Until recently, machines only made better-than-mechanical jobs better than mechanical jobs for which great skill was not required. But that is starting to change. This week, the journal Nature Medicine presents the results of an experiment in which an algorithm capable of learning achieves as good or even better than several radiologists in predicting a person’s risk of lung cancer after seeing images of the patient obtained by tomography.
Lung cancer kills more than one million people a year in the world and is the most lethal tumor among men and the second among women after the breast. Screening with low-dose tomography allows the reduction of mortality by this type of tumor, making early treatment possible, but there are still problems related to the false positives and negatives produced by this method.
To attempt to improve these results, Google and Northwestern University in Illinois (USA) proposed an artificial intelligence system that used current and past tomography imaging of patients to predict their risk of developing lung cancer. Using almost 8,000 cases, they compared their analyzes with those of six radiologists. When there were no previous CT scans of the patients and only one was present, the machine performed better than any of the doctors, with a reduction of 11% in the number of false positives and a 5% reduction in false negatives. When there were previous images of the patients, the computer model equaled the results of the radiologists.