An Efficient Hybrid Framework for Multiclass Brain Tumor Classification Using Handcrafted Features and Neural Networks

Abstract

Brain tumors are serious neurological disorders characterized by abnormal and uncontrolled cell growth within brain tissues, leading to significant disruption of normal brain function. Early detection is essential for improving clinical outcomes and enabling effective treatment. Medical imaging techniques, particularly computed tomography, are widely used for tumor detection; however, manual interpretation is time-consuming and prone to variability. To address these challenges, automated approaches based on machine learning and deep learning have been developed. Despite their effectiveness, deep learning models often require large annotated datasets, high computational resources, and may fail to capture subtle texture variations in medical images. To overcome these limitations, this paper proposes a multiclass brain tumor classification framework based on a Hybrid Feature Vector Network (HFVN). The proposed approach integrates complementary handcrafted features by combining Local Binary Patterns (LBP) and Histogram of Oriented Gradients (HOG) to capture both texture and structural information. The fused feature vector is utilized by the HFVN model for accurate classification, and its performance is compared with a CNN-based model. Experimental results demonstrate that the proposed HFVN achieves a classification accuracy of 93%, significantly outperforming the CNN model, which achieves 70%. These results highlight the effectiveness of combining handcrafted features with an efficient classification architecture for reliable multiclass brain tumor detection.