Automated Detection of Gastric Lesions in Endoscopic Images by Leveraging Attention-Based YOLOv7

Abstract

Gastric cancer is a leading cause of mortality, resulting in approximately 770000 deaths in the year 2020. Early detection theatres a vital role in facilitating targeted treatments for gastric conditions. One commonly employed method for diagnosis and treatment of gastrointestinal ailments is endoscopy. However, the effectiveness of endoscopy heavily depends on the expertise of the endoscopist. By integrating Artificial Intelligence techniques with endoscopic procedures, we can enhance the swiftness and accuracy of the diagnostic process.This study presents an automated approach that enhances the YOLO-v7 object detection algorithm through the integration of a Squeeze and Excitation attention block. This integration significantly improves the detection of small gastric lesions, demonstrating promising results. The attention-powered YOLOv7 achieved notable precision, recall, F1-score, and mean average precision values of 0.72, 0.69, 0.71, and 0.71, respectively. Additionally, the system achieved a high frame rate of 63 Frames Per Second, making it well-suited for real-time applications. Furthermore, a performance comparison with the baseline YOLOv7 model revealed a notable 10% increase in mean average precision and improved detection of small-sized lesions. The proposed architecture enables real-time lesion detection and identification, thereby supporting endoscopists in the analysis of endoscopic images, facilitating early diagnosis, and reducing reliance on the operator’s expertise.

Keywords

• Adenoma
• cancer
• convolutional neural network
• deep learning
• endoscopy
• gastrointestinal conditions