病理学诊断是胃病诊断的金标准。当下胃病已成为我国第一大常见病，紧缺的病理医师资源正面临着越来越大的挑战。随着人工智能在医疗图像领域的飞速发展，结合数字病理与人工智能，为病理医师赋能，已成为趋势。人工智能的发展离不开数据和模型。然而当前公开的胃病理数据集少之又少，符合研究需求的数据集更是寥寥无几。获得一份精确标注的自建胃病理数据集，是我们开展研究前的首要工作。另外，国内外基于卷积神经网络和Transformer架构的胃病算法也存在着精度偏低、运算成本较高等痛点。为此，本文开展了关于胃癌和胃炎分类模型的研究，并完成算法的工程化落地，来真正实现为病理医师赋能。本文的工作和创新可以概括如下：1.构建具有细胞级特征标注的大型胃病理数据集，填补了大型公开数据集的缺失。本文构建的数据集中，共采集胃数字病理图像930张。其中胃炎占610张，正常黏膜122张。胃癌数字病理图像共采集198张。2. 基于深度学习的创新性胃癌分类模型。本文提出的新型胃癌分类模型，该模型由以下两大部分构成：（1）为了解决注意力机制编码模块处理大规模图形数据时产生的复杂计算量而提出的基于残差连接网络的新型定位网络。（2）基于嵌入式位置信息向量中心化的改进，提高了位置编码的效率。3. 基于深度学习的创新性胃炎分类模型。本文提出的创新性胃炎分类模型，由以下两大部分组成：（1）基于空洞卷积和多尺寸卷积核的多尺度特征编码器。（2）基于特征图上采样和特征聚合的解码器。4.基于深度学习与PyQt的胃病临床辅助诊疗系统开发本研究中，为了实现算法的工程化落地，进行了辅助诊疗软件的开发。建立了完整的病理切片辅助诊断的实际应用流程，实现了模型部署、模型预测、图像预处理等图像管理和操作功能。关键词：数字病理图像；胃癌与胃炎；图像处理；工程化软件；深度学习

Pathological diagnosis is the gold standard for the diagnosis of gastric diseases. At present, stomach disease has become the first common disease in our country, and the scarce pathologist resources are facing more and more challenges. With the rapid development of artificial intelligence in the field of medical images, it has become a trend to combine digital pathology and artificial intelligence to empower pathologists.The development of artificial intelligence is inseparable from data and models. However, there are few publicly available gastrologic data sets, and even fewer data sets that meet the research needs. Obtaining an accurately labeled self-built gastrological data set is our first task before we start our research. In addition, stomach disease algorithms based on convolutional neural network and Transformer architecture at home and abroad also have pain points of low accuracy and high operation cost. Therefore, this paper carried out research on the classification model of gastric cancer and gastritis, and completed the engineering landing of the algorithm to truly empower pathologists.The work and innovation of this paper can be summarized as follows:1. Construct a large gastric pathology data set with cell-level feature labeling. A total of 930 gastric digital pathological images were collected in the dataset constructed in this paper. Among them, 610 were gastritis and 122 were normal mucosa. A total of 198 digital pathological images of gastric cancer were collected.2. Innovative gastric cancer classification model based on deep learning. In this paper, a new gastric cancer classification model is proposed, which consists of the following two parts: (1) A novel positioning network based on residual connection network is proposed to solve the complex computation caused by the attention mechanism coding module when processing large-scale graphic data. (2) Improvement of vector centralization based on embedded location information.3. Innovative gastritis classification model based on deep learning. The innovative gastritis classification model proposed in this paper consists of the following two parts: (1) A multi-scale feature encoder based on void convolution and multi-size convolution kernel. (2) Decoder based on sampling and feature aggregation on the feature map.4. Development of clinical auxiliary diagnosis and treatment system for gastric diseases based on deep learning and PyQtIn this study, in order to realize the engineering landing of the algorithm, the auxiliary diagnosis and treatment software was developed. A complete practical application flow of pathological section assisted diagnosis was established, and image management and operation functions such as model deployment, model prediction and image preprocessing were realized.Key words: digital pathological image; Gastric cancer and gastritis; Image processing; Engineering software; Deep learningKeywords: Digital pathological images; Gastric cancer and gastritis; Image processing; Engineering software; Deep learning