Diagnosis of the disease is no less important than its treatment. But often for various reasons, it can be difficult for a number of additional factors. And if in the case of states that are not particularly life-threatening, the delay will not give great complications, then diseases like oncology need to be diagnosed as quickly as possible. And to improve the detection of diseases, Google experts suggest using machine learning algorithms.
Google employees came to the rescue by deciding that, according to statistics, in the presence of patients with certain forms of breast and prostate cancer, the opinions of doctors can be only 48%. To solve this problem, it was proposed to use the "help AI". To create the program, researchers used images of organs affected by oncology, provided by the Medical Center of the University of Nijmegen. For these purposes, an already existing algorithm called Inception (or GoogLeNet) worked well.
During the training, the system was “fed” a large number of heatmaps, that is, images showing which parts of the tissue might be affected. After some modification of the algorithm, the heat maps prepared with its help were improved to such an extent that their accuracy reached 89% in comparison with the original images containing a large amount of “noise”. Further, it was decided to compare the results of the work and the conclusions of Inception with the conclusions that the doctors would give. As a result, after analyzing 130 images, it turned out that Inception is 16% more effective than a person determines the lesion areas.
Despite good results, the authors of the algorithm claim that the model is still far from perfect. She does not pay attention to the presence of comorbidities, which can indirectly indicate the presence of the disease. But even so, if we supplement the work of a person with a diagnostic program, this can give extremely positive results and increase the number of identified diseases that could have been missed.
The article is based on materials .
Comments
Post a Comment