Accurate multiclassification and segmentation of gastric cancer based on a hybrid cascaded deep learning model with a vision transformer from endoscopic images

Compared with other forms of cancer, gastric cancer has high mortality and incidence rates, making it a major cause of death worldwide. Accurate diagnosis is crucial in the treatment of stomach cancer. Researchers have used deep learning techniques facilitated by developments in artificial intelligence to classify and segment endoscopic images of stomach cancer. Most recent research examining endoscopic images of stomach cancer has used a binary classification system, which is insufficient for practical use. False-positives and computational costs become problematic when segmentation is applied to all the images of healthy patients obtained throughout the evaluation. Hence, the expected level of performance has not been attained in the real-time multiclassification and segmentation of stomach cancer. In this study, we present a deep learning-based technique for multiclassification of endoscopic images by combining modified GoogLeNet and vision transformer (ViT) models and identifying invasive areas based on Faster R-CNN. The classification of endoscopic images into three categories, namely, normal, early gastric cancer, and advanced gastric cancer, is accomplished by using a hybrid approach including modified GoogLeNet and vision transformer (ViT) models. Gastric cancer regions are then identified and segmented in the endoscopic images using the Faster R-CNN method. The Faster R-CNN algorithm is used with an endoscopic image as input, resulting in the generation of a bounding box and label image that accurately represents the gastric cancerous area. The proposed model achieved an accuracy, sensitivity and F1-score of 97.4%, 97.5% and 95.9%, respectively, for the classification of noncancerous, early gastric cancer and advanced gastric cancer. Furthermore, the performance of the segmentation method was also validated based on evaluation metrics and achieved 96.7%, 96.6% and 95.5% accuracy, sensitivity and F1-score, respectively, for the segmentation of noncancerous, early gastric cancer and advanced gastric cancer tissues. In conclusion, the method proposed in this study demonstrates enhanced global classification and detection performance compared to existing state-of-the-art algorithms. This finding underscores the significant potential of the proposed method in the domain of gastric endoscopic image classification and segmentation.