Texture analysis as a radiomic marker for differentiating renal tumors

Purpose

To evaluate the utility of texture analysis for the differentiation of renal tumors, including the various renal cell carcinoma subtypes and oncocytoma.

Materials and methods

Following IRB approval, a retrospective analysis was performed, including all patients with pathology-proven renal tumors and an abdominal computed tomography (CT) examination. CT images of the tumors were manually segmented, and texture analysis of the segmented tumors was performed. A support vector machine (SVM) method was also applied to classify tumor types. Texture analysis results were compared to the various tumors and areas under the curve (AUC) were calculated. Similar calculations were performed with the SVM data.

Results

One hundred nineteen patients were included. Excellent discriminators of tumors were identified among the histogram-based features noting features skewness and kurtosis, which demonstrated AUCs of 0.91 and 0.93 (p < 0.0001), respectively, for differentiating clear cell subtype from oncocytoma. Histogram feature median demonstrated an AUC of 0.99 (p < 0.0001) for differentiating papillary subtype from oncocytoma and an AUC of 0.92 for differentiating oncocytoma from other tumors. Machine learning further improved the results achieving very good to excellent discrimination of tumor subtypes. The ability of machine learning to distinguish clear cell subtype from other tumors and papillary subtype from other tumors was excellent with AUCs of 0.91 and 0.92, respectively.

Conclusion

Texture analysis is a promising non-invasive tool for distinguishing renal tumors on CT images. These results were further improved upon application of machine learning, and support the further development of texture analysis as a quantitative biomarker for distinguishing various renal tumors.

     

Intravascular ultrasound area strain imaging used to characterize tissue components and assess vulnerability of atherosclerotic plaques in a rabbit model

The purpose of this study was to investigate the association of area strain and tissue components and vulnerability of atherosclerotic plaques in a rabbit model. Forty purebred New Zealand rabbits underwent balloon-induced abdominal aorta endothelium injury, then a high-cholesterol diet for 24 weeks. Intravascular ultrasound (IVUS) images of abdominal aortas were acquired in situ and two consecutive frames near the end-diastole were used to construct an IVUS elastogram. Histologic slices matched with corresponding IVUS images were stained for fatty and collagen components, smooth muscle cells (SMCs) and macrophages. Regions-of-interest (ROIs) in plaques were classified as fibrous, fibro-fatty or fatty according to histologic study. Vulnerability indexes of ROIs were calculated as (fat + macrophage)/(collagen + SMCs). The area strain of these ROIs was calculated by use of an in-house–designed software system with a block-matching–based algorithm. Area strain was significantly higher in fatty ROIs (0.056 ± 0.003) than in fibrous (0.019 ± 0.002, p < 0.001) or fibro-fatty ROIs (0.033 ± 0.003, p < 0.001). The sensitivity and specificity of area strain for fatty ROIs characterization was 75.0% and 80.2% (area under the curve [AUC] 0.858, 95% confidence interval [CI] = 0.800–0.916, p < 0.001) and 75.0% and 75.3% (AUC 0.859, 95% CI = 0.801–0.917, p < 0.001) for fibrous ROIs, as demonstrated by receiver operating characteristic curve analysis. Area strain was positively correlated with vulnerability index (r² = 0.495, p < 0.001), fatty components (r² = 0.332, p < 0.001) and macrophage infiltration (r² = 0.406, p < 0.001); and negatively correlated with collagen and SMC composition (r² = 0.115 and r² = 0.169, p < 0.001, respectively). Area strain calculation with IVUS elastography based on digital B-mode analysis is feasible and can be useful for tissue characterization and plaque vulnerability assessment.     

Gastrointestinal stromal tumors

Background We reviewed radiologic features of gastrointestinal stromal tumors (GISTs) and correlated them with clinical and pathologic findings. Methods We investigated a series of 39 c-Kit–positive GISTs. Clinical and radiologic findings and management of these patients were recorded. Results Twenty women and 19 men (mean age 64 years) had histologically proved GIST. Tumor locations were the small bowel (n = 20), stomach (n = 14), rectum (n = 4), and omentum (n = l). Symptoms at presentation were most frequently gastrointestinal bleeding (n = 14) and abdominal pain (n = l1). Tumors were classified as very low risk (n = 2), low risk (n = 10), intermediate risk (n = 12), and high risk (n = 11). Ultrasonography, computed tomography, magnetic resonance, digital subtraction angiography, and barium series were used in the evaluation of these tumors. Most tumors were seen as well-delineated soft tissue masses with heterogeneous contrast enhancement. Necrosis, calcification, and ulceration were most commonly seen in large tumors that presented a more aggressive behavior. Conclusion GISTs can arise anywhere in the gastrointestinal tract and present a great variety of clinical and radiologic features, depending mostly on size and location.     

