How Singapore built AI to predict pneumonia severity

By Shirley Tay

The application can be used for Covid-19 patients, and will alert doctors to patients who require critical care.

Images: Changi General Hospital and IHiS

Singapore’s Changi General Hospital (CGH) and the Integrated Health Information System (IHiS) have built an artificial intelligence tool to help doctors determine the severity of pneumonia in patients.

Pneumonia is one of the leading causes of death worldwide, and the main cause of deterioration in Covid-19. The AI tool allows for the risk of patients requiring critical care to be calculated “almost instantaneously”, said Dr Charlene Liew, Project Lead and Deputy Chief Medical Informatics Officer of CGH.

Initial results have shown an approximate accuracy of 80 per cent in predicting severe pneumonia, says IHiS.
 

How it works


The model first observes a patient’s chest X-Ray scans and highlights abnormalities such as opaque areas in the lungs, IHiS tells GovInsider.

It then generates a risk score for three areas: low-risk pneumonia with anticipated short hospitalisation; the risk of mortality; and the risk of requiring critical care.

Doctors can receive an early warning for possible deterioration and take interim measures to improve patient outcomes, said Liew. The score also allows doctors to prioritise treatment for patients in critical condition.



The AI tool can be used for all forms of community-acquired pneumonia including Covid-19, IHiS says.
“Driven by the clinical care needs and resource demands of the pandemic, the CGH and IHiS teams saw the potential of AI to combat the critical needs of Covid-19,” said Professor Ng Wai Hoe, Chief Executive Officer of the hospital.

The 80 per cent accuracy of the AI model is comparable to manual pneumonia severity scoring tools, says IHiS. These manual methods take a lot of time and resources as they typically take 5 to 20 variables into account, while the AI tool computes a score almost instantly.
 

How it was created


The AI tool is known as the Community Acquired Pneumonia and Covid-19 Artificial Intelligence Predictive Engine (CAPE), and is one of over 50 solutions IHiS has created thus far.

“Technology has been a crucial enabler in every stage of our fight against Covid-19,” said Bruce Liang, CEO of IHiS.
Six data scientists from IHiS split into two teams to use different coding approaches. The teams then checked that the results achieved are the same, ensuring the consistency of the model. The machine learning model was also fed with more than 3,000 X-ray images and 200,000 data points to train it to predict risk scores accurately.

Due to the urgency and potential impact it had on managing Covid-19, the team accelerated the project and completed it in two months, says IHiS.

"Operationalising machine learning models in healthcare is extremely challenging, particularly for triaging severity of illness at the point of admission for Covid-19," said Prof Ankur Teredesai, Health Day Chair at the KDD 2020 conference. He is also Chief Technology Officer and Co-Founder of KenSci and Professor of Computer Science & Systems at the School of Engineering & Technology, University of Washington Tacoma.

"I have spent over a decade in helping health systems embed AI in workflows and was very impressed by the agility and comprehensiveness of the CAPE system as reported in the paper at KDD Health Day 2020 this August," he added.
CAPE is designed as a standalone desktop application, and can easily be installed on radiologist workstations. That allows it to be deployed quickly and reduces disruption to the existing clinician workflow.
 

Improving accuracy and scaling for wider use


Moving forward, CGH and IHiS are looking to integrate electronic medical records and clinical data from other public health institutions. That will very likely improve the accuracy of CAPE, says IHiS.

The team is also thinking of hosting CAPE as a “freeware” collaborative tool on a research platform for interested researchers to test out.

“Beyond local healthcare settings, CAPE can potentially be calibrated to identify and predict the severity of respiratory infections globally. That would be crucial during pandemics such as Covid-19, where there could be an increased need for inpatient and critical care support,” says the healthtech agency.