Improving Brain Tumor Diagnosis Accuracy: A Machine Learning Approach with CNN, RNN, and PCA

Abstract

Effective and efficient brain tumor classification from MRI scans is of critical importance as medical diagnostics in detecting early signs of the disease and treating the disease as early as possible. The focus of this paper is to propose a novel method to classify the brain tumors into 4 types of glioma, meningioma, notumor, and pituitary tumors using a combination of RNN based LSTM with PCA and SVM. To extract features from the MRI images, we use VGG19 a pre trained Convolutional Neural Network (CNN) and because the data is sequential, LSTM is utilized to process the sequential nature of the data so that the model learns the temporal relationship between multiple MRI slices. They are then applied to an SVM classifier with Principal Component Analysis (PCA) for dimensionality reduction and improved efficiency for classification. To further enhance model robustness, we combine three prominent brain MRI datasets, ensuring a diverse set of training examples. The experimental results show that the proposed LSTM-based SVM model gives 97% accuracy in all the tumor categories with high precision, recall and F1 scores. The model’s performance dominates the existing CNN based models especially in term of generalization where training and validation accuracy exhibit little change implying good overfitting prevention. Two main contributions are identified to address the problem with a hybrid approach consisting of both Deep Learning and traditional ML techniques: (a) both methods achieve high accuracy and (b) results are interpretable and scalable.