Hybrid Deep Learning Model for Endoscopic Ultrasound Image Recognition: Integrating YoloV4 and Vision Transformers

This project proposes an advanced approach for endoscopic ultrasound image recognition using data mining and hybrid deep learning techniques to address challenges such as image blurriness, noise, and difficulty in identifying organ structures. The method starts with preprocessing steps like graying, enhancement, histogram equalization, and noise reduction to improve image quality and feature extraction. An enhanced YoloV4 convolutional neural network (CNN) is used for detection and classification, trained on a comprehensive dataset for real-time, accurate detection. A hybrid model combining CNNs with Vision Transformers (ViTs) captures both local and global features, improving performance in recognizing subtle variations in organ structures. The results show that this hybrid system outperforms traditional manual methods, reducing misjudgments, increasing detection efficiency, and supporting medical professionals in making more accurate clinical decisions. This work advances automated detection technologies in healthcare, enhancing diagnostic accuracy and outcomes.