DEEP NEURAL NETWORK FOR ENHANCEMENT OF ACCURATE BRAIN TUMOR DETECTION IN MEDICAL IMAGING

Abstract

 This research paper proposed a framework designed to enhance the accuracy of brain tumor (BT) detection through the integration of deep learning (DL) and machine learning (ML) methodologies. The model implemented on publicly available MRI image dataset of brain tumors collected from the Kaggle repository. Histogram equalization is applied on the dataset to enhance the contrast features of the MRI images. These pre-processed images are then used to extract features from the images using transfer learning models such as MobileNetV3 and ResNet50. These extracted feature vectors then fed into the traditional machine learning model such as Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), K-Nearest Neighbors (KNN), and XGBoost for classification. The ResNet50 model paired with Logistic Regression achieving a peak classification accuracy of 95% alongside superior scores in precision, recall, and F1-measure. ResNet50 derived features consistently yielded the highest accuracy for ensemble methods like Random Forest and XGBoost. Features extracted by the lightweight MobileNet architecture were more compatible with SVM and KNN algorithms. These findings validate the proposition that hybrid architectures which leverage the feature extraction process of deep learning combined with the computational efficiency of machine learning classifiers can substantially improve the performance and accuracy of brain tumor diagnostic systems.