Imagine a future where you could predict, using advanced machine learning algorithms, nearly a day before a patient may have a heart attack. It's the stuff of science fiction! Actually, it's something we're able to do now thanks to the Covid-Heart predictor developed by Johns Hopkins University in the US.

Of the many uses we talk about for machine learning (ML) on the Aito blog, healthcare is one we don't often mention. Yet, great strides are being made, where ML is pushing the possibilities of what modern medicine can achieve. What once was considered science fiction is now being used across various types of medical research. Let's look at that Covid-Heart predictor, for instance.

The Covid-Heart predictor tool

The Covid-Heart machine-learning algorithm was built using more than 100 clinical data points collated by a student in the biomedical engineering department of Johns Hopkins University. These included patient demographic data, as well as lab results from a registry set up to collect Covid-19 data from every one of five Johns Hopkins Health System hospitals. Variables were added along the way to help evolve the model. This was during the early phases of Covid-19, when little was known about the virus.

Photo by Laura Adai

We now know that Covid-19 affects the heart as well as our respiratory system. This is where the Covid-Heart predictor stretches its machine-learning muscles. The ML algorithm can predict blood clotting three days in advance. This is just remarkable! Thanks to the input of electrocardiogram data, Covid-Heart became an accurate source for indicating which patients were most at risk of clotting.

The challenge begins for how to effectively roll out the Covid-Heart predictor tool across many hospitals, and make best use of the machine learning technology at our disposal. Yet, Covid-19 isn't the only area of medical concern where ML is playing a vital role in new research and development.

The necessity of reliable data

A new scientific article reveals how machine learning is helping to predict lymph node metastasis (LNM) in patients with poorly differentiated-type intramucosal gastric cancer. If you're like me, most of the medical vocabulary goes way over my head, but the article outlines other areas where ML has been used across numerous fields of medical research, such as hospital-acquired pneumonia and the survival rates of patients with chondrosarcoma (a form of cancer).

However, as with any decent ML algorithm, reliable data is crucial. Medical data, as with any other, is open to error. Adding incomplete or incorrect data to an algorithm created to help patients is a scenario none of us want to encourage. There are many things to consider when engineering an effective and reliable ML algorithm, such as how it will interface with other systems, how accurate the information is, what biases are present, and what sources of expertise are being relied upon.

This is why your machine learning journey – whether that's in medicine, supply chain, finance or any other sector – should be robustly planned so that implementation of an algorithm is as error-free as can be, with oversight in place to monitor ongoing validation.

There's so much potential for machine learning to be a game-changer in medical research and discovery, and the development of new techniques to improve our health. With so many factors to take into consideration by virtue of medicine being incredibly complex, ML has the potential to empower clinical decision makers that will benefit us all.