The application of artificial intelligence in upper gastrointestinal cancers

Abstract

Upper gastrointestinal cancers, mainly comprising esophageal and gastric cancers, are among the most prevalent cancers worldwide. There are many new cases of upper gastrointestinal cancers annually, and the survival rate tends to be low. Therefore, timely screening, precise diagnosis, appropriate treatment strategies, and effective prognosis are crucial for patients with upper gastrointestinal cancers. In recent years, an increasing number of studies suggest that artificial intelligence (AI) technology can effectively address clinical tasks related to upper gastrointestinal cancers. These studies mainly focus on four aspects: screening, diagnosis, treatment, and prognosis. In this review, we focus on the application of AI technology in clinical tasks related to upper gastrointestinal cancers. Firstly, the basic application pipelines of radiomics and deep learning in medical image analysis were introduced. Furthermore, we separately reviewed the application of AI technology in the aforementioned aspects for both esophageal and gastric cancers. Finally, the current limitations and challenges faced in the field of upper gastrointestinal cancers were summarized, and explorations were conducted on the selection of AI algorithms in various scenarios, the popularization of early screening, the clinical applications of AI, and large multimodal models.

Keywords: Artificial intelligence; Esophageal cancer; Gastric cancer; Radiomics; Upper gastrointestinal cancers.

Fig. 1

The application of artificial intelligence in upper gastrointestinal cancers research primarily focuses on four aspects: screening, diagnosis, treatment, and prognosis. CT, computed tomography; NBI, narrow-band imaging; WLI, white-light imaging.

Fig. 2

The pipeline of radiomics. The input images undergo image segmentation initially to acquire the region of interest. Subsequently, features such as texture, shape, and histogram are extracted from the segmented regions. LASSO or other methods are then employed for feature selection. Following this, machine learning methods such as K-means, SVM, and RF are utilized for modeling based on the selected features. Finally, the performance of the model is evaluated using metrics like the ROC curve, confusion matrix, and decision curve. FN, false negative; FP, false positive; LASSO, least absolute shrinkage and selection operator; mRMR, max-relevance and min-redundancy; PCA, principal component analysis; RF, random forest; ROC, receiver operating characteristic; SVM, support vector machine; TN, true negative; TP, true positive.

Fig. 3

The pipeline and classic model of deep learning in medical imaging. CNN, ViT, and GAN are commonly used in deep learning models in medical image analysis. Upon training these models, they can be applied to specific downstream tasks. The pipeline of deep learning is highly flexible, allowing for modification of model architectures and autonomous design of training tasks according to specific requirements. cCRT, concurrent chemoradiotherapy; CNN, convolutional neural network; D, discriminator; DFS, disease-free survival; DGC, diffuse gastric cancer; DL, deep learning; FC, fully connected layer; G, generator; GAN, generative adversarial networks; IGC, intestinal gastric cancer; MLP, multilayer perceptron; NACT, neoadjuvant chemotherapy; NBI, narrow-band imaging; nCRT, neoadjuvant chemoradiotherapy; OS, overall survival; ViT, vision transformer; WLI, white-light imaging.

Fig. 4

The general framework of the model proposed by Hou, consisting mainly of two modules: SE-ResNet 50 and the attentive hierarchical aggregation module. These serve as the student and teacher, respectively. The entire model is trained in a supervised learning setting with self-distillation. L_ce represents the cross-entropy loss, while D_KL denotes the KL divergence.