Endosonographic evaluation of c-kit-positive gastrointestinal stromal tumor

BACKGROUND: Endosonographic features of c-kit-positive gastrointestinal stromal tumors (GISTs) were compared with those of leiomyomas and schwannomas. METHODS: Twenty-four patients with gastric mesenchymal tumors who underwent endoscopic ultrasonography (EUS) and surgical treatment were enrolled. GISTs were defined as c-kit (CD117)-positive tumors, leiomyomas as desmin-positive and c-kit-negative tumors, and schwannomas as S-100-positive and c-kit-negative tumors. Invasion to adjacent organs or more than 20 mitotic counts per 50 high power fields indicated malignancy. RESULTS: There were 19 GISTs, three leiomyomas, and two schwannomas. All five malignant tumors were GISTs. A marginal halo was found in 12 of 19 GISTs and in both of the schwannomas, but not in any of the three leiomyomas. The echogenicities of GISTs were low but higher than that of the normal proper muscle layer, whereas those of leiomyomas and schwannomas were usually low. Lobulation of the tumor surface was documented only in GISTs, particularly in malignant ones. The tumor doubling time of a malignant GIST was 9.3 months, and that of six benign GISTs was 18.7 months (range = 10.7-28.0 months). CONCLUSION: Marginal halo and relatively higher echogenicity on EUS might suggest GIST. Marginal lobulation and a short doubling time may be signs of a malignant GIST.     

Utility of endoscopic ultrasonography-guided fine-needle biopsy (EUS-FNB) for diagnosing small subepithelial lesions (< 20 mm)

AimSubepithelial lesions (SELs) are defined as being located under the mucosa. Presently, endoscopic ultrasound-guided fine-needle aspiration (EUS-FNA) is commonly performed to diagnose SELs. With the development of new puncture needles, endoscopic ultrasound-guided fine-needle biopsy (EUS-FNB), which allows for the acquisition of large tissue samples, has been proposed. However, studies on EUS-FNB of SELs measuring < 20 mm have not yielded satisfactory results. Therefore, we aimed to assess the performance and usefulness of EUS-FNB of SELs measuring less than < 20 mm.MethodsThe present study included 62 patients who underwent EUS-FNA or EUS-FNB for SELs at our hospital between January 2015 and March 2019. EUS-FNA was performed using fine-needle aspiration needles, and EUS-FNB was performed using fine-needle biopsy needles. These needles, which come in different shapes and diameters, were compared in terms of their usefulness in performing procedures for SELs measuring ≥ 20 mm and those measuring < 20 mm.ResultsFor SELs measuring ≥ 20 mm, the use of needles with a large diameter, such as 19 or 20 G, resulted in significantly improved diagnostic rates. For SELs measuring < 20 mm, the use of FNB needles showed significantly improved diagnostic rates, regardless of the size of the puncture needles.ConclusionEven when SELs are less than 20 mm, they might have malignant potential, and histological diagnosis may be desirable in some cases. EUS-FNB has an advantage over EUS-FNA in the diagnosis of SELs measuring < 20 mm.     

超声内镜在上消化道黏膜隆起型病变诊断和治疗中的应用

目的探讨超声内镜在上消化道黏膜隆起型病变的诊断和内镜治疗中的价值。方法运用超声内镜检查114例上消化道黏膜隆起型病变,并对其中26例进行超声内镜引导下病变切除术。结果黏膜隆起病变性质以间质瘤最常见,占51.8%,其次,壁外压迫占21.9%,较少见有恶性肿瘤、脂肪瘤、异位胰腺、囊肿等。对26例黏膜下肿瘤进行切除,超声诊断和病理符合率达77.0%。手术并发症少见。结论超声内镜可初步定性上消化道黏膜隆起型病变,对黏膜下肿瘤的治疗选择有指导作用;超声内镜引导下内镜切除黏膜下间质瘤安全、有效。     

