Scientists from Google and the North American University in the Northwest have developed a system of AI able to detect lung cancer better than radiologists, which could favor the early diagnosis of a tumor that kills more than one million people in the world every year.
His description is published in the latest issue of the journal Nature Medicine and shows the "accuracy" of the new system of deep learning to predict this type of cancer, according to its authors, who nevertheless warn that these findings should be clinically confirmed even in large amounts of patients.
on "deep learning" or deep learning is a branch of artificial intelligence by which the machine learns from examples and develops increasingly complex models that simulate the functioning of the brain.
Based on this, American scientists developed a an algorithm capable of detecting malignant pulmonary nodules, sometimes small, from the chest CT (computed axial tomography) results, with commitment and precision equal to or better than that of radiologists.
To do so, they "trained" the system of 42,290 tomographic images and found that The Artificial Intelligence System is able to detect modules with 94% accuracy in 6,716 test cases,
on The model is compared to the tests made by six radiologists both when they had images from TAC previous when not, and in both cases the machine was superior to the expert radiologists,
In addition, the deep-learning system also produces less false positive results and less false-negative tumors, scientists say in a press release from the American University.
Mozziyar Etemadi, a professor of medicine and engineering at Northwestern and one of the authors of this article, explained that radiologists usually study hundreds of 2D images of a CT, but this new system allows us to analyze 3D photos,
"Artificial 3D intelligence can be much more sensitive in its ability to detect early lung cancer than the human eye looking at two-dimensional images," he sums up.
However, he pointed out that, from a technical point of view, we can talk about four dimensions, as this is not just a CT scan, but two taken at different times (current and past patient tomograms).
To build artificial intelligence to see TAC in this way, "you need a huge computer system on Google," says Etemadi, who notes that "the concept is new, but its engineering is also on a scale."
Shrava Shetty, Google AI, said in turn that "this area of research is extremely important, as lung cancer has the highest mortality among all types of cancer."
The present system examines how artificial intelligence can be used to improve and optimize the low-dose radiation dose screening process, adds this expert who says "the results are promising and we hope to continue our work with our partners and partners ".
And that is, notes the note of the American University thoracic screening serves to identify the tumor and reduce mortality, but also "high levels of error" appear.
In addition, "limited access to these medical tests means that many lung tumors are usually found at an advanced stage when they are difficult to treat."
Now, this new system is able to identify both a region of interest and whether the region has a high probability lung cancer".
"The system can categorize the lesion with more specificity, not only can we better diagnose a person with cancer, we can also say if someone does not have it, and save it from an invasive, expensive and risky lung biopsy," he summed Etemadis.