论著摘要 |【综述】基于深度学习的影像组学(DLR)及其在低级胶质瘤非侵入性IDH1预测中的用途(双语版)

2017-08-31 10:58:40 admin 6

Deep Learning based Radiomics (DLR) and its usage in noninvasive IDH1 prediction for low grade glioma.

发表日期:2017.6.14    来源:Sci Rep. 

作者:Li Z1Wang Y2,3Yu J4,5Guo Y1,6Cao W7.

作者介绍

    1.Department of Electronic Engineering, Fudan University, Shanghai, China.

    2.Department of Electronic Engineering, Fudan University, Shanghai, China. 

    3.Key laboratory of Medical Imaging Computing and Computer Assisted Intervention of Shanghai, Shanghai, China.

    4.Department of Electronic Engineering, Fudan University, Shanghai, China. 

    5.Key laboratory of Medical Imaging Computing and Computer Assisted Intervention of Shanghai, Shanghai, China.

    6.Key laboratory of Medical Imaging Computing and Computer Assisted Intervention of Shanghai, Shanghai, China.

    7.Department of micro-electronics, Fudan University, Shanghai, China.

摘要

开发了基于深度学习的影像组学(DLR),从多种磁共振(MR)图像中提取深层信息。在151例低度胶质瘤患者的资料组中验证了DLR预测异柠檬酸脱氢酶1IDH1)突变状态的表现。使用具有6个卷积层的改进卷积神经网络(CNN)结构和具有4096个神经元的完全连接层来分割肿瘤。与典型的影像组学方法一样,通过标准化CNN的最后卷积层的信息来获得图像特征,而非从分割图像计算图像特征。来自不同大小的图层的CNN特征通过Fisher矢量编码。从CNN获得维度大于1.6 * 104的高通量特征。配对t检验和F分数用于选择能够区分IDH1CNN特征。使用相同的资料组,正常影像组学方法的IDH1估值的工作特征曲线下面积(AUC)为86%,而对应DLRAUC92%。基于多模态MR图像的DLRIDH1估值的AUC进一步提高到95%。 DLR可以是一种从医学图像中提取深层信息的有力方式。

Deep learning-based radiomics (DLR) was developed to extract deep information from multiple modalities of magnetic resonance (MR) images. The performance of DLR for predicting the mutation status of isocitrate dehydrogenase 1 (IDH1) was validated in a dataset of 151 patients with low-grade glioma. A modified convolutional neural network (CNN) structure with 6 convolutional layers and a fully connected layer with 4096 neurons was used to segment tumors. Instead of calculating image features from segmented images, as typically performed for normal radiomics approaches, image features were obtained by normalizing the information of the last convolutional layers of the CNN. Fisher vector was used to encode the CNN features from image slices of different sizes. High-throughput features with dimensionality greater than 1.6*104 were obtained from the CNN. Paired t-tests and F-scores were used to select CNN features that were able to discriminate IDH1. With the same dataset, the area under the operating characteristic curve (AUC) of the normal radiomics method was 86% for IDH1 estimation, whereas for DLR the AUC was 92%. The AUC of IDH1 estimation was further improved to 95% using DLR based on multiple-modality MR images. DLR could be a powerful way to extract deep information from medical images.


慧影医疗科技(北京)有限公司

地点:北京市海淀区中关村东升科技园B2-C103

电话:400-890-9020

邮箱:radcloud@huiyihuiying.com

关闭
图片
图片
  • 人工智能诊断云平台