内镜超声检查对上消化道隆起性病灶的诊断价值

目的：探讨内镜超声检查（EUS）在上消化道隆起性病灶的诊断与鉴别诊断中的价值。方法：回顾分析2002年1月至2006年6月本院1060例上消化道隆起性病变患者的EUS诊断结果，并与其手术、病理诊断及临床随访结果对照。同时比较EUS与CT检查在黏膜下肿瘤（SMTs）定位及定性诊断中的价值。结果：1060例上消化道隆起性病变中，发生于食管者355例（33．5％），胃618例（58．3％），十二指肠87例（8．2％）；EUS诊断为间质瘤260例（24．5％），平滑肌瘤221例（20．8％），脂肪瘤17例（1．6％），腔外压迹221例（20．8％），息肉94例（8．9％），恶性肿瘤96例（9．1％），囊肿57例（5.4％），血管瘤7例（0．7％），炎性皱襞45例f4．2％），静脉曲张21例（2．0％），血肿1例f0．1％）及异位胰腺20例（1．9％）。对EUS诊断SMTs的进一步分析发现，EUS对SMTs性质和来源层次的判断准确率达90．6％和86．9％；与CT比较，EUS在SMTs定位（95．8％比72．9％，P〈0．001）和定性（93．8％比64．6％，P〈0．001）诊断方面明显优于CT，在良、恶性鉴别诊断方面，CT组与EUS组间差异无统计学意2／．（91．7％比89．6％，P〉0．05）。结论：EUS对上消化道隆起性病灶诊断有较大价值，进一步为制定临床处理方案提供重要的影像学依据。     

基于CT影像组学模型鉴别肾乏脂肪血管平滑肌脂肪瘤与均质肾透明细胞癌

目的 基于增强CT影像组学特征联合临床特征建立综合模型，验证其术前鉴别肾乏脂肪血管平滑肌脂肪瘤（fp-AML）与均质肾透明细胞癌（hd-ccRCC）的效能。方法 回顾性分析经病理证实的32例fp-AML与39例hd-ccRCC。在增强CT皮质期、实质期及排泄期图像手工勾画肿瘤ROI，提取影像特征，计算观察者间及观察者内组内相关系数（ICC），采用LASSO回归进行特征选择，通过Logistic多元回归分析构建回归方程，并计算皮质期、实质期、排泄期及三期联合的影像组学得分。通过Logistic多元回归分析建立综合模型，并绘制列线图。采用Hosmer-Lemeshow拟合优度检验评价列线图的拟合度，以ROC曲线分析检测列线图的鉴别效能，以决策曲线评价列线图鉴别fp-AML和hd-ccRCC的净获益。结果 自各期图像中提取出包括强度、形状、纹理、图像滤波在内共1 029个特征，对ICC均>0.75的特征进行LASSO选择，分别于皮质期、实质期、排泄期和三期联合获得6、6、5和7个有鉴别意义的特征，AUC分别为0.83[95%CI（0.73，0.92）]、0.80[95%CI（0.70，0.91）]、0.78[95%CI（0.68，0.89）]和0.86[95%CI（0.77，0.95）]。基于三期联合影像组学得分和临床特征的列线图的AUC为0.90[95%CI（0.81，0.99）]，以之术前鉴别fp-AML和hd-ccRCC可获得较满意的诊断净获益。结论 通过列线图表示基于增强CT影像组学特征联合临床特征建立的综合模型术前鉴别fp-AML与hd-ccRCC具有较高诊断效能，有助于术前定性诊断肾肿瘤。     

应用计算机人工神经网络系统结合CT与PET鉴别良、恶性肺结节

目的 利用肺结节CT、PET特征,开发计算机人工神经网络(ANN)辅助诊断系统,评价其对肺结节良恶性的鉴别能力。方法 连续收集112例肺内单发小结节(<3.0 cm)患者,均接受PET/CT及胸部CT检查,二者间隔小于1个月。112例患者中恶性肺结节52例,良性60例,均经组织学或临床随诊证实。利用结节的CT特征及PET特征开发计算机ANN辅助诊断系统。计算机ANN的训练及测试采用Round-Robin方法。采用ROC方法评价计算机ANN输出结果并进行统计学分析。结果 CT计算机ANN程序采用20个输入单元,包括4个临床特征及16个CT特征,ROC曲线下面积(Az)为0.83;PET计算机ANN程序采用4个临床特征及1个PET特征作为5个输入单元,Az值为0.91;CT+PET计算机ANN程序采用临床特征CT及PET所有21个输入单元,Az值为0.95。与CT计算机ANN程序、PET计算机ANN程序相比,CT+PET计算机ANN程序输出结果明显提高(P＝0.015、0.037)。 结论 CT+PET ANN计算机辅助诊断程序输出结果优于单纯PET或CT计算机ANN结果。当PET对肺结节诊断有困难时,结节的CT特征有助于鉴别诊断。     

