BRAIN TUMOR DETECTION MODEL USING DIGITAL IMAGE PROCESSING AND TRANSFER LEARNING

Abstract

MRI images are the first input used in the detection of brain tumors. The healthcare system would greatly benefit from the development of autonomous detection systems. Due to technological advancements, MRIs are now digital and can be analyzed utilizing image processing methods to automate classification methods. preprocessing steps help to improve the accuracy of brain tumor detection using digital image processing techniques. It typically includes image acquisition and normalization, image enhancement, feature extraction and feature selection. This study has looked into a technique for classifying brain MRI images using transfer learning and convolution neural networks. This study's primary objective was to develop a model for the identification of brain tumors using transfer learning techniques and techniques for processing and classifying MRI images. This study is confined to categorizing 2800 brain MRI images from Korea hospital in Ethiopia either a tumor or healthy based on their size, shape, and pattern. The suggested detection system uses a pre-trained model like VGG16 or Inception V3 and combines deep learning with transfer learning. Accuracy measurement like precision, recall and f-1 score metrics were used to illustrate the model's performance and results. The model's accuracy was increased by using a variety of model optimization strategies. The model properly classified images into classes of healthy or tumors 93.10% of the maximum Accuracy. incorporating more fully connected layers with appropriate NNs. Data augmentation is used to avoid over fitting hence the collected data is small in number for this study'. Standard datasets for in-depth experimentation were advised as tasks for the future because machine learning and deep learning algorithms need large size datasets for better performance and generalization