Register now After registration you will be able to apply for this opportunity online.
Personalized algorithms for prediction of sepsis onset
Development and implementation of personalized algorithms for personalized prediction algorithms for the detection of sepsis in different population groups.
Keywords: machine learning, deep learning, sensor fusion, predictive analysis, adverse event prediction, medical decision support, computer science, TinyML
Sepsis is a life-threatening host response to infection. Globally, sepsis is associated with high mortality, morbidity, and health costs across populations of all age groups. Patient management is highly time-sensitive as delayed treatment increases probability of mortality. In spite of decades of research, robust biomarkers for sepsis are missing due to low temporal resolution of available vital sign data and dependence of different patient monitors resulting in delays in recording of electronic medical records. We propose a unique approach to assist in diagnostics by continuously monitoring multiple vital signs to aid in disease management of neonatal sepsis patients with digital biomarkers. This approach would result in providing medical decision support using self-sustainable multi-sensor smart patch with on-board tiny machine learning based personalized medicine algorithms.
However, for accurate prognosis it is critical to personalize the model for sepsis detection using vital sign data. This project aims to develop models that are personalized to the demographic of the patient taking into account the patient history ( heart disease, kidney disease, diabetes etc.), gender and age.
**Prerequisites**
• Python (Matlab) programming;
• Basic Knowledge of machine learning algorithms such as regression, clustering and deep learning.
• Knowledge of tinyML for implementing machine learning on microcontrollers is beneficial.
Sepsis is a life-threatening host response to infection. Globally, sepsis is associated with high mortality, morbidity, and health costs across populations of all age groups. Patient management is highly time-sensitive as delayed treatment increases probability of mortality. In spite of decades of research, robust biomarkers for sepsis are missing due to low temporal resolution of available vital sign data and dependence of different patient monitors resulting in delays in recording of electronic medical records. We propose a unique approach to assist in diagnostics by continuously monitoring multiple vital signs to aid in disease management of neonatal sepsis patients with digital biomarkers. This approach would result in providing medical decision support using self-sustainable multi-sensor smart patch with on-board tiny machine learning based personalized medicine algorithms. However, for accurate prognosis it is critical to personalize the model for sepsis detection using vital sign data. This project aims to develop models that are personalized to the demographic of the patient taking into account the patient history ( heart disease, kidney disease, diabetes etc.), gender and age.
**Prerequisites** • Python (Matlab) programming; • Basic Knowledge of machine learning algorithms such as regression, clustering and deep learning. • Knowledge of tinyML for implementing machine learning on microcontrollers is beneficial.
**Project Tasks**
• List the most common algorithms for personalization of patient data and evaluate their
advantages and drawbacks.
• Use existing dataset ( PhysioNet 2019) to develop algorithms for sepsis prediction using machine learning and sensor fusion.
• Identify and cluster different patient phenotype to implement personalization algorithms
• Implement developed algorithms on microcontrollers for real time risk assessment
**Project Tasks** • List the most common algorithms for personalization of patient data and evaluate their advantages and drawbacks. • Use existing dataset ( PhysioNet 2019) to develop algorithms for sepsis prediction using machine learning and sensor fusion. • Identify and cluster different patient phenotype to implement personalization algorithms • Implement developed algorithms on microcontrollers for real time risk assessment