Core Viewpoint - The research introduces a novel multimodal vision-language model named AFLoc (Annotation-Free pathology Localization), which enables automatic localization of pathologies in medical images without the need for prior annotations from doctors, showcasing strong generalization capabilities that surpass human benchmarks in various pathology image tasks [4][9]. Group 1 - AFLoc is designed to perform pathology localization without requiring annotations, thus reducing the dependency on expert input [4][10]. - The model utilizes a contrastive learning approach based on a multi-level semantic structure, aligning diverse medical concepts with rich image features to adapt to the varied manifestations of pathologies [7][9]. - Initial experiments were conducted on a dataset of 220,000 pairs of chest X-ray images and reports, demonstrating that AFLoc outperforms current state-of-the-art methods in both localization and classification tasks without annotations [9]. Group 2 - The research validates AFLoc's generalization capabilities across different modalities, including histopathology and retinal fundus images, indicating its robustness in diverse clinical environments [9][10]. - The findings highlight AFLoc's potential to lower annotation requirements and adapt to complex clinical applications, making it a significant advancement in the field of medical imaging [10].

在胸片实验中，AFLoc在覆盖肺炎、胸腔积液、气胸等34种常见胸部疾病、涉及8个主流公开数据集的 测试中，在多项病灶定位指标上优于现有方法，并在多个病种中达到了甚至超越人类专家的水平。在眼 底影像和病理图像任务中，AFLoc同样展现出稳定的病灶定位能力，定位精度优于当前主流模型。 除病灶定位外，AFLoc还展现出强大的疾病诊断能力。在胸部X光、眼底和组织病理图像的零样本分类 任务中，其整体表现均优于现有方法。尤其在眼底视网膜病变诊断中，AFLoc的零样本分类性能甚至超 越了部分依赖人工标注数据微调的模型。 1月6日，中国科学院深圳先进技术研究院医学成像科学与技术系统全国重点实验室研究员王珊珊团队联 合清华大学助理教授周洪宇、澳门科技大学教授张康等合作者在《自然·生物医学工程》发表最新成 果，研究团队提出了一种名为AFLoc的人工智能模型，该模型不需要医生提前标注病灶，就能自动在医 学影像中"找病灶"。 研究团队介绍，"我们让AFLoc 模型同时学习两类信息，一类是医学影像本身，比如胸片、眼底照片或 病理切片，另一类是医生撰写的临床报告。通过反复'对照学习'，AFLoc模型会逐渐明白，临床报告中 提到的疾病描述 ...