Analysis and Detection of Brain Tumor Using U-Net-Based Deep Learning
Brain tumor could be a life threatening disease and the survival rate of such disease is low. It is generally the abnormal growth of cells inside the brain. Early and accurate detection of the brain tumor is very difficult. The manual segmentation of the
- PDF / 2,110,907 Bytes
- 13 Pages / 439.37 x 666.142 pts Page_size
- 91 Downloads / 259 Views
Abstract. Brain tumor could be a life threatening disease and the survival rate of such disease is low. It is generally the abnormal growth of cells inside the brain. Early and accurate detection of the brain tumor is very difficult. The manual segmentation of the brain tumor extent from 3D MRI (Magnetic Resonance Imaging) volumes is a time consuming process and depends a lot on the operator’s experience. The automatic tumor segmentation has the potential to decrease lag time between diagnosis tests and the treatment for the same. Hence, there is a high demand of time and memory efficient, and reliable computer algorithms to do this accurately and quickly. In this paper, we first highlight limitations of the image processing based solutions and subsequently present a novel deep learning based technique. The proposed technique relies on U-Net based Deep Convolutional Networks for the automatic detection and analysis of brain tumors. Keywords: Brain tumor · Segmentation · Multimodal MRI · Thresholding · Deep neural networks
1 Introduction Multimodal MRI (Magnetic Resonance Imaging) is a popular technology used for the detection of brain tumor by observing the soft tissues. The MRI images are better in terms of quality as compared to other non-invasive imaging techniques such as X-Ray or Computed Tomography. Brain tumors could be life threatening and thus the accurate and timely detection of brain tumors is important. The brain tumors are classified into two groups – benign and malignant tumors. The malignant ones consist of fast growing cancerous tissues as compared to the benign tumors. The MRI images consist of weighted images or segments: T1-weighted, T2-weighted, Flair-weighted (Fluid Attenuated Inversion Recovery) and T1c. Obtaining these images is a difficult problem because the manual segmentation is a time-consuming process and the accuracy depends a lot on the experience of the operator. The process of MRI scan is also quite exhaustive and needs a lot of efforts at the operator’s part. Any mistake(s) at the operator’s part will lead to chaos due to incorrect diagnosis. The conclusions also may vary from one operator to another operator. Hence, the efficient and reliable computer algorithms are © Springer Nature Switzerland AG 2020 K. Arai et al. (Eds.): SAI 2020, AISC 1230, pp. 161–173, 2020. https://doi.org/10.1007/978-3-030-52243-8_13
162
V. Garg et al.
required to solve this problem. The possible solutions of automatic brain tumor segmentation are image processing based and deep learning based. The solutions strictly based on image processing techniques like Thresholding Based Segmentation [1] are insufficient in brain tumor detection. The automatic segmentation and further classification of tumor into various other types from the multimodal MRI scans remains a popular area of research in the field of medical science. The proposed model in this paper extracts features using Convolutional Neural Networks (CNN) technique and then tries to learn the characteristics of tumors through extensive training o
Data Loading...
