Artificial intelligence-assisted diagnosis of early gastric cancer: present practice and future prospects

Abstract

Gastric cancer (GC) occupies the first few places in the world among tumors in terms of incidence and mortality, causing serious harm to human health, and at the same time, its treatment greatly consumes the health care resources of all countries in the world. The diagnosis of GC is usually based on histopathologic examination, and it is very important to be able to detect and identify cancerous lesions at an early stage, but some endoscopists' lack of diagnostic experience and fatigue at work lead to a certain rate of under diagnosis. The rapid and striking development of Artificial intelligence (AI) has helped to enhance the ability to extract abnormal information from endoscopic images to some extent, and more and more researchers are applying AI technology to the diagnosis of GC. This initiative has not only improved the detection rate of early gastric cancer (EGC), but also significantly improved the survival rate of patients after treatment. This article reviews the results of various AI-assisted diagnoses of EGC in recent years, including the identification of EGC, the determination of differentiation type and invasion depth, and the identification of borders. Although AI has a better application prospect in the early diagnosis of ECG, there are still major challenges, and the prospects and limitations of AI application need to be further discussed.

Keywords: Early gastric cancer; artificial intelligence; convolutional neural networks; deep learning.

Figure 1.

Structure of the classical detection model fast-cnns. Convolutional layers are used to computationally obtain different feature maps, the convolutional layers located at the beginning of the network architecture are used to detect low-level semantic features such as edges and curves, while the convolutional layers located deeper into the model architecture are used to learn more abstract semantic features. An activation function is applied to the convolution results to obtain a feature map. The pooling layer serves to reduce the computational burden on the network by decreasing the spatial dimensions of the feature map and the number of parameters in the network without losing information. By stacking several convolutional and pooling layers, feature maps with high-level semantic information can be obtained gradually. One or more fully connected layers follow the convolutional and pooling layers for integrated extraction of features to enhance the feature representation of the model. In a convolutional neural network, each neuron of the feature map is connected to the region of a neighboring neuron in the previous layer, such a neighborhood is called the neuron’s receptive field. When generating a feature map at a certain level, the convolution kernel of that level is shared by all spatial locations of the input, and this weight sharing can effectively reduce the parameters of the network and reduce the risk of overfitting of the network. Complete feature information is obtained by setting different convolution kernels at different levels. The last layer of the convolutional neural network is the output layer, which can usually be used to obtain the best parameters for a particular task and output predictions by minimizing the loss function defined on that task. (Endoscopic gastric mucosal morphology was obtained from the First Affiliated Hospital of Soochow University, with written informed consent from the patient and a statement of consent for publication.) (Figure drawing by Visio 2024).