You’re offline. This is a read only version of the page.
Skip to main content
KTH Degree Project Portal
Växla navigering
Assignments
Subscribe
About KTH Degree Project Portal
About KTH Degree Project Portal
For Employers
For Students
FAQ
English
English
Svenska
Sign in
Apply using email
Apply using URL
General
Headline
*
*
Organization/Company
*
Location
*
Assignment Type
*
Description
*
*
This project is a collaboration between The Royal Institute of Technology (KTH) and Karolinska Institutet (KI). Supervisor: Oscar Lapidus (KI) Co-supervisors: Martin Jacobsson (KTH), Rebecka Rubenson Wahlin (KI) Background: To decrease mortality and morbidity associated with severe traumatic injuries, trauma systems worldwide have adapted triage tools to rapidly identify trauma patients in need of urgent medical attention. In-hospital responses, such as trauma team activation (TTA), have also been developed to summon staff to the emergency department to prepare for the arrival of a severely injured patient. In regional trauma systems encompassing multiple emergency hospitals, severely injured patients are often directed to the regional trauma center in accordance with prehospital trauma guidelines. However, at non-trauma center emergency hospitals this may pose challenges when determining the need for TTA for these patients. Previous investigations have shown that adherence to national guidelines for TTA is relatively low in clinical practice, and that the criteria may be unsuitable in trauma cohorts where there is heavy selection towards patients with non-apparent injuries. No previous studies have attempted to determine the need for in-hospital TTA using artificial intelligence trained on a large dataset based on the local trauma cohort. Aim: This thesis project aims to investigate the feasibility and accuracy of a novel method of trauma triage using machine learning by analysing data from the Swedish Trauma Registry and pre-hospital medical records. For this purpose, a total of ≈4000 trauma patients treated in Sweden during 2019-2022 have been identified. Task: To develop the machine learning model using supervised learning methodology. Any available data at the time of arrival to hospital can constitute a feature of the model, such as prehospital vital signs and a free-text description of the injuries. Here, we may use modern large language models, such as the new open source GPT-SW by AI Sweden. All necessary data is available at the start of the project. Application: The machine learning model will be adjusted to provide a percentage estimate of the likelihood of severe injuries (injury severity score, ISS ≥15). An ROC analysis will be used to determine the appropriate threshold for TTA in relation to current triage guidelines. The rate of under/overtriage will be calculated and compared to the results of previous investigations. Student: This thesis project is open to Master or Civ.Ing. students (or equivalent) with knowledge in data analytics, machine learning, and/or medical engineering. Previous medical knowledge is not required, but a genuine interest is desirable.
There are no records to display.
You don't have permissions to view these records.
Error completing request.
Loading...
Create
×
Close
Edit
×
Close
View details
×
Close
Delete
×
Close
Are you sure you want to delete this record?
Error
×
Close
We're sorry, an error has occurred.
Deadline for application
*
*
Publish date
*
*
Credits
30 hp
15 hp
15-30 hp
Application Channel
*
Email
URL
Both
Application Email
*
*
Application URL
*
Application Documents
*
Your application can be written in Swedish or English and should include a CV and transcript.