The experiment to put the Mammogram analysis process ”in the hands” of a computer algorithm was successful. Artificial intelligence turned out to be more efficient at researching breast X-rays than a single radiologist, and as effective as a team of two doctors.
Artificial intelligence will support radiologists at work? Cancer diagnosis
The ”Nature” magazine described a research conducted jointly by Google Health, Google’s subsidiary Deep Mind (creator of an artificial neural network), the Northwestern University from US and Cancer Research UK Imperial Centre (CRUK), as well as the local state hospital of the Surrey county.
We have decided to see if artificial intelligence can support radiologists in more accurate analyzing of radiological test results. At the first try, our model was trained and configured to work on a set of anonymous data of 76 thousand patients from UK and 15 thousand from America. The results were evaluated based on the second set of anonymous data about 25 thousand British women and 3 thousand patients from USA – informs Google in a press release.
The results turned out to be promising. The AI algorithm reduced (compared to the analysis conducted by radiologists) the number of false positives, or false alarms (when disease is mistakenly detected in a Healthy body) among American patients by 5.7 percent and by 1.2 percent among British patients.
Artificial intelligence provided analysis (Cancer diagnosis) that contained 9.4 percent less false negatives (when a sick body is mistakenly determined to be Healthy) in the results from the United States and 2.7 percent when it came to the results of UK patients.
We wanted to see the degree to which the algorithm we created is flexible and can work on data from the Health care system of any country. That’s why during the next try we trained the AI model on data from UK so that it could later analyze data from USA. This separate experiment witnessed a reduction of false positives by 3.5 percent and of false negatives by 8.1 percent. It means that our computer model is capable of working on new data sets while maintaining higher efficiency than human experts, claims Google.
The algorithm was more efficient at reading X-rays than any of the six radiologists participating in the test. On the other hand, it was just as efficient as two radiologists reading the same Mammogram. Whereas in the British public Health care, for example, this exact form of double verification of patients is the norm, the problem is weariness.
Efficiency and effectiveness of artificial intelligence
An AI algorithm can work around the clock. On top of that, it maintains effectiveness of a two-man team of medics in a situation where there is no access to the patient’s history or earlier Mammogram results (which was the assumption of the experiment). Meanwhile, British hospitals are desperate to find at least 1,000 radiology specialists. The existing ones are overworked and would be happy to welcome additional support from artificial intelligence computers.
There is no point expecting, however, that AI could replace people in a hospital. After all, it is people who have written the algorithm and taught it to work on a specific data set. There is also no permission to authorize a computer program to make independent, final evaluations of people’s Health.
As noted by BBC, even in case of allowing AI to help with diagnostics, the supervision and the final verdict would be up to humans. It doesn’t change the fact that an algorithm could take some load off the shoulders of radiologists and enable them to do some equally important work at a ”different section”. The initial analysis could be performed within seconds from conducting the tests.
Those are pretty promising yet early test results. They suggest that in the future it could be possible to conduct screening faster and more accurately. This means faster service and less worries for the patient – says Sara Hiom, chief of the diagnostic department at CRUK.
I will never forget my first computer. I was interested in new possibilities just as much as the computer itself. This fascination was so strong that I became NERD in a second 🙂 I am an IT technician by profession. Before I started my own laptop repair service, I gained experience for 8 years. My friends keep asking me:
„Hey Joshua, which computer is better for gaming?”, „I can't pick the graphics card, help me!” , „What do you think about off lease computers?” e.t.c.
That's why I got used to giving advice. And just to be honest - I like it very much. So I decided to share my knowledge with other people in need.