Towards Design of Brain Tumor Detection Framework Using Deep Transfer Learning Techniques

Introduction

Recently, Brain Tumor (BT) detection becomes a fundamental research challenge due to increase in cases worldwide and this problem has attracted researchers to find out AI-based detection tools for early diagnosis There are primary and secondary BT. In primary BT, a tumor grows in the brain, it can be described as ‘high’ and ‘low’ grade tumor. High grade tumor grows faster as compared to low grade whose growth is slower. The secondary BT are the tumor that grows in another part of body such as lung, breast, etc., and then spread through the brain, it is also called as metastatic. Figure 1. shows some of the types of BT and a healthy brain image. BT is the abnormal growth of cells in the brain. There are many methods that are used for detection of BT with high accuracy. The rise in artificial intelligence (AI) and machine learning (ML) field help in BT surgery. Brain surgery with AI is resulting safer and more efficient and precise. These methods are performing better in different field like early diagnosis of BT, surgery, optimizing the surgical plan, better prediction the prognosis and providing efficient support during the operation.

Figure 1.

Types of BT and No-tumor brain

Early detection and identification of BT are crucial for the patient's efficient and prompt therapy. Our visual cortex's capacity to discern levels of MRI (Magnetic Resonance Imaging) images limits our ability to identify BT. So, the next technology, known as CAD (Computer-Aided Diagnosis), was invented to help radiologists detect different types of tumors and provide improved visualization capabilities. This technology automatically analyses photos and recognizes BT, as well as performs numerous operations such as segmentation, classification, and others that help doctors better comprehend and save their patients' lives, as well as researchers working in these disciplines to analyze these BT. Also, due to the improved diagnosis findings obtained by this technology, the odds of surgery are lowered. Image processing, computer vision, and image segmentation are some of the methods that may be used to determine the nature of a tumor, measure its size and depth, and better comprehend its structure. For automated detection, classification, and segmentation, these approaches have been demonstrated to be accurate and efficient.

These methods employ AI subfields that are divided into two categories: machine learning (ML) and deep learning (DL). ML employs a variety of feature selection methods, which are then used for classification and segmentation. In DL, a convolutional neural network (CNN) is utilized to extract the hidden pattern from BT images and to classify and segment BT using labelled images. SVM, KNN, Random Forest, CNN, UNET, and other ML and DL algorithms are only a few examples. This paper focuses on design of Brain Tumor framework using latest Deep transfer learning techniques. The main contribution of this work is list out as follows:

• 1.

  Design of brain tumor detection framework using latest deep transfer learning approaches such as DenseNet201, DenseNet169, DenseNet121, MobileNet_v2, VGG19, VGG16, and Xception.

• 2.

  Performance evaluation of the proposed framework is evaluated using benchmark dataset in terms of accuracy and loss.

In the later section, a related work is presented, Next, describe the description of the proposed framework using deep transfer learning models. Section 4 presents result analysis and discussion. Finally, section 5 concludes the paper.

Related Work

In this section we have discuss various related study which help us to learn and build the proposed work. In this various feature extraction, pre-processing, CNN architecture, different pre-trained model and their performance and approaches are discussed.