EFFICIENT TUMOUR DETECTION FROM BRAIN MR IMAGE WITH MORPHOLOGICAL PROCESSING AND CLASSIFICATION USING UNIFIED ALGORITHM

Abstract

Brain diseases caused due to malignant are the biggest concern among all the age groups. Studies show that almost 80% of death cases are reported due to presence of malignant tumour. Hence diagnosing brain tumour at an early stage would increase the survival rate. Magnetic resonance imaging (MRI) plays a major role in diagnosing tumours in human brain. However, it is considered to be a time consuming and tedious process which could lead to deviation in the opinion of radiologists. This has led to the development of computer-based automatic extraction of tumour cells from the images obtained by MRI. This paper proposes an efficient tumour detection mechanism from MR images using morphological processing and unified algorithm. A neural network that uses bounding boxes and associated class probabilities detects the packets of tumour that exist in a full MR image. Simulated results of the proposed technique on the BRATS 2016 dataset show that a detection accuracy of 95.97% is achieved, while reducing the likelihood of false positives. This approach is compared with other detection methods such as DPM and R-CNN and the analysis proves that our method proposed outclasses the other detection methods.