Artificial Neural Networks for Recognition of Brain Tumors on MRI Images

Abstract

In this paper, artificial neural networks for recognition of brain tumors on MRI images are analyzed. This analysis allows to choose the most appropriate neural network architecture and various preprocessing techniques to increase the precision of tumor instance recognition. Understanding the image and extracting information from it to accomplish some result is an important area of application in digital image technology. Image recognition has quickly found its use in medicine and specifically oncology. Precise recognition masks may not be critical in other cases, but marginal recognition errors in medical images may render the results unreliable for clinical use. Therefore, biomedical problems require much higher boundary detection precision to improve further analysis. Various methods and algorithms for image recognition and segmentation are considered. The advantages and disadvantages of neural network architectures (ResNet, U-Net, SegNet, YOLO v3) are considered and analyzed in more detail. Additionally, the analysis of data preprocessing methods was carried out, as well as a study of the input data. Comparison of different artificial neural networks algorithms and architectures will achieve the highest accuracy of recognition. During the comparison, a system of the U-Net architecture with additional processing methods was selected as the final model. Its accuracy reached 94%, which is a significant result compared to manual image recognition.