Google Cloud Research Says that AI Won’t Replace Radiologists…Yet

The medical community has viewed the advent of artificial intelligence (AI) with both fear and excitement. The technology held a major platform at this year’s Radiological Society of North America conference, and companies like GE Healthcare, Siemens, and NVIDIA are rapidly rolling out AI-based devices. This onset of machine and deep learning programs has threatened to replace human radiologists. However, according to research from Google Cloud, AI is capable of only automating a portion of a radiologist’s responsibilities.

Google Cloud’s head of research and development, Jai Li, studied how machine learning can apply to disease diagnosis in clinical environments where physicians lack data points and resources. “We want to build an AI-powered tool to assist [radiologists’] diagnoses, but in order to do so, traditionally, we’ll need a large amount of labeled data. That goes back to the exact problem that we want to solve. Put back the burden on our radiologists again to label a large amount of data,” she told the Massachusetts Institute of Technology’s EmTech.

Through machine learning, Li and her team analyzed abnormalities in chest x-rays using a supplementary data set. They found that AI could only diagnose a limited amount of patient data. Machine learning systems still need to digest more comprehensive treatment, diagnosis, and health history components in order to produce effective assessments.

Li’s work could be the beginning of Google Cloud’s medical technology program, meaning the Cloud’s design could ultimately be a respite for clinicians who lack AI training.