Abstract
Intensive care units (ICUs) in low- and middle-income countries (LMICs) typically suffer from insufficient staff expertise and lack of resources. These ICUs normally manage different patient cohorts to those in high income countries, for example dengue, tetanus, tuberculosis and HIV patients. The Vietnam ICU Translational Applications Laboratory (VITAL) project, which hosts this PhD project, aims at developing, and testing the utility of, affordable technology including wearable devices, artificial intelligence (AI)-enabled imaging and smart usage of patient data to support critical care clinical decision making in Vietnamese ICUs.Ultrasound (US) imaging is affordable, portable, and safe, and can be used to investigate many body organs. As a result, US can be an invaluable tool in a resource limited ICU setting. However, US requires extensive operator experience to be carried out effectively. Such expertise is scarce in LMICs, where there are few specialists and formal US training is uncommon.
AI in US is an exciting prospect and has the potential to optimize existing resources and help overcome workforce shortages by assisting US system operation, measurement of biometric parameters from images, image interpretation, and providing insights that can help patient management. As a result, AI can make US accessible and ultimately improve patient outcomes in resource-limited settings. Although there are barriers to deploying AI-enabled US at scale in LMICs, a strategy of promoting local innovation and initiative can accelerate progress towards sustainable AI-enabled US implementation.
A challenge to accelerate the use of AI in US in LMICs relates to the quantity and quality of the available data. Most algorithms are trained with data from high-income countries (HICs), which may not be representative of LMIC populations (both in terms of patients and diseases). This means that even if AI algorithms are approved commercially, their efficacy and translatability may still be inhibited by the lack of data from LMICs.
This thesis evaluates the clinical usability of AI-enabled ultrasound tools in the ICU in a LMIC. To this end, my work has focused on i) identifying the main challenges and opportunities for AI-enabled US in the ICU of a LMIC, ii) collecting and curating a comprehensive US dataset, iii) adapting pre-existing computational methods for real-time classification and quantification of lung, heart and muscle US videos, and iv) clinical translation of such methods to enable non-experts to obtain expert-quality scans in ICU in LMICs.
| Date of Award | 1 Jun 2024 |
|---|---|
| Original language | English |
| Awarding Institution |
|
| Supervisor | Andrew King (Supervisor), Alberto Gomez Herrero (Supervisor) & Reza Razavi (Supervisor) |