delivery it can aid predicting inform decisions regarding administration surfactant.

Language: Английский

Citations

7
Limitations of the Cough Sound-Based COVID-19 Diagnosis Artificial Intelligence Model and its Future Direction: Longitudinal Observation Study DOI Creative Commons
Jina Kim, Yong‐Sung Choi, Young Ju Lee

et al.

Journal of Medical Internet Research, Journal Year: 2024, Volume and Issue: 26, P. e51640 - e51640

Published: Feb. 6, 2024

Background The outbreak of SARS-CoV-2 in 2019 has necessitated the rapid and accurate detection COVID-19 to manage patients effectively implement public health measures. Artificial intelligence (AI) models analyzing cough sounds have emerged as promising tools for large-scale screening early identification potential cases. Objective This study aimed investigate efficacy using a diagnostic tool COVID-19, considering unique acoustic features that differentiate positive negative We investigated whether an AI model trained on sound recordings from specific periods, especially stages pandemic, were applicable ongoing situation with persistent variants. Methods used 3 data sets (Cambridge, Coswara, Virufy) representing different pandemic Our was Cambridge set subsequent evaluation against all sets. performance analyzed based area under receiver operating curve (AUC) across measurement periods Results demonstrated high AUC when tested set, indicative its initial effectiveness. However, varied significantly other sets, particularly detecting later variants such Delta Omicron, marked decline observed latter. These results highlight challenges maintaining backdrop evolving virus. Conclusions While offer noninvasive method their effectiveness is challenged by emergence new virus Ongoing research adaptations methodologies are crucial address these limitations. adaptability evolve underscores foundational technology not only current but also future outbreaks, contributing more agile resilient global infrastructure.

Language: Английский

Citations

2
Artificial intelligence in the healthcare sector: comparison of deep learning networks using chest X-ray images DOI Creative Commons
Muhammed Akif Yenikaya, Gökhan Kerse, Onur Oktaysoy

et al.

Frontiers in Public Health, Journal Year: 2024, Volume and Issue: 12

Published: April 10, 2024

Artificial intelligence has led to significant developments in the healthcare sector, as other sectors and fields. In light of its significance, present study delves into exploring deep learning, a branch artificial intelligence.

Language: Английский

Citations

2
Few-shot meta-learning for pre-symptomatic detection of Covid-19 from limited health tracker data DOI Creative Commons
Atifa Sarwar, Abdulsalam Almadani, Emmanuel Agu

et al.

Smart Health, Journal Year: 2024, Volume and Issue: 32, P. 100459 - 100459

Published: Feb. 27, 2024

Detecting (or screening for) Covid-19 even before symptoms fully manifest, could enable patients to receive timely and life-saving treatment. Prior work has demonstrated that heart rate step data from low-end wearables analyzed using deep learning models can detect reliably. However, significant individual differences in vital sign manifestation (high inter-subject variability) present a challenge the generalization of across diverse users. The limited amount many medical scenarios further exacerbates this issue. Consequently, neural network learn varied patterns are compelling. Meta-learning emerged as powerful technique for tackling various machine challenges, including insufficient data, domain shifts datasets, issues with generalization. This study proposes MetaCovid, adaptation framework employs meta-learning address variability between subjects only two days order manifest. MetaCovid leverages measurements collected consumer-grade health trackers over preceding 2 days, extracts 45 digital bio-markers (features), which along raw fed into GRU-based an attention mechanism, followed by uncertainty filtering. is trained OC-MAML, one-class few-shot MAML variant adapts target distribution/user samples majority class. generalized well relatively small, publicly available achieving recall 0.81 0.92, detecting 61% (14 out 23) 50% (17 34) users infected symptom onset. When OC-MAML was excluded ablation study, F2 score dropped 36%, highlighting indeed facilitates sensing varying patterns. Notably, outperforms current state-of-art method predicting early on day N compared 28 reducing requirements 93%. To best our knowledge, first propose utilizing mitigate screening. We believe will pave way innovative interventions accurate help contain spread infectious diseases future.

Language: Английский

Citations

0
Cough Sound Based Deep Learning Models for Diagnosis of COVID-19 Using Statistical Features and Time-Frequency Spectrum DOI
Jina Kim, Jinseok Lee

Published: July 15, 2024

This paper presents a deep learning model that can classify COVID-19 patients through cough sounds. The sound data were selected from the Cambridge set which is crowedsourced collected sounds application. Virufy and Coswara sets also for external testing. For waveform, we extracted Variable frequency complex demodulation (VFCDM) image applied to Xception, as pre-trained model. Then zero crossing rate (ZCR), spectral roll-off (SR), centroid (SC), bandwidth (SB), concatenated them output node of Results evaluated by using area under receiver operating curve. set: 0.9346, 0.9244, 0.8250.

Language: Английский

Citations

0
Machine Learning–Based Prediction of Suicidal Thinking in Adolescents by Derivation and Validation in 3 Independent Worldwide Cohorts: Algorithm Development and Validation Study (Preprint) DOI Creative Commons
Hyejun Kim, Yejun Son, Hojae Lee

et al.

Published: Dec. 29, 2023

BACKGROUND Suicide is the second-leading cause of death among adolescents and associated with clusters suicides. Despite numerous studies on this preventable death, focus has primarily been single nations traditional statistical methods. OBJECTIVE This study aims to develop a predictive model for adolescent suicidal thinking using multinational data sets machine learning (ML). METHODS We used from Korea Youth Risk Behavior Web-based Survey 566,875 aged between 13 18 years conducted external validation 103,874 Norway’s University National General 19,574 adolescents. Several tree-based ML models were developed, feature importance Shapley additive explanations values analyzed identify risk factors thinking. RESULTS When trained South 95% CI, XGBoost reported an area under receiver operating characteristic (AUROC) curve 90.06% (95% CI 89.97-90.16), displaying superior performance compared other models. For United States Norway, achieved AUROCs 83.09% 81.27%, respectively. Across all sets, consistently outperformed highest AUROC score, was selected as optimal model. In terms predictors thinking, feelings sadness despair most influential, accounting 57.4% impact, followed by stress status at 19.8%. age (5.7%), household income (4%), academic achievement (3.4%), sex (2.1%), others, which contributed less than 2% each. CONCLUSIONS integrating diverse 3 countries address suicide. The findings highlight important role emotional health indicators in predicting Specifically, identified significant predictors, stressful conditions age. These emphasize critical need early diagnosis prevention mental issues during adolescence.

Language: Английский

Citations

0