Efficient brain tumor detection in MR images using deep feature-extracted machine learning

Brain tumors are a leading cause of cancer-related deaths in individuals over 65, emphasizing the need for early detection for better patient outcomes. Accurate and timely detection of brain tumors in magnetic resonance (MR) imaging allows medical professionals to make prompt, informed decisions, leading to effective treatment plans and quality patient care. However, achieving high accuracy in tumor detection while minimizing computation time remains challenging, especially in resource-limited clinical settings. This study combines deep feature-based machine learning (ML) techniques with transfer learning strategies to enhance both execution time and classification accuracy in brain tumors detection. It utilizes a publicly available dataset on Kaggle that is curated explicitly for brain tumor detection in MR images. Specifically, this paper evaluates the performance of four pre-trained deep learning models, VGG-16, VGG-19, MobileNet-v2, and DenseNet-121, combined with four ML classifiers: random forest, decision tree, support vector machine, and Gaussian Naïve Bayes. The research comprises two stages: first, ML classifiers are applied to image features extracted from pre-trained convolutional neural network-based deep learning models. Next, principal component analysis (PCA) reduces the dimensionality of these deep feature vectors before reapplying the classifiers. Results show that brain tumor classification without PCA achieves high accuracy (up to 98.61%) but requires significant computational time (509.43 seconds). In contrast, PCA maintains comparable accuracy (98.84%) while reducing execution time to under 40 seconds. The findings confirm the potential of the proposed method as a practical and efficient method for medical imaging, contributing to more effective and timely brain tumor detection.