Publicação
Development of Artificial Intelligence Algorithms for Early Diagnosis of Sepsis
| Resumo: | Sepsis is a prevalent syndrome that manifests itself through an uncontrolled response from the body to an infection, that may lead to organ dysfunction. Its diagnosis is urgent since early treatment can reduce the patients’ chances of having long-term consequences. Yet, there are many obstacles to achieving this early detection. Some stem from the syndrome’s pathogenesis, which lacks a characteristic biomarker. The available clinical detection tools are either too complex or lack sensitivity, in both cases delaying the diagnosis. Another obstacle relates to modern technology, that when paired with the many clinical parameters that are monitored to detect sepsis, result in extremely heterogenous and complex medical records, which constitute a big obstacle for the responsible clinicians, that are forced to analyse them to diagnose the syndrome. To help achieve this early diagnosis, as well as understand which parameters are most relevant to obtain it, an approach based on the use of Artificial Intelligence algorithms is proposed in this work, with the model being implemented in the alert system of a sepsis monitoring platform. This platform uses a Random Forest algorithm, based on supervised machine learning classification, that is capable of detecting the syndrome in two different scenarios. The earliest detection can happen if there are only five vital sign parameters available for measurement, namely heart rate, systolic and diastolic blood pressures, blood oxygen saturation level, and body temperature, in which case, the model has a score of 83% precision and 62% sensitivity. If besides the mentioned variables, laboratory analysis measurements of bilirubin, creatinine, hemoglobin, leukocytes, platelet count, and Creactive protein levels are available, the platform’s sensitivity increases to 77%. With this, it has also been found that the blood oxygen saturation level is one of the most important variables to take into account for the task, in both cases. Once the platform is tested in real clinical situations, together with an increase in the available clinical data, it is believed that the platform’s performance will be even better. |
|---|---|
| Autores principais: | Manoel, Francisca de Almeida Nunes Tudela |
| Assunto: | Sepsis Early Diagnosis Artificial Intelligence Machine Learning Alert System Monitoring |
| Ano: | 2022 |
| País: | Portugal |
| Tipo de documento: | dissertação de mestrado |
| Tipo de acesso: | acesso aberto |
| Instituição associada: | Universidade Nova de Lisboa |
| Idioma: | inglês |
| Origem: | Repositório Institucional da UNL |
| Resumo: | Sepsis is a prevalent syndrome that manifests itself through an uncontrolled response from the body to an infection, that may lead to organ dysfunction. Its diagnosis is urgent since early treatment can reduce the patients’ chances of having long-term consequences. Yet, there are many obstacles to achieving this early detection. Some stem from the syndrome’s pathogenesis, which lacks a characteristic biomarker. The available clinical detection tools are either too complex or lack sensitivity, in both cases delaying the diagnosis. Another obstacle relates to modern technology, that when paired with the many clinical parameters that are monitored to detect sepsis, result in extremely heterogenous and complex medical records, which constitute a big obstacle for the responsible clinicians, that are forced to analyse them to diagnose the syndrome. To help achieve this early diagnosis, as well as understand which parameters are most relevant to obtain it, an approach based on the use of Artificial Intelligence algorithms is proposed in this work, with the model being implemented in the alert system of a sepsis monitoring platform. This platform uses a Random Forest algorithm, based on supervised machine learning classification, that is capable of detecting the syndrome in two different scenarios. The earliest detection can happen if there are only five vital sign parameters available for measurement, namely heart rate, systolic and diastolic blood pressures, blood oxygen saturation level, and body temperature, in which case, the model has a score of 83% precision and 62% sensitivity. If besides the mentioned variables, laboratory analysis measurements of bilirubin, creatinine, hemoglobin, leukocytes, platelet count, and Creactive protein levels are available, the platform’s sensitivity increases to 77%. With this, it has also been found that the blood oxygen saturation level is one of the most important variables to take into account for the task, in both cases. Once the platform is tested in real clinical situations, together with an increase in the available clinical data, it is believed that the platform’s performance will be even better. |
|---|