Giải pháp kết hợp Vision Transformer và Active Learning trong phát hiện và phân loại tổn thương gan

Abstract

This paper proposes a combined solution of Vision Transformer (ViT) and Active Learning to detect and classify liver lesions in CT images. The study was conducted on 2008 CT images using DenseNet-121, VGG-16, and Vision Transformer models, achieving a maximum accuracy of 0.99. Active Learning was applied to reduce manual labeling costs and optimize the machine learning process. The results demonstrate the potential of this method in medical diagnosis support systems.