Application of text-to-image diffusion model technology in the teaching of pathology
The slice images used in the teaching of pathology are processed in the steps of sampling,fixation,dehydration and trans-parency,wax immersion,embedding,slicing,staining,and sealing.The procedure is complex and prone to generate low-quality ima-ges.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 Lan-guage-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.
万方数据
