AI displays potential to detect diseases with impressive accuracy


Published Date : Oct 01, 2019

Artificial intelligence has the potential at par with humans to detect diseases by means of medical images, as revealed following investigation carried out by a team of researchers. The accuracy of machines to be at par with humans is postulated following systematic understanding of published studies on the topic.

Medical imaging is thus far considered the most valuable means for diagnosis of medical conditions, explained the researchers. However, the demand for diagnostic imaging outpaces the bandwidth of available specialists, especially in under-developed and developing countries.

In such scenarios, deep learning holds immense promise for medical diagnostics, stated the authors of the study. To validate this, researchers aimed to gauge the diagnostic precision of deep learning algorithms versus health-care professionals to classify diseases employing medical imaging.

Whilst small in absolute numbers, percentage accuracy of deep learning algorithms favorable

Meanwhile, under such attempts, only 14 deep learning based algorithms produced satisfactory results for classification of diseases using medical imaging. This is out of more than 20,000 studies that involved artificial intelligence applications to classify diseases.

Collectively, results of 14 studies suggest that deep learning algorithms are correct eighty seven percent time to detect a disease, in comparison to eighty six percent of time correctly detected by healthcare professionals.

One of the authors of the study who is an eye specialist at University Hospitals Birmingham, explained the study is a step forward for reality check of hype about artificial intelligence.

Earlier, several headlines of AI outperforming humans made a bang. But it can only be considered equivalent, stated lead author of the study.

This is based on review of few high-quality studies to evaluate the performance of AI in real clinic settings. This had good chances of change of results of meta-analysis to prevent real analysis of performance of AI. Due to such limitations, real diagnostic potential of AI remains uncertain.