论著摘要 |【多模态】治疗前18 F-FDG摄取数据与CT纹理特征联合用于放射性肺炎诊断的影像组学模型(双语版)

2017-09-01 10:40:17 admin 8

Incorporation of pre-therapy 18 F-FDG uptake data with CT texture features into a radiomicsmodel for radiation pneumonitis diagnosis.

发表日期:2017.6.15   来源:Med Phys. 

作者:Anthony GJ1Cunliffe A1Castillo R2Pham N3Guerrero T4Armato SG 3rd1Al-Hallaq HA5.

作者介绍

    1.Department of Radiology, The University of Chicago, Chicago, IL, USA.

    2.Department of Radiation Oncology, The University of Texas Medical Branch, Galveston, TX, USA.

    3.Baylor College of Medicine, Houston, TX, USA.

    4.Department of Radiation Oncology, Oakland University William Beaumont School of Medicine, Royal Oak, MI, USA.

    5.Department of Radiation and Cellular Oncology, The University of Chicago, Chicago, IL, USA.

目的

为了确定从PET扫描到CT肺纹理特征标准摄取值(SUV)的添加是否能改善放射治疗患者放射性肺炎(RP)诊断的影像组学基础模型。

To determine whether the addition of standardized uptake value (SUV) from PET scans to CT lung texture features could improve a radiomics-based model of radiation pneumonitis (RP) diagnosis in patients undergoing radiotherapy.

方法和材料

收集了96例食管癌患者的匿名数据(18例≥2RP阳性病例),包括治疗前PET / CT扫描,治疗前/治疗后诊断CT扫描和RP状态。从诊断CT扫描计算20个纹理特征(一阶,分形,法则滤波器和灰度共生矩阵),并与肺的解剖匹配区域进行比较。通过计算接收器操作特性曲线(AUC)下面积来评估分类器性能(纹理,SUV,或组合)。对于每个纹理特征,创建由纹理特征值的平均变化和治疗前SUV标准偏差(SUVSD)组成的逻辑回归分类器,并将其与使用ANOVA进行多重比较校正(P < 0.0025)的仅有纹理特征的分类器进行比较。

Anonymized data from 96 esophageal cancer patients (18 RP-positive cases of Grade ≥ 2) were collected including pre-therapy PET/CT scans, pre-/post-therapy diagnostic CT scans and RP status. Twenty texture features (first-order, fractal, Laws' filter and gray-level co-occurrence matrix) were calculated from diagnostic CT scans and compared in anatomically matched regions of the lung. Classifier performance (texture, SUV, or combination) was assessed by calculating the area under the receiver operating characteristic curve (AUC). For each texture feature, logistic regression classifiers consisting of the average change in texture feature value and the pre-therapy SUV standard deviation (SUVSD ) were created and compared with the texture feature as a lone classifier using ANOVA with correction for multiple comparisons (P < 0.0025).

结果

虽然临床参数(平均肺剂量,吸烟史,肿瘤位置)在有或无症状RP患者中没有显着差异,但SUV和纹理参数与RP状态显著相关。单一纹理特征分类器的AUC分别在高剂量(≥30Gy)和低剂量(<10Gy)区域的0.580.810.530.71之间。单独SUVSDAUC0.6995%置信区间:0.54-0.83)。将SUVSD添加到逻辑回归模型中显著改善了181411纹理特征的模型拟合,并且分别在低,中,高剂量区域中将特征上的平均AUC增加了0.08,0.060.04

While clinical parameters (mean lung dose, smoking history, tumor location) were not significantly different among patients with and without symptomatic RP, SUV and texture parameters were significantly associated with RP status. AUC for single-texture feature classifiers alone ranged from 0.58 to 0.81 and 0.53 to 0.71 in high-dose (≥ 30 Gy) and low-dose (< 10 Gy) regions of the lungs, respectively. AUC for SUVSD alone was 0.69 (95% confidence interval: 0.54-0.83). Adding SUVSD into a logistic regression model significantly improved model fit for 18, 14 and 11 texture features and increased the mean AUC across features by 0.08, 0.06, and 0.04 in the low-, medium-, and high-dose regions, respectively.

结论

SUVSD添加到单一纹理特征可以平均提高分类器性能,但是当将SUVSD添加到已经有效的单独使用纹理的分类器时,改进的幅度变小。这些研究结果表明使用多种成像模式的信息有更准确地评估RP的潜能。



Addition of SUVSD to a single-texture feature improves classifier performance on average, but the improvement is smaller in magnitude when SUVSD is added to an already effective classifier using texture alone. These findings demonstrate the potential for more accurate assessment of RP using information from multiple imaging modalities.

关键词

计算机断层扫描,正电子发射断层扫描,标准摄取值,放射性肺炎,影像组学,纹理分析

CT; PET; SUV; radiation pneumonitis; radiomics; texture analysis


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