7 research outputs found
Unsupervised Clustering and Automatic Language Model Generation for ASR
No full textI hereby declare that I am the sole author of this thesis. This is a true copy of the thesis, including any required final revisions, as accepted by my examiners. I understand that my thesis may be made electronically available to the public. ii The goal of an automatic speech recognition system is to enable the computer in understanding human speech and act accordingly. In order to realize this goal, language modeling plays an important role. It works as a knowledge source through mimicking human comprehension mechanism in understanding the language. Among many other approaches, statistical language modeling technique is widely used in automatic speech recognition systems. However, the generation of reliable and robust statistical model is very difficult task, especially for a large vocabulary system. For a large vocabulary system, the performance of such a language model degrades as the vocabulary size increases. Hence, the performance of the speech recognition system also degrades due to the increased complexity and mutual confusion among the candidate words in the language model. In order to solve these problems, reduction of language model size as well as minimization o
Characterising acute and chronic care needs: insights from the Global Burden of Disease Study 2019
No full textChronic care manages long-term, progressive conditions, while acute care addresses short-term conditions. Chronic conditions increasingly strain health systems, which are often unprepared for these demands. This study examines the burden of conditions requiring acute versus chronic care, including sequelae. Conditions and sequelae from the Global Burden of Diseases Study 2019 were classified into acute or chronic care categories. Data were analysed by age, sex, and socio-demographic index, presenting total numbers and contributions to burden metrics such as Disability-Adjusted Life Years (DALYs), Years Lived with Disability (YLD), and Years of Life Lost (YLL). Approximately 68% of DALYs were attributed to chronic care, while 27% were due to acute care. Chronic care needs increased with age, representing 86% of YLDs and 71% of YLLs, and accounting for 93% of YLDs from sequelae. These findings highlight that chronic care needs far exceed acute care needs globally, necessitating health systems to adapt accordingly.
© 2025. The Author(s)
Global, regional, and national incidence of six major immune-mediated inflammatory diseases: findings from the global burden of disease study 2019
No full textBackground: The causes for immune-mediated inflammatory diseases (IMIDs) are diverse and the incidence trends of IMIDs from specific causes are rarely studied. The study aims to investigate the pattern and trend of IMIDs from 1990 to 2019. Methods: We collected detailed information on six major causes of IMIDs, including asthma, inflammatory bowel disease, multiple sclerosis, rheumatoid arthritis, psoriasis, and atopic dermatitis, between 1990 and 2019, derived from the Global Burden of Disease study in 2019. The average annual percent change (AAPC) in number of incidents and age standardized incidence rate (ASR) on IMIDs, by sex, age, region, and causes, were calculated to quantify the temporal trends. Findings: In 2019, rheumatoid arthritis, atopic dermatitis, asthma, multiple sclerosis, psoriasis, inflammatory bowel disease accounted 1.59%, 36.17%, 54.71%, 0.09%, 6.84%, 0.60% of overall new IMIDs cases, respectively. The ASR of IMIDs showed substantial regional and global variation with the highest in High SDI region, High-income North America, and United States of America. Throughout human lifespan, the age distribution of incident cases from six IMIDs was quite different. Globally, incident cases of IMIDs increased with an AAPC of 0.68 and the ASR decreased with an AAPC of −0.34 from 1990 to 2019. The incident cases increased across six IMIDs, the ASR of rheumatoid arthritis increased (0.21, 95% CI 0.18, 0.25), while the ASR of asthma (AAPC = −0.41), inflammatory bowel disease (AAPC = −0.72), multiple sclerosis (AAPC = −0.26), psoriasis (AAPC = −0.77), and atopic dermatitis (AAPC = −0.15) decreased. The ASR of overall and six individual IMID increased with SDI at regional and global level. Countries with higher ASR in 1990 experienced a more rapid decrease in ASR. Interpretation: The incidence patterns of IMIDs varied considerably across the world. Innovative prevention and integrative management strategy are urgently needed to mitigate the increasing ASR of rheumatoid arthritis and upsurging new cases of other five IMIDs, respectively. Funding: The Global Burden of Disease Study is funded by the Bill and Melinda Gates Foundation. The project funded by Scientific Research Fund of Sichuan Academy of Medical Sciences & Sichuan Provincial People's Hospital ( 2022QN38). © 2023 The Author(s
