Oncologists lost to the computer

A computer program better than doctors diagnoses brain cancer from MRI images

marks, Geektimes

Magnetic resonance imaging (MRI) is a tomographic method of examining internal organs and tissues using the phenomenon of nuclear magnetic resonance. MRI images have become familiar to ordinary people thanks to a large number of medical series, where "doctors" look at such images with a serious look, instantly establishing the correct diagnosis.

In fact, everything is much more complicated. Even the most trained medic can make a mistake. It is especially difficult to diagnose brain cancer using MRI scans. A person in general often makes mistakes, and such mistakes often lead to sad consequences. Well, what if we involve computer systems in the work of doctors? After all, they have learned to make diagnoses based on the descriptions of diseases and the results of patient tests. Scientists from Case Western Reserve University decided to test the abilities of computer systems in terms of diagnosing oncological diseases using MRI brain scans.

As it turned out, not in vain. The program developed within the framework of the project makes a diagnosis more accurately than a human doctor. The system can determine, for example, what is an unusual formation in the brain of a patient who has previously been diagnosed with cancer. Is this site a group of dead cells killed by radiation, or is it cancer that has returned? The computer, after a thorough analysis of the image, can determine all this.

"One of the most pressing issues for medicine is planning a possible treatment method, if the patient has already been diagnosed with cancer, and now it is necessary to determine whether the cells have died after radiation treatment or the tumor has not gone away," says Pallavi Tiwari, one of the developers of this system. "On an MRI, it all looks almost the same."

But the treatment of radiation necrosis and cancer are radically different. This is the problem – if you make a mistake, the patient will receive the wrong treatment that he needs, and the situation may worsen significantly. It is possible to distinguish necrosis from a tumor, but a biopsy is needed for this. And this is expensive, and it takes a lot of time to analyze. Plus, a biopsy is an invasive operation, which can also have a negative impact on the development of the patient's disease.

To develop the program, the researchers used a machine learning methodology. Scientists used MRI images, according to which doctors had previously correctly diagnosed, uploading these images to the neural network. Not only doctors take part in the project, engineers, scientists of other fields, physicists are also working on this system. To train the system, doctors used images of 43 University Hospitals Case Medical Center patients.

The team managed to develop algorithms that are able to distinguish between both types of deviations and make the correct diagnosis. "Algorithms see things that doctors simply cannot see. The computer system conducts a large number of measurements of images, trying to determine the presence of a tumor or radiation necrosis of brain tissue," says another participant in the experiment. Malignant tumors and the consequences of radiation necrosis still differ, but these differences are so small that they are almost impossible to recognize with the naked eye.

MRI images of the brain of a person with radiation necrosis (above) and with a recurrence of a brain tumor (below). The structure of the tumor is more heterogeneous (red) than the structure of radiation necrosis. Figure from the article by Tiwari et al. Computer-Extracted Texture Features to Distinguish Cerebral Radionecrosis from Recurrent Brain Tumors on Multiparametric MRI: A Feasibility Study (American Journal of Neuroradiology, 2016) – VM. And if doctors are trying to find heterogeneities in the structure of the studied tissues, then the computer does not only that.

He studies the images as thoroughly as possible, analyzing the images by pixels. The structure of the tumor in the images looks more homogeneous, although the differences with the consequences of radiation necrosis are minimal, as mentioned above.

To test the efficiency of the new method, the authors of the project decided to use MRI images of patients who had been correctly diagnosed. Human doctors and the computer system tried to identify the pictures. A total of 15 images were selected. One of the doctors made the correct diagnosis from the pictures of 7 patients. The other is eight. The computer program correctly diagnosed 12 cases out of 15 from the same images.

The developers of the system say that when it was created, it was conceived as an additional diagnostic tool. It can be used in conjunction with other diagnostic methods, which will radically improve the accuracy of diagnosis for patients with malignant brain tumors and for patients with radiation necrosis of brain tissue.

Now the development exists as a prototype. Scientists are constantly refining and supplementing it, hoping to make it even more accurate. To do this, scientists upload into the system a large number of MRI images with a previously confirmed diagnosis of a number of patients from various hospitals. After the methodology is finalized, according to the developers, it can be used as an additional diagnostic method in clinical settings.

Portal "Eternal youth" http://vechnayamolodost.ru  19.09.2016