Artificial intelligence has been of little use for diagnosing covid-19

IS THERE no issue artificial intelligence can’t tackle? Methods such as deep learning have found uses in from algorithms that recommend what you ought to purchase next to ones that predict someone’s voting habits. The effect is that AI has developed a somewhat mystical reputation as an instrument that can digest many types of data and accurately predict many different outcomes, an ability that may be of particular use for solving previously impenetrable problems within healthcare.

However, AI is no panacea. Too often, it is considered too quickly and within an impulsive way, leading to claims that it works when it doesn’t. It has become increasingly apparent through the covid-19 pandemic, as many AI researchers try their hand at healthcare – without much success.

It really is no wonder many persons think healthcare is a promising area for AI as hospitals generate plenty of data, which deep learning depends on. The partnership has recently borne fruit, with AI systems in a position to help identify cancer earlier and better predict which treatments persons will react to.

In the original stages of the pandemic, there is a deluge of publications wanting to do the same for covid-19. Specifically, there are a huge selection of papers claiming that machine-learning techniques may use chest scans to quickly diagnose covid-19 also to accurately predict how patients will fare.

My colleagues and I viewed every such paper that was published between 1 January 2020 and 3 October 2020 and found that do not require produced tools that might be sufficient to use in a clinical setting (Nature Machine Intelligence, doi.org/gjkjvw). Something has truly gone seriously wrong when a lot more than 300 papers are published that contain no practical benefit.

Our review discovered that there were often issues at every stage of the development of the tools mentioned in the literature. The papers themselves often didn’t include enough detail to replicate their results.

Another issue was that many of the papers introduced significant biases with the info collection method, the development of the machine-learning system or the analysis of the results. For instance, a substantial proportion of systems made to diagnose covid-19 from chest X-rays were trained on adults with covid-19 and children without it, so their algorithms were much more likely to be detecting whether an X-ray originated from an adult or a child than if see your face had covid-19.

Though authors might have been motivated by the desire to build up models that could help people, in their haste, many of the publications didn’t consider how, or whether, these models could pass regulation requirements to be used in practice.

The papers also suffer from publication bias towards excellent results. For instance, imagine a theoretical research group that carefully develops a machine-learning model to predict covid-19 from a chest X-ray and it finds that doesn’t outperform standard tests for the condition. This finding isn’t of interest to numerous journals and is hard to communicate. It is far easier to create a model with poor rigour that gives excellent performance and publish this.

While machine learning has great promise to find solutions for many healthcare problems, it must be done just as carefully as whenever we develop other tools in healthcare.

If we take as much care in developing machine-learning models as we do with clinical trials, there is no reason why these algorithms won’t become part of routine clinical use and help us all push towards the perfect of more personalised treatment pathways. But there is no rushing that.

More on these topics:

  • artificial intelligence
  • covid-19

IS THERE no issue artificial intelligence can’t tackle? Methods such as deep learning have found uses in from algorithms that recommend what you ought to purchase next to ones that predict someone’s voting habits. The effect is that AI has developed a somewhat mystical reputation as an instrument that can digest many types of data…

IS THERE no issue artificial intelligence can’t tackle? Methods such as deep learning have found uses in from algorithms that recommend what you ought to purchase next to ones that predict someone’s voting habits. The effect is that AI has developed a somewhat mystical reputation as an instrument that can digest many types of data…

Leave a Reply

Your email address will not be published.