Segmentation

Dec 22, 2020

A brain tumor is the occurence in which a vast number of abnormal cells grow inside the brain. Moreover, there is an abounding number of existing brain tumors some of which can be classified as noncancerous(benign), and others as cancerous(malignant). Consequently, brain tumors start from the brain, or other parts of the body to which can eventually lead up to the brain. This process can be referred to as secondary brain tumors. One of the most common primary brain malignancies is gliomas. Moreover, "glioma therapy is a combination of surgery, chemo - and radiotherapy." Nonetheless, when it comes to brain tumor segmentation, the process might seem fairly complicated on the outside, whereas it can be accomplished using the proper tools. Such tools include handheld technologies such as Raman Spectroscopic Probes, 5-ALA Fluroscence Visualization and Raman Spectroscopy. In fact, Dr. Sofie Van Cauter stated in her article, "when dealing with a brain tumor, radiologists look at multiple MRI sequences performed in multimodal imaging protocol." After all, brain tumor segmentation is the process of separating the tumor from normal brain tissues. Nevertheless, this process is challenging given that the brain is in an irregular form of state and confusing boundaries of tumors. On the other hand, recently there has been a massive growth in the use of deep convolutional neural networks (DCNN) to which each test produced an adequate result in natural and medical imaging segmentation tasks. Without a doubt, there was a slight concern, one in which the models are difficult to obtain from a large dataset. Similarly, Magnetic Resonance Imaging(MRI) has been commonly used in the clinical routine. This can be the fact that it can overture high spatial resolution images with high constrast amidst soft tissues. The challenges that can arise from brain tumor segmentation can be scaled down with the use of AI technologies by developing an algorithm for determining a faster brain tumor subtype accuracy. In general, an "accurate segmentation of brain tumors from MRI images represents a crucial and a challenging task in diagnosis and treatment planning." Currently, image segmentation is a growing terrain when it comes to medical imaging, as it abides with the abstraction in images from the evolving era of interest.

Current Situation

One of the most common treatments for brain tumor segmentation is surgery. This process employs the slowness of tumor growth to make the segmentation procedure doable. However, with the process of localizing the diameter of each tumor in the brain can become difficult to quickly, and one step to overcome this challenge is to utilize MRI scanning, which can be beneficial in providing detailed scans of the brain.

As a result, image scanning can produce both low and high grade outputs in classifying the segmented areas. In other words, "spectroscopic tools have the theoretical advantage of accurate tissue identification, coupled with the potential for manual intraoperative adjustments to improve visualization of remaining tumor tissue that would otherwise be difficult to detect." Moreover, there still consists a major challenge in determining accurate segmentation when performing image abstraction that can successfully treat each patient. As Sajid Iqbal stated in his research, "There may exist multiple tumors of different types in a human brain at the same time." Each stochastic model can produce complex presence when extracting from the tumor's texture and can result in a geographical changeable component that may differ from the actual output. The following figure shows the results of several MRI segmentation image scans using HGG and LGG cohort cases when inspecting each individual brain in classifying a tumor patch.

In contrast, scientists and researchers have discovered an architect for treating brain tumor by using machine learning techniques such as deep learning and CNN efficiency. The use of AI technologies can further improve the overall process needed to accurately segment brain tumors by applying segmentation algorithms which can reduce several challenges in diagnosis accuracy and speed performance.

To demonstrate, CNNs are willing to facilitate the copied MRI image scan and perform the burden of tumor segmentation. Second, the overall procedure comprises network training and using a hand operated segmented database that is later stabilized in the final image output. Lastly, the segmented images can be refined expeditiously to envision analytical sequelae including therapy & survival feedback. The amount of data obtained from regular MRI scanning is large and can take up to several hours to process. In other words, the use of Convolutional Neural Networks can speed up processing time by applying automatic segmentation for each patient whether they persist HGG(Higher Grade Gliomas) or LGG(Lower Grade Gliomas) tumors. Generally, patients with LGG tumors achieve higher accuracy during segmentation compared to those that contain HGG tumors. In contrast, most of the segmentations are fulfilled by the use of a wrapper library in Keras and implemented with TensorFlow.

History

The procedure behind brain tumor segmentation dates to the development of CT and MRI scanners. On the contrary, CT(Computer Tomography) Scanners appeared in the medical field in 1957 but was first initialized in 1971 on a patient's brain. This process made image scanning more efficient from the traditionalized manual labor. To illustrate, CT Scanners were able to process segmented data points in a patient's brain at "forty 8 mm slices" every 5 to ten seconds. Over time, this was able to achieve a 512 by 512 matrix image by scanning different parts of the brain within a matter of seconds. To demonstrate, CT Scanners were able to successfully help determine a patient's tumor and subtle delineation with detailed pixel verdict for identifying the patient's dilemma and how to resolve it. In contrast, MRI scanning came into play in 1977 when the first patient was successfully scanned for any possible deterioration. This led to establishing a full-body MRI machine for undergoing multiple images scans at once. With the creation of AI and machine learning protocols, in the early 1960's, scientists and researchers have adopted multiple deep learning protocols that are geared towards helping perform automatic segmentations on patients by reducing the processing time and execution.

Conclusion

It can be quite a challenge when performing brain tumor segmentations. With the help of AI and machine learning, CNNs can speed up the mechanism for segmentation by reducing the estimated time required to render each scanned image produced by the MRI. The process of image scanning has improved from using recursive sequences to dynamic programming implementation by magntitude structure, even when max-pooling layers are precise. Furthermore, deep neural networks can transcend in analyzing images, detection and segmentation. Without a doubt, deep learning networks have been shown to cultivate adequate procedures in processing medical imaging segmentation tasks. As each algorithm gets smarter by overcoming its mistakes, the process of brain tumor segmentation will eventually present accurate diagnostic readings from overcoming slow stuttering and weak performances.