Gaussian Weighted Deep CNN with LSTM for Brain Tumor Detection

Abstract

Brain tumor is contemplated as a cruel illness in which the accuracy of images has a vital task. Accurate identification of tumor aid in exactly finding the injured portion and so decrease the mortality rate. Given that, learning the hidden pattern is significant to get an enhanced diagnosis and image quality. But, acquiring accurate diagnosis considering different lesion cases is a key concern. To overcome existing works problem, the automatic detection of brain tumor patients from their tumor images, Gaussian Weighted Deep Convolutional Neural Network with LSTM (GWDeepCNN-LSTM), is introduced. GWDeepCNN-LSTM technique includes many layers. First, brain MR images are collected from the given database. The image preprocessing is performed using a Gaussian weighted non-local mean filter where the noisy pixels are eliminated. Then, the segmentation is employed Hartigan's segmentation method to partition the image into similar regions. Followed by feature extraction is performed to extract the more informative features, such as texture, color and intensity, from the segmented image. Later, the classification of brain MR images is executed via Long short-term memory (LSTM). From that, the input image is classified as normal or tumor with higher accuracy. GWDeepCNN-LSTM performs better with an accuracy of disease detection and minimal time and error rate.