Global mortality associated with 33 bacterial pathogens in 2019: a systematic analysis for the Global Burden of Disease Study 2019
No full textBackground Reducing the burden of death due to infection is an urgent global public health priority. Previous studies have estimated the number of deaths associated with drug-resistant infections and sepsis and found that infections remain a leading cause of death globally. Understanding the global burden of common bacterial pathogens (both susceptible and resistant to antimicrobials) is essential to identify the greatest threats to public health. To our knowledge, this is the first study to present global comprehensive estimates of deaths associated with 33 bacterial pathogens across 11 major infectious syndromes.Methods We estimated deaths associated with 33 bacterial genera or species across 11 infectious syndromes in 2019 using methods from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2019, in addition to a subset of the input data described in the Global Burden of Antimicrobial Resistance 2019 study. This study included 343 million individual records or isolates covering 11 361 study-location-years. We used three modelling steps to estimate the number of deaths associated with each pathogen: deaths in which infection had a role, the fraction of deaths due to infection that are attributable to a given infectious syndrome, and the fraction of deaths due to an infectious syndrome that are attributable to a given pathogen. Estimates were produced for all ages and for males and females across 204 countries and territories in 2019. 95% uncertainty intervals (UIs) were calculated for final estimates of deaths and infections associated with the 33 bacterial pathogens following standard GBD methods by taking the 2.5th and 97.5th percentiles across 1000 posterior draws for each quantity of interest.Findings From an estimated 13.7 million (95% UI 10.9-17.1) infection-related deaths in 2019, there were 7.7 million deaths (5.7-10.2) associated with the 33 bacterial pathogens (both resistant and susceptible to antimicrobials) across the 11 infectious syndromes estimated in this study. We estimated deaths associated with the 33 bacterial pathogens to comprise 13.6% (10.2-18.1) of all global deaths and 56.2% (52.1-60.1) of all sepsis-related deaths in 2019. Five leading pathogens-Staphylococcus aureus, Escherichia coli, Streptococcus pneumoniae, Klebsiella pneumoniae, and Pseudomonas aeruginosa-were responsible for 54.9% (52.9-56.9) of deaths among the investigated bacteria. The deadliest infectious syndromes and pathogens varied by location and age. The age-standardised mortality rate associated with these bacterial pathogens was highest in the sub-Saharan Africa super-region, with 230 deaths (185-285) per 100 000 population, and lowest in the high-income super-region, with 52.2 deaths (37.4-71.5) per 100 000 population. S aureus was the leading bacterial cause of death in 135 countries and was also associated with the most deaths in individuals older than 15 years, globally. Among children younger than 5 years, S pneumoniae was the pathogen associated with the most deaths. In 2019, more than 6 million deaths occurred as a result of three bacterial infectious syndromes, with lower respiratory infections and bloodstream infections each causing more than 2 million deaths and peritoneal and intra-abdominal infections causing more than 1 million deaths.Interpretation The 33 bacterial pathogens that we investigated in this study are a substantial source of health loss globally, with considerable variation in their distribution across infectious syndromes and locations. Compared with GBD Level 3 underlying causes of death, deaths associated with these bacteria would rank as the second leading cause of death globally in 2019; hence, they should be considered an urgent priority for intervention within the global health community. Strategies to address the burden of bacterial infections include infection prevention, optimised use of antibiotics, improved capacity for microbiological analysis, vaccine development, and improved and more pervasive use of available vaccines. These estimates can be used to help set priorities for vaccine need, demand, and development. Copyright (c) 2022 The Author(s). Published by Elsevier Ltd. This is an Open Access article under the CC BY 4.0 license
