HYBRID TUMOR CLASSIFICATION USING VISION TRANSFORMERS AND PSO-OPTIMIZED FEATURE SELECTION WITH XGBOOST

Abstract

Accurate tumor classification from medical images is a crucial task in computer-aided diagnosis, requiring high precision and effective feature extraction techniques. This study proposes a hybrid tumor classification framework that integrates Vision Transformers (ViT), Particle Swarm Optimization (PSO)-based feature selection, and XGBoost classification to enhance prediction performance. Initially, medical images are processed using a pretrained Vision Transformer to extract deep, high-level features that capture global contextual information more effectively than traditional Convolutional Neural Networks (CNNs). The extracted feature vectors are then optimized using Particle Swarm Optimization, which selects the most relevant and discriminative features while reducing dimensionality and eliminating redundant data. The optimized feature set is subsequently fed into an Extreme Gradient Boosting (XGBoost) classifier to perform the final tumor classification. This hybrid approach leverages the powerful representation capability of Vision Transformers, the optimization efficiency of PSO, and the high accuracy of XGBoost. Experimental results demonstrate that the proposed method achieves superior performance in terms of accuracy, precision, recall, and F1-score compared to conventional CNN-based and standalone machine learning models.