News Release

人工神经网络的脑出血探测

Peer-Reviewed Publication

Proceedings of the National Academy of Sciences

Deep Learning Algorithm

image: A deep learning algorithm recognizes abnormal CT scans of the head in neurological emergencies such as aneurysm rupture with accuracy similar to highly-trained specialist physicians. The images show the same head CT scan, with the image on the right highlighted to show abnormalities identified by the algorithm. The algorithm also classifies the pathological subtype of each abnormality (red = subarachnoid hemorrhage). view more 

Credit: Image courtesy of Weicheng Kuo, Christian H?ne, Pratik Mukherjee, Jitendra Malik, and Esther Lim Yuh

科研人员报告了一个算法,它可能有潜力改善对神经急症的诊断。对头部的计算机断层成像(CT)扫描被用于迅速识别诸如创伤性脑外伤、中风以及破裂动脉瘤等病。为了解释这类图像,受过训练的人类专家在含有大量像素的有噪声的、低对比度3D堆叠灰阶图像中寻找常常微小而微妙的异常情况。Jitendra Malik、Esther Yuh及其同事开发了一个神经网络PatchFCN,并且用4396个CT扫描训练了这个网络,从而探测异常,其准确性类似于人类专家。这个迅速评估整个头部的算法以4名经过认证的放射科医师作为基准进行了检测。在200次扫描组成的一个测试样本中,这组作者报告说,PatchFCN的性能类似于放射科医师的准确性,在某些案例中识别出了这些放射科医师遗漏的异常情况。这组作者说,这个算法表现出了高的准确性、进行异常情况的像素级描绘的能力,以及把异常情况归类为不同的病理亚型的能力。

###


Disclaimer: AAAS and EurekAlert! are not responsible for the accuracy of news releases posted to EurekAlert! by contributing institutions or for the use of any information through the EurekAlert system.