Still appears to work better in combination with radiologists

Researchers from Google and medical systems in the US and Britain may be providing clarity on a path forward for radiology interpretations being supplemented by AI, and vice versa. A paper recently published in the journal Nature is the result of one of Google’s ventures into medicine. Researchers have trained computers to recognize patterns and interpret images, and tested on images with known diagnoses. The A.I. system performed better than radiologists. A.I. improves outcomes in mammograms: the researchers saw reductions in both false negatives (where the image is mistakenly read as normal and cancer is missed) and false positives (where it is mistakenly read abnormal but there is no cancer).

Curiously, the paper demonstrated the AI system created an improvement that was four times greater among American radiologists than British radiologists.

While mammograms are commonly ordered to identify breast cancer, this first line imaging (when done by humans) still misses about twenty percent of cancers. AI, however, does not have a perfect record either. The researchers discovered that while AI picked up cancers that radiologists missed, radiologists picked up cancers the AI missed. So, perhaps the best use for the technology now and in the near future is to use it in conjunction with radiologists, each complimenting the other.

Read the study here about how A.I. improves outcomes in mammograms: International evaluation of an AI system for breast cancer screening

You can read some of our prior coverage regarding AI in health care here, here, here, and here.