Age-sex differences in the global burden of lower respiratory infections and risk factors, 1990-2019: results from the Global Burden of Disease Study 2019
No full textBackground The global burden of lower respiratory infections (LRIs) and corresponding risk factors in children older than 5 years and adults has not been studied as comprehensively as it has been in children younger than 5 years. We assessed the burden and trends of LRIs and risk factors across a groups by sex, for 204 countries and territories.Methods In this analysis of data for the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2019, we used dinician-diagnosed pneumonia or bronchiolitis as our case definition for LRIs. We included International Classification of Diseases 9th edition codes 079.6, 466-469, 470.0, 480-482.8, 483.0-483.9, 484.1-484.2, 484.6-484.7, and 487-489 and International Classification of Diseases 10th edition codes A48.1, A70, B97.4 B97.6, 109-115.8, J16 J16.9, J20-121.9, J91.0, P23.0 P23.4, and U04 U04.9. We used the Cause of Death Ensemble modelling strategy to analyse 23109 site-years of vital r *stration data, 825 site-years of sample vital registration data, 1766 site-years of verbal autopsy data, and 681 site-years of mortality surveillance data. We used DisMod-MR 2.1, a Bayesian metaregression tool, to analyse age sex-specific incidence and prevalence data identified via systematic reviews of the literature, population-based survey data, and daims and inpatient data. Additio y, we estimated age sex-specific LRI mortality that is attributable to the independent effects of 14 risk factors.Findings Globally, in 2019, we estimated that there were 257 million (95% uncertainty interval [UI] 240-275) LRI incident episodes in males and 232 million (217-248) in females. In the same year, LRIs accounted for 1.30 million (95% UI 1.18-1.42) male deaths and 1.20 million (1.07-1.33) female deaths. Age-standardised incidence and mortality rates were 1.17 times (95% UI 1.16-1.18) and 1.31 times (95% UI 1.23-1.41) greater in males than in fe es in 2019. Between 1990 and 2019, LRI incidence and mortality rates declined at different rates across age groups and an increase in LRI episodes and deaths was estimated among all adult age groups, with males aged 70 years and older having the highest increase in LRI episodes (126.0% [95% UI 121.4-131.1]) and deaths (100.0% [83.4-115.9]). During the same period, LRI episodes and deaths in children younger than 15 years were estimated to have decreased, and the greatest dedine was observed for LRI deaths in males younger than 5 years (-70.7% [-77.2 to 61.8]). The leading risk factors for LRI mortality varied across age groups and sex. More than half of global LRI deaths in children younger than 5 years were attributable to child wasting (population attributable fraction [PAF] 53.0% [95% UI 37.7-61.8] in males and 56.4% [40.7-65.1] in females), and more than a quarter of LRI deaths among those aged 5-14 years were attributable to household air pollution (PAF 26.0% [95% UI 16.6-35.5] for males and PAF 25.8% [16.3-35.4] for females). PAFs of male LRI deaths attributed to smoking were 20.4% (95% UI 15.4-25.2) in those aged 15-49 years, 305% (24.1-36. 9) in those aged 50-69 years, and 21.9% (16. 8-27. 