Net for catching cancer

Convolutional neural network determines the types of cancer cells

Sergey Syrov, XX2 century

Malignant tumors differ from each other and even within the same diagnosis may contain different types of cancer cells. Identifying specific cell types for each case of the disease can significantly improve the effectiveness of treatment. The problem is that it is long and difficult to do this – you have to rely on imperfect human vision and put up with a large share of mistakes.

The development of AI systems may open a new era of cancer diagnosis and treatment. A scientific group from Osaka University showed (an article about the work was published in the journal Cancer Research) how existing problems can be overcome with the help of computer technology.

Scientists have built a system capable of detecting various types of cancer cells by simply scanning images obtained using micrography with an accuracy exceeding human capabilities.

Cancer cells under the microscope. Images that were used to train a convolutional neural network.

The basis of the system is a convolutional neural network, a form of artificial intelligence modeled on the principles on which animal vision is built. It is reported that AI, after training, is able to determine cancer cells and their resistance to radiation exposure.

"First, we trained our system on eight thousand images of cells obtained using a phase contrast microscope," says Hideshi Ishii, the author of the publication. "Then we tested the accuracy of her work on another two thousand images to make sure whether she captures the features that distinguish mouse cancer cells from human, and radioresistant cancer cells from radiosensitive."

It is expected that the development and implementation of such systems in clinical practice will increase the chances of patients to recover. It will be possible not to waste the time and resources of the patient's body on obviously ineffective procedures.

Work on the system continues – scientists hope to train it on more types of cancer cells. The ultimate goal is to create a universal system that will automatically identify and distinguish all pathological cells.

Portal "Eternal youth" http://vechnayamolodost.ru