The Hype about AI in Healthcare: An Introduction to Machine Learning

Updated: Mar 25

Artificial intelligence is a hot topic of discussion in healthcare. While it does have dystopian connotations it is expected to have huge impacts in the coming years across many industries that deal with large datasets - one of which is healthcare.

In this educational piece, we’ll be covering an exciting derivative of artificial intelligence: machine learning.

What is artificial intelligence?

Artificial intelligence (AI) is simply a computer capable of human-like intelligence. Within that, machine learning (ML) is one of the AI domains. ML involves computers learning by pattern recognition without being explicitly programmed to do so. ML has further subtypes such as ‘neural networks’ and ‘deep learning' and it is an active area of research in computer science and many fields of its application.

Programmed algorithms are already used to make decisions in healthcare from simple triaging to giving life-changing diagnoses. These rule-based algorithms are an integral part of healthcare delivery across all medical specialities in some form. They help clinicians make decisions for the delivery of safe and effective healthcare.

Radiology, dermatology, ophthalmology, cardiology, and histopathology are some of the medical specialties most expected to be assisted by ML due to elements of pattern recognition and the availability of large datasets. AI healthcare research is already published on medical imaging, endoscopy, electrocardiograms (ECGs) and vital signs.

How does ML work?

ML works by developing a ‘trained’ algorithm that can sort data. The data can be in any form (e.g. images, text and videos). The ‘untrained’ ML algorithm is initially provided with a defined and sorted training dataset that it can learn the meaning of data from. For example, feeding an untrained ML algorithm a defined data set of skin lesions that outputs a trained algorithm that can sort the images into ‘cancer’ and ‘not cancer’. The ML algorithm learns these patterns and provides a diagnosis accordingly. It is difficult to know how the algorithm made a specific decision due to its nature of not being explicitly programmed and thus sometimes being referred to as a ‘black box’ decision-maker. However, there is ongoing research to help improve interpretability for regulatory and evaluation purposes.

ML systems learn by several different methods depending on the task needed to be performed. There are three main types of tasks needed in healthcare: 1) classification (sorting data to a defined set of classes), 2) regression (predicting numerical outputs based on observed patterns) and 3) clustering (finding classes within data) [1].

Gradient descent is the most common algorithm learning method applied in healthcare for training models on large datasets. It is a part of supervised learning where the ML algorithm has a ‘teacher’, and we know what the outputs are going to be. In contrast, unsupervised learning gives the algorithm the freedom to find patterns within the dataset without known outputs.

Gradient descent works by making a random guess and then quantifying that error from the true value and then gradually reducing the error to find the best-fitting model by taking more targeted guesses.

Applications of machine learning in the NHS


Addenbrooke’s hospital in Cambridge became one of the first hospitals in the world to use machine learning to predict the oxygen needs of COVID-19 patients.

Cancer diagnosis

ML subset: Neural networks

Neural networks, also known as artificial neural networks (ANNs) are ML algorithms that are designed as a circuit of neurons represented as nodes. They work by first mapping nodes in a given dataset by their respective weights given by the strength of connections within that dataset. Like a neural pathway, it will follow from input to output by multiplying the input with weights and nodes and then summing this up to give an output accordingly which is transformed by an activation function; this is simply a way to interpret the output [2]. Deep learning involves neural networks with multiple node layers and there are two main types of these algorithms. Convolutional neural networks (CNNs) excel at medical imaging and is the most researched. Recurrent neural networks (RNNs) excel at sequential data [3].

ML outputs then need to be assessed for performance and clinical significance. Research papers evaluate this using a plot of true-positive and false-positive rates of the trained ML algorithm against doctors’ assessments. The area under the receiver operating characteristic (ROC) curve plot is used to express the level of accuracy which is widely used in medical research. However, AI healthcare randomised controlled trials (RCTs) make up a very tiny proportion of the research papers published to date, with most AI healthcare research being on retrospective and prospective studies.

There are ethical concerns regarding the use of ML as algorithms are known to be biased and greater care needs to be taken to ensure this is properly addressed before clinical implementation [4]. Guidelines are being developed to ensure safe research and clinical adoption of ML systems [5]. As more and more researchers take this on, the field will carry on advancing and holds the potential for data-driven medicine that is evidence-based.

Editor's key takeaways: Machine Learning

Machine learning has been around for a long time. In recent times companies across many industries are using countless ML models for a range of purposes. This is thanks to standardised cloud-based ML platforms.

ML is rapidly becoming an important method across biomedical discovery, clinical research and medical diagnostics/devices.

ML approaches can automatically analyse data inputs and learn the structure already resident in that data in order to provide a solution. This can help clinicians identify, diagnose and treat disease.

Such tools can uncover exciting new possibilities for researchers, doctors and patients, allowing them to make more informed decisions and achieve better outcomes.

However, it must be acknowledged that ML is only as effective as the data we’re able to access and analyse. Therefore, advancements in big data and analytics will play a key role in how useful ML can be for healthcare delivery.

Like other forms of medical intervention, AI technologies can cause harm to patients; this means ethical judgment and use of policies (such as NHS Code of Conduct for data-driven health and care technology) are essential to ensure safe use of AI technologies.

While ML does have its limitations, it is moving beyond the hype and is becoming a promising tool for the improvement of patient care. In future, it will have a transformative impact in healthcare as the field continues to make breakthroughs.

References and resources



Kalp Patel

Content Writer

Kalp is a fourth-year medical student at the University of Nottingham. He previously completed the NHS Scientist Training Programme in Clinical Engineering and Medical Physics and has worked in the NHS for five years as a gastrointestinal physiology clinical scientist. Interested in medical technology, AI/ML, workflow optimisation, academic research, and surgery. Kalp is a Content Writer at MedTech Foundation.

Edward Mbanasor


Edward is an intercalating fourth-year medical student at the University of Leeds currently studying MSc Enterprise and Entrepreneurship. He has a strong interest in healthcare entrepreneurship, digital healthcare and digital transformation. Edward is the National Content Director at MedTech Foundation.

38 views0 comments

Recent Posts

See All