Editor's Note
When deep learning algorithms were compared with health-care professionals in classifying diseases using medical imaging, diagnostic performance was equivalent between the two.
In this meta-analysis of 14 studies, researchers found that deep learning systems correctly detected a disease state 87% of the time, compared with 86.4% for healthcare professionals, and the systems gave an all-clear 92.5% of the time, compared with 90.5% for human experts.
Diagnosis of disease using deep learning algorithms holds enormous potential, the authors say. However, they point out that there are key issues in the design and reporting in present research. These issues are pertinent for ensuring studies of deep learning diagnostics are of sufficient quality to evaluate the algorithms' performance.
