Optimised CNN based Brain Tumour Detection and 3D Reconstruction

Abstract

ABSTRACT The brain tumour classification and its early identification are significant for treating the tumour efficiently. The 3D visualisation and detection of the brain tumours from MRI is an error-prone task and computationally time-consuming one. This paper offers a 3D reconstruction scheme for reconstruction of brain tumour. Initially, the preprocessing is performed by the process of average filter and histogram equalisation. Subsequently, the K-means clustering-based segmentation is carried out. As a novelty, to detect the brain tumour, an optimised convolutional neural network (CNN) is employed, in which the training process is carried out by a new Levy-adopted Tunicate Swarm Algorithm (L-TSA) through the optimal weight tuning. At last, the 3D reconstruction of brain tumours is performed by the improved marching cube (MC) algorithm for accurate 3D construction of the tumour region. At last, the performance of developed approach is carried out in MATLAB, and the results are verified for the extant schemes concerning certain metrics like sensitivity, FNR, specificity, FPR, accuracy, FNR, precision, MCC, NPV and F1-score. Accordingly, the accuracy of the adopted CNN+L-TSA algorithm attains a higher value of ~0.934 to extant schemes, including CNN+GWO, CNN+EHO, CNN+TSA and CNN+CUGWA, correspondingly.