报告题目：Data-Efficient Ophthalmic Medical Image Analysis
报告人：唐晓颖 副研究员 南方科技大学
报告时间：2022年6月22日 上午10：00-11：00
报告地点：十教410会议室
报告摘要：Medical AI has achieved performance comparable to that of experts in many applications. However, these achievements rely heavily on a large number of expert annotations. Manual annotation is tedious, labor-intensive, and time-consuming; and a large number of unlabeled data have not been fully and effectively explored under the paradigm. Therefore, it is necessary to study data-efficient medical image analysis (MIA) algorithms. In this talk, we will share some of our solutions for ophthalmic MIA: (1) A unified vision transformer architecture (named Uni4Eye) based on masked image modeling paradigm for self-supervised pretraining; (2) A lesion-based contrastive learning paradigm for data-efficient diabetic retinopathy (DR) grading; (3) A one-shot anomaly detection framework (named LesionPaste) for DR anomaly detection.
唐晓颖，约翰霍普金斯大学博士，南方科技大学副研究员、博士生导师；深圳市海外高层次引进人才；卡内基梅隆大学电气与计算机工程系客座教授；约翰霍普金斯大学电气与计算机工程系客座教授；科技部重点研发计划课题负责人；国家自然科学基金面上项目及青年项目负责人；深圳市优秀青年基础研究项目负责人；深圳市基础研究面上项目负责人；MICCAI领域主席、分会场主席、本地主席；IEEE高级会员；主要研究方向为医学图像分析、模式识别、医学AI等。
