Artificial Intelligence is on a Par with Human Experts in Medical Diagnoses

A latest published report found that Artificial Intelligence can at best be equivalent to human experts when it comes to making image-based medical diagnoses. Today, AI research within medicine is growing rapidly and it has the potential to transform the medical research from the bottom. Experts say it will ease the strain on resources, free up time for physician-patient interactions as well as aid in tailored diagnoses.

Since the sector is overwhelmed by impecunious research, the latest findings are based on a few numbers of studies. And the researchers found one of the burgeoning applications is the use of AI in medical images, a field relies on Deep Learning, wherein a series of labelled images are fed into algorithms that pull-out features within them and learn how to classify similar images.

This approach has shown promise in disease treatment, from eye conditions to cancer. However, responding to questions on how such deep learning systems measure up to human skills, researchers conducted the first comprehensive review of published studies on the issue that revealed humans and machine are on a par.

According to the University Hospitals Birmingham NHS foundation trust and a co-author of the study, Prof Alastair Denniston, the results were encouraging but the study was a reality check for some of the hype about AI.

In their writing to the Lancet Digital Health, Denniston, Dr Xiaoxuan Liu, the lead author of the study, and colleagues noted how they focused on research papers published since 2012, an essential year for deep learning. In over 20,000 germane studies, only 14 studies, which all based on human disease, reported good quality data, tested the deep learning system with images from a distinct dataset to the one used to train it, and demonstrated the same images to human experts.

Afterward, the team pooled up the most promising results from each of the 14 studies that found deep learning systems correctly detected a disease state 87% of the time, compared with 86% for healthcare professionals, as well as correctly gave the all-clear 93% of the time that outperformed the human experts with 91%.

With these outcomes, Denniston is very optimistic about the potential of AI in healthcare, with saying such deep learning systems could perform as a diagnostic tool and assist in tackling the backlog of scans and images. In the same context, Liu said such systems could prove useful in places which lack experts to interpret images. It would be significant to leverage deep learning systems in clinical trials to weigh, whether patient outcomes improved compared with current practices.