ACNN-LSTM Hybrid Deep Learning Framework for Accurate Brain Tumour Classification

Abstract

Brain tumors are neurological diseases that are life- threatening and must be detected early and properly to maximize patient outcomes and allocate the treatment options. The most popular non-invasive analysis of the brain, Magnetic Resonance Imaging (MRI), is the imaging technique that provides high- resolution images of soft tissues and tumor mass. Nevertheless, the interpretation of MRI images by radiologists typically takes time, making them subjective and diagnostically inconsistent. This paper, to deal with these issues, presents a hybrid deep learning model, which is a composite of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) to classify brain tumors efficiently. The CNN component performs the high-level tasks of spatial and structural feature extraction of MRI images, and the LSTM component provides sequence dependencies and spatial correlation between extracted features and therefore improves the discriminative ability of the model. The developed CNN–LSTM hybrid network was developed and tested on the publicly available MRI brain tumor database containing glioma, meningioma, pituitary, and normal cases. Preprocessing of images was done through normalization, image size resizing as well as augmentation to maintain generalization and eliminate over fitting. Experimental findings reveal that the offered model is characterized by a total accuracy of 97.6 percent in classification, which is higher than conventional CNN and other previously existing models. The measures of quality used to assess the performance like precision, recall, and F1-score confirm the strong and reliable performance of the model in classifying tumor and non-tumor. Also, the Grad-CAM visualization methods were used to visualize the model and interpret the decision making process to enable clinical transparency and explain ability. The system is software-based, scalable and can be intertwined with the medical diagnostic workflow to be used in real-time clinical practice. Overall, the CNN–LSTM hybrid model can be considered as a strong, interpretable, and efficient framework that can be used in automated brain tumor classification and help radiologists acquire diagnoses and be able to plan the treatment course without drawing too much on human potential and increasing diagnostic latency.