AI demonstrates human-level capacity to diagnose disease

GlobalData Healthcare 1 October 2019 (Last Updated October 1st, 2019 15:35)

Deep learning diagnostics holds immense potential to enhance both the speed and accuracy of disease diagnoses.

AI demonstrates human-level capacity to diagnose disease

A joint research effort, published on 24 September in the journal Lancet Digital Health, conducted by researchers in the UK, Germany, Switzerland and the US suggest that image-based artificial intelligence (AI) disease diagnosis performance is comparable to healthcare professionals. 

Deep learning diagnostics

The study analysed the results of papers published between 2012 and 2019 that directly compared the diagnostic accuracy of deep learning models to healthcare professionals and found the accuracy rate for AI diagnoses was 87% compared to 86.4% for healthcare professionals.

Deep learning diagnostics holds immense potential to enhance both the speed and accuracy of disease diagnoses. The US Food and Drug Administration (FDA) has already approved several products in this space, such as IDx-DR (IDx Technologies Inc.), which detects diabetic retinopathy through deep learning methods, and the Viz.ai Contact application (Viz.ai, Inc.), which can diagnose large vessel occlusion strokes. AI-based innovations bring the promise of high levels of diagnostic accuracy and speed, which in the case of stroke can be vital to positive patient outcomes.

While the authors of the meta-analysis study indicate that AI-based frameworks can perform on par with professionals in the field, they emphasised the lack of clinically relevant evaluation and direct head-to-head comparisons between AI and professionals throughout the vast majority of the literature. 

However, the authors also note that the most recently published papers in the field tended to include both of these evaluation criteria. Growing confidence in AI-based diagnostics is expected to drive new competitors into the space to enhance disease diagnostic capabilities. The potential of standardised accuracy and cost-saving measures of AI-based disease diagnostic frameworks will drive further adoption of these products.