The neural network has learned to recognize a brain tumor from MRI images Person

The neural network has learned to recognize a brain tumor from MRI images

The method is based on deep learning methods, it is aimed at reducing the level of uncertainty of the neural network model in the analysis of medical images. The fact is that doctors are engaged in marking up medical images, manually performing a huge amount of time-consuming work. But this is fraught with errors and inaccuracies, and, therefore, may affect the work of the neural network model. At the same time, the problem of neural networks is that a person fully trusts the solutions of the model. But you can't blindly rely on its conclusions, instead you need to take into account how confident the model is in its prediction. This is especially important for solving problems in the field of medicine.

Compared with the basic neural networks for processing medical images, the accuracy of the algorithm of ITMO researchers is 3% higher, and the calibration of the model has become twice as good. In fact, this is quite a lot – such an algorithm will not only be able to determine the pixels representing the tumor, but also predict its boundaries much better, and will also tell the doctor more accurately which pixels of the prediction he is more or less sure of.

The algorithm we created solves the problem of tumor segmentation. Let's say there is an image of the brain, the neural network will receive it and convert it into a binary image with pixels labeled 0 or 1, each of which will correspond to a healthy area of tissue or neoplasm. The algorithm also allows you to see areas in which the machine learning model is less confident, which require closer attention of the doctor. Usually these are the boundaries of the tumor. Our model is better calibrated because it has a much lower percentage of uncertainty," says Natalia Khanzhina, the author of the project, a graduate student at the ITMO Faculty of Information Technology and Programming.

The algorithm was tested on an open BraTS dataset, which includes 45 thousand magnetic resonance imaging images. This is very valuable for the scientific community, because the method is accessible, universal, and, therefore, can be applied to solve other problems.

Natalia is working on the research together with Maxim Kashirin, a graduate of the Master's program in Machine learning and Data Analysis at ITMO University. The scientists' project was selected for presentation at MICCAI, a prestigious international conference on medical image processing. Previously, the method was already presented at UAI, one of the most respected machine learning conferences in the world.

The material was provided by the ITMO University press service