超声内镜检查术诊断消化道黏膜下病变的回顾性分析

目的 评估超声内镜检查术（EUS）诊断消化道黏膜下病变（SML）的临床价值。方法 回顾性分析2020年1月-2021年12月该院消化内科使用EUS诊断,并经组织病理学证实为SML患者的病例资料。结果 共有142例消化道SML患者进行了EUS检查。其中,135例为实体瘤,7例为壁外压迫。实体瘤中,EUS诊断黏膜层病变14例,黏膜肌层病变20例,黏膜下层病变62例,固有肌层病变39例,其诊断准确率分别为：100.0%、100.0%、95.0%和89.7%。EUS诊断平滑肌瘤30例（21.1%）,间质瘤29例（20.4%）,神经内分泌肿瘤25例（17.6%）,异位胰腺15例（10.6%）,息肉14例（9.9%）,脂肪瘤11例（7.7%）,囊肿9例（6.3%）,壁外压迫7例（4.9%）,颗粒细胞瘤2例（1.4%）。107例行内镜治疗或手术切除后送病检,99例病理与EUS诊断相符,EUS总体诊断准确率为92.5%。结论 EUS对消化道SML诊断的准确性与病变的起源有关,起源于黏膜层与黏膜肌层的诊断准确性最高,在鉴别壁内病变与壁外压迫方面,也具有较好的诊断价值。EUS在一定程度上可以判断SML的性...     

胃肠道间质瘤病理危险度分级的CT放射组学模型研究

目的 构建胃肠道间质瘤(GISTs)病理危险度分级的CT放射学模型并验证其效果.方法 回顾性分析沧州市中心医院收治的140例GISTs患者术前动脉期CT图像.将患者依据2008年美国国立卫生研究院(NIH)分类标准分为极低危组(n=8)、低危组(n=47)、中危组(n=33)和高危组(n=52)4组,并将患者按随机数字...     

不同阈值勾画18F-FDG PET/CT心脏肿瘤ROI鉴别其良恶性

目的 评价采用不同阈值勾画¹⁸F-FDG PET/CT图像中病灶ROI鉴别心脏良恶性肿瘤的价值。方法 纳入64例心脏肿瘤患者（65个病灶）,根据病理学或长期随访结果分为良性组（n=27）和恶性组（n=38）。采用3种阈值（Th 2.5、Th 40%和Th bgd）勾画心脏肿物,测量并比较病灶¹⁸F-FDG PET/CT代谢和形态学参数,包括最大标准摄取值（SUV_max）、平均标准摄取值（SUV_mean）、代谢肿瘤体积（MTV）、病灶糖酵解总量（TLG）、最大CT （CT_max）值、平均CT （CT_mean）值及CT值中位数（CT_median）。以多因素Logistic回归分析良、恶性组差异具有统计学意义的参数,绘制受试者工作特征（ROC）曲线,计算曲线下面积（AUC）,评价3种阈值诊断心脏良恶性肿瘤的效能。结果 采用3种阈值勾画ROI得到的肿瘤SUV_max一致,其鉴别诊断心脏良恶性肿瘤的AUC均为0.92;良、恶性肿瘤之间SUV_mean、MTV及TLG差异均具有统计学意义（P均<0.05）。根据SUV_mean诊断良、恶性肿瘤时,以阈值Th 40%的诊断效能最佳;以MTV及TLG诊断时,阈值Th 2.5均具最佳效能。多因素Logistic回归结果显示Th 40%与Th bgd诊断效能均较好,且Th bgd的敏感度及准确率均高于Th 40%。结论 根据PET代谢参数可鉴别心脏良恶性肿瘤。SUV_max截断值为6.75时,鉴别心脏良恶性肿瘤具有较好效能;采用多参数评价时,3种阈值中,以Th bgd效能最优。     