3) in those aged 70 years and older. PAFs of female LRI deaths attributed to household air pollution were 21.1% (95% UI 14.5-27.9) in those aged 15-49 years and 18 " 2% (12.5-24.5) in those aged 50-69 years. For females aged 70 years and older, the leading risk factor, ambient particulate matter, was responsible for 11-7% (95% UI 8.2-15.8) of LRI deaths.Interpretation The patterns and progress in reducing the burden of LRIs and key risk factors for mortality varied across age groups and sexes. The progress seen in children you - than 5 years was dearly a result of targeted interventions, such as vaccination and reduction of exposure to risk factors. Similar interventions for other age groups could contribute to the achievement of multiple Sustainable Development Goals targets, induding promoting wellbeing at all ages and reducing health inequalities. Interventions, including addressing risk factors such as child wasting, smoking, ambient particulate matter pollution, and household air pollution, would prevent deaths and reduce health disparities.Copyright 2022 The Author(s). Published by Elsevier Ltd
Global, regional, and national incidence of six major immune-mediated inflammatory diseases : findings from the global burden of disease study 2019
No full textDATA SHARING STATEMENT : Data used for the analyses are publicly available from the Institute of Health Metrics and Evaluation (http://www.healthdata.org/; http:// ghdx.healthdata.org/gbd-results-tool).BACKGROUND : The causes for immune-mediated inflammatory diseases (IMIDs) are diverse and the incidence trends of IMIDs from specific causes are rarely studied. The study aims to investigate the pattern and trend of IMIDs from 1990 to 2019. METHODS : We collected detailed information on six major causes of IMIDs, including asthma, inflammatory bowel disease, multiple sclerosis, rheumatoid arthritis, psoriasis, and atopic dermatitis, between 1990 and 2019, derived from the Global Burden of Disease study in 2019. The average annual percent change (AAPC) in number of incidents and age standardized incidence rate (ASR) on IMIDs, by sex, age, region, and causes, were calculated to quantify the temporal trends. FINDINGS : In 2019, rheumatoid arthritis, atopic dermatitis, asthma, multiple sclerosis, psoriasis, inflammatory bowel disease accounted 1.59%, 36.17%, 54.71%, 0.09%, 6.84%, 0.60% of overall new IMIDs cases, respectively. The ASR of IMIDs showed substantial regional and global variation with the highest in High SDI region, High-income North America, and United States of America. Throughout human lifespan, the age distribution of incident cases from six IMIDs was quite different. Globally, incident cases of IMIDs increased with an AAPC of 0.68 and the ASR decreased with an AAPC of −0.34 from 1990 to 2019. The incident cases increased across six IMIDs, the ASR of rheumatoid arthritis increased (0.21, 95% CI 0.18, 0.25), while the ASR of asthma (AAPC = −0.41), inflammatory bowel disease (AAPC = −0.72), multiple sclerosis (AAPC = −0.26), psoriasis (AAPC = −0.77), and atopic dermatitis (AAPC = −0.15) decreased. The ASR of overall and six individual IMID increased with SDI at regional and global level. Countries with higher ASR in 1990 experienced a more rapid decrease in ASR. INTERPRETATION : The incidence patterns of IMIDs varied considerably across the world. Innovative prevention and integrative management strategy are urgently needed to mitigate the increasing ASR of rheumatoid arthritis and upsurging new cases of other five IMIDs, respectively.The Global Burden of Disease Study is funded by the Bill and Melinda Gates Foundation. Support from Scientific Research Fund of Sichuan Academy of Medical Sciences & Sichuan Provincial People’s Hospital; Shaqra University; the School of Pharmacy, University of Botswana; the Indian Council of Medical Research (ICMR); an Australian National Health and Medical Research Council (NHMRC) Investigator Fellowship; the Italian Center of Precision Medicine and Chronic Inflammation in Milan; the Department of Environmental Health Engineering of Isfahan University of Medical Sciences, Isfahan, Iran; National Health and Medical Research Council (NHMRC), Australia; Jazan University, Saudi Arabia; the Clinician Scientist Program of the Clinician Scientist Academy (UMEA) of the University Hospital Essen; AIMST University, Malaysia; the Department of Community Medicine, Kasturba Medical College, Mangalore, Manipal Academy of Higher Education, Manipal, India; a Kornhauser Research Fellowship at The University of Sydney; the National Research, Development and Innovation Office Hungary; Taipei Medical University; CREATE Hope Scientific Fellowship from Lung Foundation Australia; the National Institute for Health and