Select Page

In May of 2019, Google teamed up with Northwestern Medicine to conduct a unique medical experiment. Researchers applied a deep-learning algorithm to the CT scans of 42,290 patient’s lungs. The goal was to see how well the algorithm could predict a person’s potential to develop lung cancer.

 

CT scans are normally examined one at a time by highly-trained radiologists. A joint team of Google and Northwestern specialists cobbled together a special machine-learning model to read the images instead. Then they compared the results with that of the six experienced radiologists.

 

The result showed that the machine-learning model was able to detect cancer 5% more often than the human team. Furthermore, the algorithm was 11% less likely to deliver a false positive.

 

The results have generated tremendous excitement in the medical community. It showed the potential held by large-scale pattern recognition in creating predictive diagnostic models when handled by technological means rather than a human-only process.

 

While 42,290 CT scans may seem like a large number, the sampling is actually quite small in terms of what would be handled by AI-assisted (artificial intelligence) algorithms. Furthermore, the larger the sample size available in any specific diagnostic category would bring the added advantage of creating an enormous database through which greater connections could be drawn between current data and previously collected data.

 

The result would be an automated system that would get better over time at finding the potential for disease faster and more accurately. That would translate to many more lives saved while also reducing the danger and enormous expense of surgery and other complicated procedures.

 

Another company, Atomwise, is using supercomputers that can pour through massive databases that ferret out suggested therapies for diseases based on what is found at the molecular level. Atomwise is now conducting a virtual search for existing prescription medicines that could be re-tasked to treat Ebola. An AI-assisted machine-learning approach has already identified two drugs that may reduce the infective potential of Ebola considerably.

 

AI-enabled machine-learning leveraging innovative algorithms may be the key to finding a cure or vaccine for COVID-19 much earlier than otherwise thought possible, researchers said.

 

The same potential exists for all types of diseases. In the near future, diagnosing disease using machine-learning AI systems will play a major role in medicine.