Artificial intelligence examining medical images can correctly identify diseases with greater speed and accuracy than doctors, raising the possibility that automated diagnosis may soon be common in parts of the world where specialists are scarce.
Using medical images, deep learning systems correctly identified a disease 87% of the time, compared with 86% for healthcare professionals, according to a review of research published in the British medical journal Lancet. AI correctly cleared patients 93% of the time, while experts did so only 91%. The images analyzed were drawn from 14 studies — narrowed from more than 20,000 published since 2012, a milestone year in the field of deep learning.
“Diagnosis of disease using deep learning algorithms holds enormous potential,” the review’s coauthors said. They added that their review revealed the limits of current technology and a lack of research directly comparing humans vs. machines in clinical practice.
Experts have stressed the need to temper expectations over AI’s use in medicine and questioned the value that it adds to clinical practice, the Guardian reported. Failure to understand why deep learning algorithms mis-diagnose certain cases is critical to the technology’s success in diagnosis and treatment.
- Applications for artificial intelligence in the healthcare industry are rapidly gaining pace, with analysts at Gartner predicting that 75% of healthcare service providers will be using AI to improve clinical or operational outcomes by 2021.
- Startups developing AI imaging and diagnostics solutions represent 29% of the best funded healthcare AI startups and raised a combined $1.5 billion in financing, in an analysis by CB insights.
- San Francisco-based Freenome has raised $238.35 million in disclosed equity funding, with investors that include Andreessen Horowitz, Google Ventures, and Founders Fund. Boston-based Path AI raised $74.94 million in disclosed equity funding, with investment from LapCorp in July.
- China and U.K.-based healthcare startups have benefited from government efforts, driving investment and forging public-private partnerships. U.K. companies benefit from a broad library of anonymized patient data, which citizens share with the national healthcare service.
- “Matter with mind; a neurological research robot”, the first paper on AI recorded in the U.S. National Library of Medicine, dates to 1951. Today, more than 16,000 papers are published each year.