Care Research Manchester Biomedical Research Centre and an NIHR Clinical Lectureship in Respiratory Medicine; Kasturba Medical College, Mangalore and Manipal Academy of Higher Education, Manipal; Author Gate Publications; the Cleveland Clinic Foundation and Nassau University Medical center; the Italian Ministry of Health (RRC); King Abdulaziz University (DSR), Jeddah, and King Abdulaziz City for Science & Technology (KACSAT), Saudi Arabia, Science & Technology Development Fund (STDF), and US-Egypt Science & Technology joint Fund: The Academy of Scientific Research and Technology (ASRT), Egypt; partially supported by the Centre of Studies in Geography and Spatial Planning; the International Center of Medical Sciences Research (ICMSR), Islamabad Pakistan; Ain Shams University and the Egyptian Fulbright Mission Program; the Belgian American Educational Foundation; Health Data Research UK; the Spanish Ministry of Science and Innovation, Institute of Health Carlos III, CIBERSAM, and INCLIVA; the Clinical Research Development Unit, Imam Reza Hospital, Mashhad University of Medical Sciences; Shaqra University; Saveetha Institute of Medical and Technical Sciences and SRM Institute of Science and Technology; University of Agriculture, Faisalabad-Pakistan; the Chinese University of Hong Kong Research Committee Postdoctoral Fellowship Scheme; the institutional support of the Department of Microbiology and Immunology, Faculty of Pharmacy, Zagazig University, Egypt; the European (EU) and Developing Countries Clinical Trials Partnership, the EU Horizon 2020 Framework Programme, UK-National Institute for Health and Care Research, the Mahathir Science Award Foundation and EU-EDCTP.http://www.thelancet.comam2024School of Health Systems and Public Health (SHSPH)SDG-03:Good heatlh and well-bein
Uso de inteligencia artificial en ortodoncia para el apoyo en la toma de decisión de extracción y seleccion entre los primeros y segundos premolares mediante el uso de variables cefalometricas extraidas de una muestra de población del Valle del Cauca
No full textLa extracción dental en ortodoncia constituye una decisión clínica irreversible con importantes implicaciones estéticas y funcionales, que exige un delicado equilibrio entre la generación de espacio para resolver discrepancias dentoalveolares y la preservación de la armonía facial. Esta intervención, particularmente controvertida, presenta complejidades adicionales al seleccionar entre primeros y segundos premolares, fundamentada en múltiples variables cefalométricas, clínicas y demográficas cuya interpretación varía significativamente entre poblaciones. El presente estudio propone una estrategia para apoyar en la toma de decisiones de extracción y selección entre primeros y segundos premolares en pacientes adultos sometidos a tratamiento ortodóncico, utilizando medidas cefalométricas extraídas de proyecciones 2D obtenidas de imágenes CBCT corregidas a posición natural de cabeza (PNC). Con aprobación del Comité de Ética de la Universidad Autónoma de Occidente (Acta No. 01-2024), se analizó una muestra de 500 pacientes de Cali, Valle del Cauca, incluyendo 325 mujeres (65%) y 175 hombres (35%), con distribución étnica de 88.4% mestizos y 11.6% afrocolombianos. Mediante encuesta a 16 ortodoncistas locales, se seleccionaron 36 variables consideradas determinantes: 26 cefalométricas, 7 clínicas y 3 demográficas. La metodología implementó tres componentes esenciales: primero, un algoritmo para la corrección de PNC en los planos axial, coronal y sagital mediante referencias anatómicas estables (suturas frontocigomáticas, cavidades oculares y línea Sella-Nasion), logrando un error cuadrático medio inferior a 0.064° en los tres planos. Segundo, un modelo de aprendizaje profundo para ubicación automática de 26 puntos cefalométricos, con error radial medio de 0.97 mm y tasa de detección del 90% a 2 mm. Tercero, un modelo de ensamble jerárquico para la decisión de extracción dental, compuesto por ensambles específicos para arcadas superior e inferior. Este sistema alcanzó una exactitud de 99%, precisión de 97.9%, sensibilidad de 100% y F1-Score de 98.9%, superando significativamente estudios previos con exactitudes máximas de 97.97%. Para la selección entre primeros y segundos premolares, se desarrolló un enfoque secuencial donde el modelo inferior (GradientBoosting) precede al superior (RandomForest), integrando la predicción inferior como variable adicional en el análisis superior. Esta aproximación algorítmica logró exactitudes de 90% y 95% para arcadas inferior y superior respectivamente, mejorando considerablemente el máximo reportado de 84.2%. Las técnicas de interpretabilidad (SHAP y gráficos de dependencia parcial condicionados) identificaron umbrales críticos para variables determinantes: L1-APog (4.58 mm para extracción general; 6.2 mm para selección de premolares), apiñamiento inferior (-1.01 mm para extracción; -8.3 mm para selección) y U1-APog (9.5 mm). Estos umbrales optimizan la toma de decisiones al representar puntos de inflexión donde cada variable incrementa significativamente su contribución a la predicción de extracción. Esta investigación constituye un aporte original a la ortodoncia clínica, proporcionando una herramienta diagnóstica de apoyo adaptada específicamente para población mestiza colombiana del Valle del Cauca.Dental extraction in orthodontics is an irreversible clinical decision with significant aesthetic and functional implications, requiring a delicate balance between creating space to resolve dentoalveolar discrepancies and preserving facial harmony. This intervention, particularly controversial, presents additional complexities when choosing between first and second premolars, based on multiple cephalometric, clinical, and demographic variables whose interpretation varies significantly across populations. The present study proposes a strategy to support decision-making regarding extractions and selection between first and second premolars in adult patients undergoing orthodontic treatment, utilizing cephalometric measurements extracted from 2D projections obtained from CBCT images corrected to natural head position (NHP). With approval from the Ethics Committee of Universidad Autónoma de Occidente (Act No. 012024), a sample of 500 patients from Cali, Valle del Cauca, was analyzed, comprising 325 women (65%) and 175 men (35%), with an ethnic distribution of 88.4% mestizos and 11.6% Afro-Colombians. Through a survey of 16 local orthodontists, 36 determining variables were selected: 26 cephalometric, 7 clinical, and 3 demographic. The methodology implemented three essential components: Firstly, an algorithm for NHP correction in axial, coronal, and sagittal planes using stable anatomical references (frontozygomatic sutures, orbital cavities, and SellaNasion line) was achieved with a mean squared error below 0.064° in all three planes. Secondly, a deep learning model for automatic localization of 26 cephalometric points was developed, achieving a mean radial error of 0.97 mm and a detection rate of 90% within 2 mm. Thirdly, a hierarchical ensemble model was designed for the extraction decision, composed of specific ensembles for upper and lower arches. This system reached an accuracy of 99%, precision of 97.9%, sensitivity of 100%, and an F1 score of 98.9%, significantly surpassing previous studies that reported maximum accuracies of 97.97%. For the selection between first and second premolars, a sequential approach was developed where the lower arch model (GradientBoosting) precedes the upper arch model (RandomForest), integrating the lower arch prediction as an additional variable in the upper analysis. This algorithmic approach achieved accuracies of 90% and 95% for the lower and upper arches, respectively, considerably improving the previously reported maximum of 84.2%. Interpretability techniques (SHAP and conditional partial dependence plots) identified critical thresholds for determinant variables: L1-APog (4.58 mm for general extraction; 6.2 mm for premolar selection), lower crowding (-1.01 mm for extraction; -8.3 mm for selection), and U1APog (9.5 mm). These thresholds optimize clinical decision-making by representing inflection points where each variable significantly increases its contribution to extraction prediction. This research represents an original contribution to clinical orthodontics, providing a diagnostic tool specifically adapted for the Colombian mestizo population, enhancing consistency and precision in a clinically irreversible decision-making processTesis (Doctor en Bioingeniería)-- Universidad Autónoma de Occidente, 2025DoctoradoDoctor(a) en Bioingenierí
