Özet:

U-net, one of the methods of evolutionary nerve network, is a image division technique developed for medical image analysis, which can accurately divide images into sections using limited amounts of educational data. It is in a simple, flexible and expandable structure and provides high-quality pixels-level split results. Thanks to these features, it provides a very high benefit within medical visualization communities, and U-net and variations are widely used for medical visualization division tasks. Tomography (CT) scans, magnetic resonance (MR) scans, X-rays, and U-net gives successful results in large main image modes to microscopy. In addition, while U-net is largely used in dividing tasks, there are examples of U-net use in other applications. U-net’s potential for use in medicine and other fields increases every day. When we study the studies in the medical field in which the U-net and its varieties are used in architecture, the most studied area is the brain, and the most studied method of imaging is MR. In MR technology, image is created through radiofrequency waves in a powerful magnetic field environment. The MR technique that does not contain radiation and does not give any medication to the patient is used in the imaging of soft tissues. MR Imaging is often used in radiology to investigate the anatomy and physiology of the body, to identify pathologies, including muscle and joint diseases and abnormalities, neurological conditions, such as tumors, inflammation and stroke, abnormalities in the heart and blood vessels. The most common type of tumor in the brain in adults is glial tumor, which causes cancer and has a higher rate of death. Glyoma, one of the glial tumors, accounts for 75% of primary brain tumors in adults. Trusted division algorithms can help diagnose and analyze diseases related to the brain by allowing doctors to quantitatively study the tissues and structures. But the division of brain tissues is a very challenging task due to the inner and mixed shape of brain tissues, the dissemination of non-specific intensity, uncertain boundaries, the low contrast between its noisy structure and neighbouring brain tissues. When there are glyal tumors, the division process becomes even more complicated due to the diverse structure of the tumor that contains active and necrotic (dead) sections. All glial tumors do not have a clear limit between the dead and active parts, and some of the tumors do not have necrotic sections, while some do not have a division. In literature, the U-net architecture has been successfully used in the segmentation of brain glioma tumors, overcoming the aforementioned challenges. In this study in recent years using U-net architectures on brain MR images BRATS data sets have collected various studies that do glioma tumor segmentation and comparative information about them has been provided.

Anahtar Kelimeler:

Özet:

U-net, one of the convolutional neural network methods, is an image segmentation technique developed for medical image analysis that can precisely segment images using a limited amount of training data. It has a simple, flexible and expandable structure and offers high quality pixel-level segmentation results. Thanks to these features, it provides a very high benefit in medical imaging communities, and the U-net and its variations are widely used for medical imaging segmentation tasks. U-net gives successful results in major main image modalities such as Tomography (CT) scans, Magnetic resonance (MR) scans, X-rays and Microscopy. Also, although the U-net is largely used in segmentation tasks, there are examples of using U-net in other applications as well. The potential of U-net's use in medicine and other fields is increasing day by day. When examining the studies in the medical field where U-net and its varieties are used architecturally, the most studied area is the brain, and the most studied imaging method is MR. In the MRI technique, images are created by radiofrequency waves in a strong magnetic field environment. The MR technique, which does not contain radiation and does not give any medication to the patient, is used in the imaging of soft tissues. MRI is widely used in radiology to investigate the anatomy and physiology of the body, to detect pathologies including muscle and joint diseases and abnormalities, tumors, neurological conditions such as inflammation and stroke, and abnormalities in the heart and blood vessels. Glial tumors are the most common type of tumors in the brain that cause cancer and have a high mortality rate. Gliomas, one of the glial tumors, constitute 75% of primary brain tumors in adults. Reliable segmentation algorithms can help physicians to diagnose and analyze brain-related diseases by allowing them to quantitatively examine tissues and structures. However, the segmentation of the brain tissues is a very challenging task due to the intertwined and mixed shape of the brain tissues, the heterogeneous density distribution, the vague boundaries, the noisy nature, and the low contrast between neighboring brain tissues. When it comes to glial tumors, the segmentation process becomes more complicated due to the multifarious nature of the tumor, which contains active and necrotic (dead) parts. In all glial tumors, there is no distinct boundary between dead and active parts, and some tumors have necrotic parts, while some of them do not exist, making segmentation difficult. In the literature, U-net architectures have been successfully used in the segmentation of brain glioma tumors, overcoming the mentioned difficulties. In this review, various studies using U-net architectures on BRATS datasets for glioma tumor segmentation on brain MRI images were compiled in recent years and comparative information about them was presented.