High-tech healing: Evolving healthcare with AI

On Friday 3 November 2023, the Faculty of Information Technology hosted ‘SHARE 2023 – Sharing about Healthcare Artificial intelligence REgionally’ in partnership with Monash University Malaysia , Apollo Hospitals , The Apollo University and the Indian Institute of Technology, Bombay (IITB).

Here are some insights from the day into the diverse healthcare challenges being addressed with AI:

The Internet of Health Things in Smart AI – Architecture, Application and Challenges

Presented by Dr Satyanarayana Rentala, The Apollo University

The Internet of Health Things (IoHT) is a specialist branch focusing on interconnected medical technologies such as devices, sensors and systems to improve patient care. It can involve multiple types of computing:

• Cloud: Enabling data storage and access via the internet to save physical hard drive space.
• Fog: Providing decentralised computing architecture that extends resources, such as storage, closer to the devices generating the data – allowing for real-time collection and analysis.
• Edge computing: Bringing computation and data storage closer to where it’s needed or generated, facilitating faster response times and lower latency.

Among many applications, the IoHT has enabled continuous healthcare management, remote patient monitoring and telemedicine, driving more timely interventions and fewer hospital visits.?

What’s more, leveraging wearable devices has allowed for the real-time tracking of health conditions and early disease detection. And the mass data generated and circulated through the IoHT has provided crucial insights to advance the state-of-the-art in healthcare.

Multi-modal Medical AI

Presented by Associate Professor Zongyuan Ge, Monash University Australia

Data digitalisation, data linkage and continuous monitoring have revolutionised healthcare by enabling seamless interoperability. One innovation borne from this is multimodal artificial intelligence, AI that can comprehensively understand, interpret and interact with information from multiple sources.?

This has been applied in a range of specialist fields such as dermatology, which only has around 600 registered practitioners in Australia to provide intensive consultations.?

Research is underway developing a machine learning model that can process diverse information, such as family history, genomic sequencing and 3D body scans, on top of dermoscopic images. This multimodal AI will alleviate the workload on dermatologists, making them more accessible, wile helping to produce more holistic and accurate diagnoses of each patient’s condition.

Another specialist area using multimodal AI is ophthalmology, where researchers are trialling ‘AI retinal experts’ that can analyse every part of the eye in detail, and combine these examinations with other image and text reports.

Not only is this beneficial for early disease diagnosis, but it also allows patients to receive more comprehensive information about their conditions with the opportunity to interact and ask questions.?

Point of Care (PoC) MRI Program

Presented by Associate Professor Zhaolin Chen, Monash University Australia

90% of the world doesn’t have access to MRI technology despite it being one of the most powerful yet non-invasive diagnostic tools in modern medicine.?

The causes? Lack of a skilled workforce, no funding, strict government legislations – to name a few. MRI technologies also consume the most energy in radiology, contributing to 1% of carbon emissions.

To address the pressing accessibility issue and environmental impact, researchers are developing a PoC MRI program designed to enable rapid diagnoses and treatments for patients regardless of socioeconomic status and location.?

Not only can it improve and produce high-quality images at low magnetic field strength, but its mobility will allow MRI technology to be transported to people globally, especially in remote areas.

Data Driven Solutions for Health Equity

Presented by Dr Sujoy Kar, Apollo Hospitals Group

Is AI widening the equity gap in healthcare??

Health equity is a global challenge, perpetuated by a lack of physical access and unbiased data, as well as the internet divide. What’s more, human rights and social justice add a layer of complexity that together, make equity a steep, ever-evolving objective.

With the world growing more dependent on technology, it’s critical that we embed standards of accountability into the algorithms that underpin it.?

Researchers in Apollo Hospital are addressing bias from the start, having developed an ethics framework to validate all AI models they create. It involves four key facets:

• Ethical considerations such as fairness, integrity, inclusion, accountability, non-maleficence – factors that govern the development and delivery of AI.?
• Adoption such as interoperability, safety, benchmarking, universalisation – factors that influence the rate and ease of adoption.
• Suitability such as addressing bias, accuracy, human-machine hand-offs and real-world needs – factors that determine whether AI is fit-for-purpose.
• Explainability such as counterfactual reasoning, decision verification, evidence-driven – factors that shape transparency, trust and understanding in an AI model.

Advancing Healthcare Diagnostics: Fine-Tuning Large Language Models (LLMs) for Edge-Based Medical Support

Presented by Dr Gopi Krishna Guntupalli, The Apollo University

LLMs possess billions of parameters which allow them to capture and generate a wide range of linguistic patterns and knowledge.

This equips AI to provide comprehensive answers to questions and engage in creative writing, debugging, testing, translating, customer service – the possibilities are endless. In healthcare, it can significantly aid diagnosis, treatment planning, drug discoveries, clinical trial designs and education.

Driving their evolution, researchers are developing ‘edge-based LLMs’ which are LLMs deployed on devices at the fringes of a network like smartphones and laptops. This close proximity to end users means that data doesn’t have to travel far, leading to lower latency and faster response times.?

Edge-based LLMs also ensure better privacy and security as the data is processed on the devices rather than sent to a cloud server, minimising opportunities for malicious attacks. The lack of a centralised server also reduces costs and increases reliability by eliminating the need for an internet connection.

AI Federated Models in Medicine

Presented by Dr Yasmeen George, Monash University Australia

Training generalisable and robust machine learning and deep learning models require large and diverse datasets. But there are strict regulations around patient privacy and data security, and governance issues concerning who can access the data and how they will use it.

To navigate this challenge, researchers are advancing federated learning – a collaborative approach where models are trained on local sites which gather data, such as hospitals. Anonymous model weights are then fed to a central server which aggregates and redistributes them back to the sites.?

The aggregated weights augment each local AI model’s performance to that of the collective while minimising risks by reducing discrepancies in training parameters across sites. And throughout the process no one is able to see or touch the data, enhancing security.

A specific use case for federated learning has been to train AI on medical images of tumours, addressing differences in the quality and quantity of data labels – driving more consistent, accurate and reliable diagnoses.

Towards Precision Oncology using Machine Learning on Medical Images

Presented by Professor Armit Sethi, IITB and Professor Swapnil Rane TMC

Oncologists conduct deep analyses of tissues and build them into diagnostic information to guide predictions and treatments. Interpretations are done on a case-by-case basis and are largely manual, time-consuming processes.?

What’s more, a digital version of a tissue slide is incredibly large, and 25 to 40 are usually generated per patient – resulting in an annual storage requirement that can reach 450TB. Not to mention that the slides themselves can have holes and disruptive artefacts making examinations difficult.?

Researchers are creating a databank to train AI to make meaningful and accurate predictions based on tissue analysis, supporting oncologists with the intensive workload and process.

The algorithm will also be able to identify and screen artefacts, and present a measure of usability for the whole image which can be taken for further processing following a human-in-the-loop approach.?

Through events like SHARE, we’re strengthening the ever-evolving community of practice and collaborations globally. We’re proud to unite brilliant minds in health and technology, inspiring new ways of applying AI to drive better patient outcomes for all.

– Professor Chris Bain, Lead, ADAM and the Faculty of IT’s Digital Health theme

Want to learn more about the event and the projects? Watch the replay now .

Initiated by the Alliance for Digital health At Monash (ADAM) , this full-day event gathered researchers and clinicians across three countries to share their cutting-edge projects applying AI in healthcare.

Attendees tuned in from local venues at Monash University 's Clayton and Kuala Lumpur campuses, as well as IITB in Mumbai, Apollo University in Chittoor and Apollo Hospital in Hyderabad.