![]()
神外前沿讯，美国国密歇根大学和加利福尼亚大学旧金山分校的研究人员在最新出版的学术顶刊Nature杂志主刊（影响因子64.8）上，发表了人工智能技术和光学成像技术在脑胶质瘤手术中应用的最新进展，其研发FastGlioma模型在检测术中肿瘤浸润情况方面，远远优于目前的影像引导和荧光引导辅助手段，这有望给神经系统肿瘤手术带来革命性进展。![]()
FastGlioma互动演示链接：https://fastglioma.mlins.org（可点击阅读原文进入）Foundation models for fast, label-free detection of glioma infiltration
https://www.nature.com/articles/s41586-024-08169-3
Nature (2024)Cite this article
作者：Akhil Kondepudi, Melike Pekmezci, Xinhai Hou, Katie Scotford, Cheng Jiang, Akshay Rao, Edward S. Harake, Asadur Chowdury, Wajd Al-Holou, Lin Wang, Aditya Pandey, Pedro R. Lowenstein, Maria G. Castro, Lisa Irina Koerner, Thomas Roetzer-Pejrimovsky, Georg Widhalm, Sandra Camelo-Piragua, Misha Movahed-Ezazi, Daniel A. Orringer, Honglak Lee, Christian Freudiger, Mitchel Berger, Shawn Hervey-Jumper & Todd Hollon
根据论文信息显示，脑胶质瘤（ glioma）治疗中的一个关键挑战是在手术过程中检测肿瘤浸润情况，以实现安全的最大程度切除。但遗憾的是，在大多数胶质瘤患者术后都发现有可安全切除的残留肿瘤，这导致了肿瘤的早期复发以及患者生存率的下降 。为此，该研究团队开发出一款名为FastGlioma的人工智能（AI）模型。在神经外科手术中，FastGlioma将显微光学成像和人工智能相结合，用10秒就判断出是否还有残留的脑胶质瘤。在针对弥漫性胶质瘤患者的前瞻性、多中心、国际性测试队列临床试验中，FastGlioma检测并计算肿瘤残余量准确率高达平均约92%，而且仅在3.8%的情况下遗漏了高风险肿瘤残余，而利用图像和荧光引导的手术遗漏率接近25%。研究人员使用超过11000份手术样本和400万个显微图像对视觉基础模型进行了预训练。这些肿瘤样本通过受激拉曼组织成像拍摄。使用受激拉曼组织成像获取全分辨率图像大约需要100秒；而“快速模式”下的低分辨率图像则仅需10秒。利用该模型还能最大程度减少对放射成像、对比增强或荧光标记的依赖，并推广到其他脑肿瘤诊断中。![]()
A patient with a suspected diffuse glioma undergoes surgical resection. During tumour resection, the surgeon samples tissue from the surgical margin. The portable SRH imaging system acquires microscopy images in the operating room, performed by a single technician using simple touchscreen instructions. A freshly excised surgical specimen is loaded directly into a custom microscope slide and inserted into the SRH imager without the need for tissue processing. Additional details on image acquisition are provided in Extended Data Fig. 1. SRH images can be virtually stained using an H&E-like colour scheme for clinician review as shown above20. A whole-slide SRH image is divided into patches and each patch undergoes a feedforward pass through a patch tokenizer (Extended Data Fig. 3a). The patch tokens, plus an appended classification token <CLS>, are then input into a whole-slide SRH encoder that is a vision transformer. The patch tokenizer and whole-slide encoder are pretrained as a visual foundation model using large-scale self-supervision (Extended Data Fig. 3b). For tumour-infiltration scoring, a slide scorer model is fine-tuned to output a normalized continuous score between 0 and 1 that predicts the degree of tumour infiltration within the whole-slide image that corresponds to a four-tier whole-slide ordinal infiltration scale as defined by expert neuropathologists (Extended Data Figs. 2 and 4). Ordinal labels are weak because they apply to the slide level only. Despite the weak labels, FastGlioma provides regional interpretability by identifying areas within whole-slide SRH images with a high probability of tumour infiltration. 学术动态往期：
学术动态第235期 | 柳叶刀报道中药FYTF919治疗急性脑出血的临床试验结果
学术动态第234期 | 麻省理工通过7T功能磁共振绘制出最全大脑皮层功能图谱 发表在Neuron
学术动态第233期 | 华山医院神经外科合作发现胶质瘤相关癫痫发生关键机制 发表在神经科学顶刊Neuron杂志
学术动态第232期 | 内镜垂体手术的AI辅助解剖识别 Nature子刊报道
学术动态第231期 | 天坛医院谢坚团队提出岛叶胶质瘤新分型 发表在Cancer Medicine
学术动态报道第1-230期（2017-2024）
神外前沿-中国神经外科新媒体；
联系邮箱：shenwaiqianyan@qq.com