Gastrointestinal Stromal Tumors

Objective. The purpose of this series was to determine the spectrum of findings on gray scale trans‐abdominal ultrasonography (TAUS) in pathologically proven cases of primary gastrointestinal stromal tumors (GISTs) and correlate them with gross morphologic and pathologic findings. Methods. The series included 18 patients with a primary GIST tumor detected on preoperative TAUS. The ultrasonographic findings were evaluated for features such as tumor size, shape, margin, echogenicity, and presence of fluid components, and the features were compared with morphologic and pathologic findings. Results. All of the primary GISTs were hypoechoic extraluminal masses with well‐delineated margins. Eight GISTs were homogeneously solid masses, and 8 were heterogeneously solid masses that contained a large central area of lower echogenicity (n = 4) or multiple internal hypoechoic irregular spaces (n = 4) corresponding to necrosis and hemorrhage. Other tumors had a cystic appearance (n = 1) or showed a dual hyperechoichypoechoic echo structure (n = 1). Three tumors showed intratumoral gas due to fistulization into the bowel lumen, which appeared as hyperechoic foci or a linear hyperechoic area with acoustic shadowing. The heterogeneous tumors were significantly larger (P = .03) and had higher mitotic counts (P = .05). Gastrointestinal stromal tumors with high malignant potential tended to be large and showed intratumoral heterogenicity with areas of lower echogenicity. Conclusions. Gastrointestinal stromal tumors showed varied patterns on TAUS. The ultrasonographic pattern depended on the tumor size and mitotic activity. Ultrasonographic features suggesting high malignant potential were size and internal heterogenicity with the presence of intratumoral hypoechoic areas.     

MRI纹理分析评价注意缺陷多动障碍

目的 观察MRI纹理分析诊断注意缺陷多动障碍(ADHD)及分型的效果.方法 基于纽约大学医学中心公开MRI数据选取88例ADHD患者(ADHD组)及67名健康受试者(对照组),将ADHD组分为注意力缺陷为主型(ADHD-D亚组(n=32)和混合型(ADHD-C)亚组(n=56),提取并比较受试者脑白质和脑灰质的纹理特征...     

基于颅脑T1WI对比增强图像构建卷积神经网络模型鉴别肺癌与乳腺癌脑转移

目的 分析基于对比增强颅脑T1WI(T1CE)构建的卷积神经网络(CNN)模型鉴别肺癌与乳腺癌脑转移的效能。方法 回顾性分析97例经手术病理证实的乳腺癌脑转移(39例)与肺癌脑转移(58例)患者，于颅脑T1CE中手动勾画病灶ROI并提取其影像组学特征，采用单因素分析及最小绝对收缩和选择(LASSO)算法进行特征降维，筛选最优特征；分别构建传统机器学习支持向量机(SVM)、随机梯度下降(SGD)、K邻近(KNN)、决策树(DT)、随机森林(RF)、极端随机树(ET)、逻辑回归(LR)模型及深度学习CNN模型，将按7∶3数据分为训练集和验证集，采用受试者工作特征(ROC)曲线评估8种模型鉴别验证集中肺癌与乳腺癌脑转移的效能。结果 共纳入202个脑转移癌，含乳腺癌、肺癌脑转移各101个。基于颅脑T1CE提取1 050个特征，经单因素分析及LASSO算法降维后得到5个最优特征；以之构建的SVM、SGD、KNN、DT、RF、ET、LR及CNN模型鉴别验证集肺癌与乳腺癌脑转移的曲线下面积(AUC)分别为0.88、0.83、0.87、0.74、0.84、0.86、0.88及0.90,其中CNN模型的...     

