文生图扩散模型技术在病理学教学中的应用  

Application of text-to-image diffusion model technology in the teaching of pathology

作　　者：陈张一 朱朝阳[1] 危晓莉[1] 彭慧琴[1] CHEN Zhangyi;ZHU Chaoyang;WEI Xiaoli;PENG Huiqin(Basic Medical Experimental Teaching Center,Zhejiang University,Hangzhou 310058,China)

机构地区：[1]浙江大学基础医学实验教学中心,杭州310058

出　　处：《基础医学教育》2024年第11期980-985,共6页Basic Medical Education

基　　金：浙江大学首批AI For Education系列实证教学研究一般基金资助项目。

摘　　要：目前,病理教学中切片图像需要经过取材、固定、脱水透明、浸蜡、包埋、切片、染色和封片等步骤,操作复杂且容易质量不高,较为短缺。为解决这一问题,应用文生图技术制作数字病理切片图像。文生图技术是AIGC的一个重要分支,通过CLIP技术将提示词与病理图像进行关联,并运用Stable Diffusion模型自动随机生成病理切片图像,并有效提高生成图像速度,减少图像学习时间。通过算例分析,应用SD1.5预训练模型的微调技术可以有效提高病理切片图像质量和生成速度,生成的图库已能应用于学生的读片训练和随堂测试。The slice images used in the teaching of pathology are processed in the steps of sampling,fixation,dehydration and transparency,wax immersion,embedding,slicing,staining,and sealing.The procedure is complex and prone to generate low-quality images.Therefore,pathological sections are in short supply.To solve this problem,the authors used technology of text-to-image,a branch of AIGC(Artificial Intelligence Generated Content),to produce digital pathological images.They used CLIP(Contrastive Language-Image Pretraining)technology to associate prompt words with pathological images,and then used the Stable Diffusion model to automatically generate pathological slice images randomly,improving the speed of image generation and reducing image learning time.Based on the analysis of the examples,the fine-tuning technique of SD1.5 pre-training model can be used to improve the quality and generation speed of pathological section images.The generated image library has been applied to students’training of reading images and class test.

关 键 词：病理学 病理切片 教学改革 文生图 图像学习 Stable Diffusion模型 

分 类 号：R36[医药卫生—病理学]