Catalysing AI Adoption to Boost Community Healthcare

Explainable AI as a Service aims to make AI understandable and easily configurable for use by healthcare providers, practitioners and patients

By 2030, chronic diseases such as hyperglycaemia (Type II diabetes), hyperlipidaemia and hypertension, collectively known as 3H, will be the top three diseases affecting 18-69 year olds. In particular, the number of people with Type II diabetes will hit an estimated 1 million if nothing is done to change the trajectory of this disease.

The third grant awardee in the AI in Health Grand Challenge – a team of 19 principal investigators from National University of Singapore (NUS) and National University Health System (NUHS) – is aiming to address this by catalysing the mass adoption of AI in healthcare.

About Explainable AI
The team is developing tools and techniques that will make AI understandable and easily configurable so that it can be used by non-data scientists such as clinicians. It is building a platform to enable the delivery of AI as a service, so that it can be applied more widely in areas such as precision medicine, preventive advice and automatic lifestyle coaching. Prototype AI devices are also being developed for deployment and testing in a community setting, to provide healthcare support for patients with chronic diseases.

12-month report card

Forming the backbone of the team’s work is PANDA, a platform which enables AI to be delivered as a service to solve the problems of 3H. Advanced AI techniques were developed to boost the training and inference capabilities of PANDA, which also has built-in automatic model selection and automatic hyperparameter tuning capabilities. This makes it easy to use, especially for healthcare practioners who do not have much background in AI and data science.

In the back end, GPU resource scheduling has been automated to optimise the use of processing power for deep learning training jobs.

The team has also developed a table-top telehealth device called MEDDi (medical digital intermediary) which incorporates vital signs sensing, biomarker sensing, video analytics, a chatbot and an AI-enabled intervention mechanism. MEDDi can be deployed in polyclinics and homes so that patients with chronic diseases can measure and monitor vital health parameters without having to go to the hospital.

For AI-enabled intervention, the device is equipped with CURATE.AI, a software module  which integrates information from various sensors to calibrate and modulate medication dosage, especially in a multi-drug scenario. This enables healthcare providers to make informed decisions about dosage adjustment and behavioural therapy.

The MEDDi chatbot supports users through natural communication channels such as text and voice. For voice communications, the speech recognition system also supports Singlish, which includes code switching between English and Mandarin within the same sentence.

For vital signs sensing, the team has developed and deployed a network of wireless sensors embedded in clothing, which use near-field communications to provide continuous measurement of human physiological signals. For example, self-powered socks equipped with sensing capabilities enable long-term monitoring of the user’s physiological status  including his/her gait in real-time. Another sensor attached to the skin measures pulse wave velocity, making it easy to monitor blood pressure.

Various AI techniques have also been adapted for use in different healthcare-related scenarios. For example, Foodlg is a dietary health app which facilitates food journaling with automatic food image recognition using a convolutional neural network. Pilot studies of Foodlg as an auxiliary healthcare solution are being carried out in collaboration with several hospitals and health establishments. 

Another AI technique, reinforcement learning, has been applied to train the system to improve the patient’s glycaemic control and avoid hypoglycaemia and other complications. 

In related work, the team has also recruited volunteer patients for a Familial Hypercholesterolemia study which focuses on high-risk hyperlipidemia case identification and genetic studies of familial cholesterol risk. A precision cost-benefit calculator has also been developed to support decisions on the use of expensive or branded medicines, and machine learning models are being used to identify genetic signatures of statin-induced muscle aches.

Benefits to Primary Care Teams

With MEDDi, the burden on hospitals can be alleviated by transferring some of the health care functions to polyclinics and eventually to the patient’s home.

One of the main reasons patients with chronic diseases need to go to hospitals is the lack of medical equipment at home to measure and monitor vital health parameters. It is estimated that MEDDi can reduce the number of hospital visits by addressing up to 80 percent of a patient’s healthcare needs at home.

For doctors, AI-as-a-Service delivered through PANDA enables them to use AI intuitively to enhance precision medicine, optimise drug combination therapy and give precise advice on 3H prevention and management.

Benefits to End-users

For patients, devices like MEDDi provide the convenience of monitoring vital health parameters from home, allowing them to make fewer trips to the hospital. At the same time, the integration of this data and application of the right AI models ensure they receive effective interventions and better quality care.

The general public also benefits from the use of AI techniques in apps such as FoodLg app for automatic analysis of food and nutrient intake and coaching.

What’s Next

The team is continuing with extensive data collection efforts to support AI training, validation and testing. Preparations are also underway for the MEDDi prototype to be deployed for beta testing.

For more details, please visit

About the Team

Lead Principal Investigator: Professor Ooi Beng Chin (NUS)

Co-Principal Investigators:

  • Dr Ngiam Kee Yuan (NUHS)
  • Professor Dean Ho (NUS)
  • Professor Wong Lim Soon (NUS)
  • Associate Professor Xiao Xiaokui (NUS)
  • Associate Professor Ng Teck Khim (NUS)
  • Assistant Professor Bryan Low Kian Hsiang (NUS)
  • Assistant Professor Wang Wei (NUS)
  • Professor Lim Chwee Teck (NUS)
  • Associate Professor Qiu Anqi (NUS)
  • Professor Li Haizhou (NUS)
  • Associate Professor Mehul Motani (NUS)
  • Associate Professor Vincent Lee Cheng Kuo (NUS)
  • Assistant Professor John Ho SY (NUS)
  • Assistant Professor Edward Chow Kai-Hua (NUS)
  • Assistant Professor Benjamin Tee Chee Keong (NUS)
  • Assistant Professor Feng Mengling (NUS)

Host Institution: National University of Singapore (NUS)

Partner Institution(s): National University Health System (NUHS)

Other achievements:
The Explainable AI team has papers featured in some of the top conferences including Neural Information Processing Systems (NeurIPS 2019), the Association for Computing Machinery’s Special Interest Group on Management of Data (ACM SIGMOD 2020) and the International Conference on learning Representatins (ICLR 2020), as well as some of the top journals.

About the AI in Health Grand Challenge

The AI in Health Grand Challenge is a five-year, two-stage programme with a total funding quantum of $35 million. AI Singapore, together with an International Review Panel, selected three projects to be awarded Stage 1 funding of $5million per project for the first two years. The projects focused on applying AI technologies in innovative ways across the continuum of 3H (hyperlipidemia, hyperglycemia, hypertension) care.