Deep Learning Based Approaches for Gastric Cancer Detection and Lesion Segmentation in Gastrointestinal Endoscopy Images

Gastric cancer (GC) is one of the foremost contributors to cancer-related mortality globally, presenting major public health challenges. Both genetic factors and lifestyle choices contribute to its prevalence, particularly among individuals under 50 years old. Traditional classification systems, such as the Lauren and WHO frameworks, often struggle to guide treatment effectively due to the complex nature of these tumours. Recent progress in artificial intelligence (AI), particularly deep learning (DL) models like Convolutional Neural Networks (CNNs), is proving valuable for enhancing early detection through endoscopic imaging. Methods such as Mask RCNN, U-Net R-CNN, GAS-Net and GoogLeNet have demonstrated good accuracy rates in identifying and segmenting tumours, which may lead to more personalized and timely treatment options. Among all these, GoogLeNet emerged as the highest-performing model, achieving an accuracy of 96.7%, primarily due to the use of the Inception-v4 architecture combined with a transfer learning algorithm, which improved its ability to detect and segment tumours. This study reviews these AI-based diagnostic methods, emphasizing their potential to improve patient outcomes. However, further extensive trials are necessary to confirm the reliability of these models across diverse patient populations.