Tai Vovan and Dinh PhamToan published a study in Neural Computing and Applications that combines Vision Transformer with an improved genetic algorithm. The approach achieved classification accuracies of 98.9% and 94.2% on two medical image datasets. The framework extracts 768-dimensional features from images and uses a genetic algorithm to eliminate 71% to 81% of redundant features, improving computational efficiency. The study also uses SHAP to analyze the impact of each feature on the classification results.
Source:BIOENGINEER.ORG · Read Original
Leave a Reply