Introduction to AI

Artificial Intelligence (AI) has changed almost every aspect of modern technology, and it is inevitable that we will soon witness its impact on healthcare. It is therefore vital that healthcare professionals (HCPs) have a basic understanding of how AI works and what impact it will have on our future practice.

Ways AI is integrated into healthcare : 

1. Medical education and teaching e.g. ChatGPT creating ‘Observed Structured Clinical Examination’ (OSCE) scenarios.
2. Medical Research: use of deep learning AI models to analyse clinical data. 
3. Question Banks: AI answers students’ questions on online question banks and simulation training.
4. Healthcare administration: AI booking appointments, medical ethics and law (e.g. confidentiality). 
5. Going forward it could potentially be integrated into clinical practice. 

Through this series of articles, readers should be able to articulate the differences between machine learning and deep learning. The reader should also gain an appreciation of how AI can fit into healthcare.

In the upcoming editions, the reader can look forward to learning about the advantages and disadvantages of AI, types of machine learning, including the latest generative AI, and an explanation on deepfakes, hallucinations and how to recognise them.

 The big question: What is AI?

AI started as a conviction that we can replicate human intelligence using machines (i.e. computers). It has developed into an umbrella term encompassing traditional machine learning and contemporary deep learning. 

Machine Learning : 

Traditional machine learning is a large group of statistical methods employed to help us understand patterns in large datasets in profound ways that no ordinary human being can process in a reasonable amount of time. It uses mathematical methods to work out trends and predict outcomes without being explicitly programmed to do so. This is important in healthcare as it thrives when large quantities of data are used and can provide insights that humans would have missed. A real-life example of this would be Pfizer using machine learning technology to help speed up the process of creating a COVID-19 vaccine. They used Smart Data Query (SDQ), one of many machine learning tools, to check over their patient trial data before it was analysed by researchers. This tool allowed data to be reviewed in 22 hours which otherwise would have taken a team of humans 30 days to do.

Deep Learning :

Deep learning is a subset of machine learning (Figure 1). Its defining feature is that it mimics the human brain by implementing many layers of “neurones” that process data in sequence. A clever mathematical equation is implemented to allow learning within the connections between these layers of neurons. This mathematical equation is simply a more complex form of calculus you would have encountered in school. As these neurons are connected, they form a network, therefore named a neural network.

Neural networks are in essence algorithms that can identify complex relationships within a dataset. Their key advantage over traditional machine learning is that they can find some really deep patterns due to the many neurons that are layered to create one model. To illustrate how a deep learning model is built, we can imagine the following analogy: 

1. Creating a neural network is like teaching a toddler to identify a cat. A parent would show the toddler thousands of pictures of cats and “non-cats”. 
2. The toddler is not answered initially but would guess as to whether the picture is showing a cat. The parent would then reveal the answer to the toddler. The toddler then learns from either his mistakes or successes to recognise features common to cats.
3. By the end, the parent would hope that the toddler will be able to identify a cat when given a novel image.

This whole process is done through creating levels of abstraction, using knowledge from the preceding level of abstraction to form a whole concept. This is what a computer is doing through deep learning but on a larger scale. The process is automated and requires no human input apart from the data which the model is learning from. Its effectiveness and efficiency have given the method incredible scalability that has demonstrated widespread use in almost every aspect of modern technology

Further Reading/References:

Iqbal Sarker (2022) ‘AI-Based Modelling : Techniques, Applications and Research Issues Towards Automation, Intelligent and Smart Systems’

Pfizer (2024) ‘mRNA and Artificial Intelligence for Advanced Vaccine Innovation’.

Jong Taek Kim (2018) ‘Application of Machine and Deep Learning Algorithms in Intelligent Clinical Decision Support Systems in Healthcare’

Written by : Nûr-al-ayn Nisar (FY1)

Edited by : Sun Lin (IMT)