CT纹理分析鉴别诊断胰腺导管腺癌、胰腺神经内分泌肿瘤及实性假乳头状肿瘤

目的 探讨CT纹理特征诊断及鉴别诊断胰腺导管腺癌（PDAC）、胰腺神经内分泌肿瘤（PNET）及实性假乳头状瘤（SPTP）的可行性。方法 回顾性分析经病理证实的98例PDAC、62例SPTP及39例PNET患者的CT资料，于肿瘤横断面最大层面沿肿瘤边界手动勾画ROI，提取46个CT纹理特征。按二分类（PDAC vs rest；SPTP vs rest；PNET vs rest）和三分类（PDAC vs SPTP vs PNET）分组方式将数据分组。以单因素回归分析每个纹理特征鉴别二分类各组的诊断效能，并计算AUC；基于随机森林算法选择特征后，采用6种机器学习分类器（LDA、K-NN、RF、Adabost、NB、NN）对二分类和三分类分组进行分类，以多因素回归分析分类器的诊断效能，基于十折交叉验证标准计算AUC。结果 采用单个纹理特征鉴别胰腺肿瘤时，低密度短域补偿和灰度不均匀性分别对PDAC vs rest和SPTP vs rest有较好鉴别能力（AUC=0.73、0.79，P<0.01），而总和均值对PNET vs rest具有极好鉴别能力（AUC=0.90，P<0.01）。分类器鉴别PDAC vs rest、SPTP vs rest、PNET vs rest的诊断效能很好或极好，最大AUC分别为0.88（RF）、0.86（RF）和0.94（Adaboost）。分类器鉴别三分类分组的准确率均较好，以RF最高（0.80）。结论 CT纹理分析可鉴别PDAC、SPTP和PNET；采用机器学习算法可进一步提高鉴别诊断效能。     

基于非下采样双树复轮廓波变换的小波纹理特征识别肺良恶性结节CT图像

目的 观察基于非下采样双树复轮廓波变换（NSDTCT）的小波纹理特征在识别肺良恶性结节CT图像中的应用价值。方法 从肺结节患者的CT图像中分别提取基于NSDTCT和基于Contourlet变换的小波纹理参数,对高维纹理参数采用单因素分析、Lasso回归等方法进行降维。对降维后的纹理参数分别构建诊断良恶性肺结节的支持向量机分类诊断模型,绘制ROC曲线,比较2种方法的诊断效能。结果 采用NSDTCT方法,基于经Lasso降维且自变量数目较少的纹理参数构建的诊断模型分类效果最好,判断良恶性肺结节的准确率为98.37%,AUC为1.00;采用Contourlet变换方法,基于全部提取纹理参数构建的模型分类效果最好,诊断准确率为56.05%,AUC为0.73;2个模型的ROC曲线的AUC差异有统计学意义（Z=6.430,P<0.001）。结论 基于NSDTCT的纹理分析方法对判断良恶性肺结节的准确性较高。     

MRI纹理分析预测乳腺癌腋窝淋巴结转移

目的 评价MRI纹理分析预测乳腺癌腋窝淋巴结(ALN)转移的价值.方法 以172例乳腺癌为训练组,分为ALN阳性亚组(n=79)和阴性亚组(n=93),提取增强MRI及ADC图中ALN纹理特征,比较2亚组间纹理特征差异;筛选纹理特征,构建多因素Logistic回归模型,并对模型进行内部验证.以另外37例乳腺癌为测试组,...     

3.0T磁共振T1WI对比增强图像纹理分析鉴别眼眶淋巴瘤与炎性假瘤

目的:探讨纹理分析在鉴别眼眶淋巴瘤与炎性假瘤中的价值。方法:纳入经手术病理证实的眼眶淋巴瘤25例和炎性假瘤24例,使用MaZda软件提取T 1WI对比增强(contrast-enhanced T 1-weighted images,CE-T 1W)肿瘤纹理特征。比较两组游程长矩阵(run-length matrix,RLM)的20个纹理特征参数,对差异有统计意义的参数进行主成分分析(principal component analysis,PCA)。采用多变量logistic回归模型分别对在两组间差异有统计学意义的特征参数及主成分进行建模,绘制受试者工作特征曲线(receiver operating characteristic curve,ROC)以评价模型效能。结果:淋巴瘤组灰度不均匀度(grey level non-uniformity,GLNU)、长游程优势(long run emphasis,LRE)的4个参数高于炎性假瘤(P<0.001),短游程因子(short run emphasis,SRE)、游程图像分数(fraction of image in runs,FIR)的4个参数低于炎性假瘤组(P<0.001);游程长不均匀度(run length non-uniformity,RLNU)的4个参数在两组间差异无统计学意义(P>0.05)。在两组间差异有统计学意义的参数中,以GLNU的参数建立的多变量logistic回归模型的诊断效能最高,灵敏度、特异度及曲线下面积(area under curve,AUC)分别为95.9%、72.0%、0.905;PCA提取出的2个主成分建立的多变量logistic回归模型的灵敏度、特异度及AUC分别为87.5%、88.0%、0.913。结论:基于CE-T 1W的RLM特征可有效鉴别眼眶淋巴瘤与炎性假瘤,其中以GLNU的诊断效能最高。