Abstract
Cancer is one of the most lethal diseases in the world. A brain tumor is a form of cancer that develops in the brain's glial cells. Magnetic resonance imaging (MRI) is a prominent imaging tool for detecting brain tumors. It includes four different modalities that neurologists use to determine the location and kind of tumor. The suggested approach uses a 2D U-Net model to separate the brain tumors sub regions. To prevent excessive preprocessing and GPU utilization, the authors utilize the patching approach to partition the picture slices into distinct patches in this study. Second, they leverage the squeeze and excitation blocks to more effectively map low-level features to high-level features than a basic U-Net. The suggested technique yields DICE scores of 0.85, 0.87, and 0.90 for the three tumor categories of enhancing tumor, whole tumor, and tumor core, respectively. The results outperform the most recent approaches, including the major papers from the Brats 2019 competition.
TopIntroduction
Glioma is a form of brain tumor that develops from glial cells. It is classified as either High-Grade Glioma (HGG) or Low-Grade Glioma (LGG). The survival rate of LGG patients is significantly higher than that of HGG patients. A brain tumor is one of the major factors that increase the global death rate. In addition, its survival rate is lower than that of other cancer types. Brain tumors can be separated into two categories. 1) Primary brain tumor that originates and spreads inside the brain. 2) All other tumors are secondary or metastatic tumors that originate in other organs of the human body.
Tumors in the brain can be further classified into two types. Malignant and benign tumors. Malignant tumors consist of cancerous cells, which are life-threatening to the human body. The following are some common types of malignant tumors. 1) Carcinoma. 2) Sarcoma. 3) Germ Cell. 4) Blastoma (Peddinti et al, 2021) (Jiang et al, 2019). While Benign tumors do not contain cancerous cells, their survival rate is high as compared to malignant tumors. But some types of benign tumors become cancerous if appropriate and timely treatment is not provided to the patient. They do not harm surrounding tissues, nor do they spread to the other organs of the body. The following are some common types of benign tumors. 1) Adenomas. 2) Fibroids. 3) Hemangiomas. 4) Lipomas.
The following sections make up the remaining portion of our work.
First, the background part describes the history of brain tumors and the method that had been used to solve segmentation problem. Second, in the related work section, we look at the various strategies used in the literature to solve the tumor segmentation problem, including traditional and deep learning approaches.
Third, the methodology part delves further into the dataset, preprocessing, the U-Net model, and our suggested strategy. Fourth, the output of the model we proposed is discussed in this section.
All of the results are displayed in image form to allow for a more visual examination of the tumor data. Fifth, we summarize our findings and discuss our tumor segmentation analysis in the conclusion. Sixth, in this section, we include all of the sources we used in this work for literature and other purposes. Seventh, in the last section we discuss several important terms and definitions connected to our work.
Key Terms in this Chapter
Dice Coefficient: The dice coefficient is the measure used to determine the similarity of two data sets. It compares the pixel to the expected image and ground truth pixel values.
Jaccard Index: It compares the members of two sets to determine the shared and unique members. The more similar the two sets are, the higher the percentage.
Accuracy: The proportion of all data points that were correctly predicted. It is the ratio of true positives and true negatives to false positives, false negatives, and true positives, divided by all of these.
Hausdorff Distance: The Hausdorff distance can be defined as if there are two sets and one point in one set has the longest distance with the most adjacent point in other sets.
Magnetic Resonance Imaging (MRI): It is a medical imaging technique that produces three-dimensional precise structural images using a magnetic field. It is frequently employed in illness detection, diagnosis, and therapy monitoring.
Sensitivity: It is the parameter used to assess a model's capability to predict the proportion of real positive cases that were predicted as positive (or true positive).
Specificity: It is the parameter used to assess a model's capability to predict the proportion of actual negative cases that were predicted by model as negative (or true negative).