Prediction of IDH-1 mutation status in WHO grade Ⅱ and Ⅲ gliomas by radiomics combined with T1-weighted contrast-enhanced image
-
摘要:
目的 探讨T1加权对比增强成像(T1-weighted contrast-enhanced image, T1CE)的影像组学特征及临床相关参数预测Ⅱ、Ⅲ级胶质瘤的异柠檬酸脱氢酶-1(isocitrate dehydrogenase-1, IDH-1)基因突变的诊断效能。 方法 选择空军军医大学第二附属医院2017年1月—2019年7月间经手术病理证实的135例患者Ⅱ、Ⅲ级胶质瘤的MRI资料(IDH-1野生型组51例，IDH-1突变型组84例)。利用ITK-SNAP在T1CE上手动绘制全肿瘤强化部分的感兴趣体积(volume of interest, VOI)，使用A.K.软件从VOI中提取1 044个影像组学特征。采用随机森林(random forest, RF)算法和5折交叉验证法验证影像组学模型以预测Ⅱ、Ⅲ级胶质瘤IDH-1突变的诊断效能。 结果 发病部位和结节状/环形强化在IDH-1突变型与IDH-1野生型组间差异有统计学意义(均P < 0.05)，2组间的其他MRI形态学特征差异无统计学意义。影像组学模型的ROC曲线下面积为0.794，灵敏度为61.4%，特异度为76.7%，准确性为70.9%。在RF分类器特征选择之后，选取前30个最优特征对IDH-1突变进行预测，与采用所有组学特征的效能相当，并且明显降低了冗余信息。 结论 影像组学联合T1CE能有效预测Ⅱ、Ⅲ级胶质瘤的IDH-1突变状态。RF分类器模型在预测IDH-1突变方面具有潜力，将有可能为胶质瘤患者早期诊断和个体化治疗方案提供影像学依据。 Abstract:Objective To explore the diagnostic efficiency of T1-weighted contrast-enhanced image (T1CE) radiomic features and clinical-related parameters in predicting isocitrate dehydrogenase-1 (IDH-1) gene mutations in WHO grade Ⅱ and Ⅲ gliomas. Methods MRI data of 135 patients with WHO grade Ⅱ and Ⅲ gliomas (51 cases in the IDH-1 wild type group and 84 cases in the IDH-1 mutant type group) confirmed by surgery and pathology from the Second Affiliated Hospital of Air Force Military Medical University between January 2017 and July 2019 were selected. The volume of interest (VOI) of the whole tumor-enhanced part was manually drawn on T1CE using ITK-SNAP, and 1 044 radiomic features from the VOI were extracted by using the software of A.K. software. Random forest (RF) algorithm and 5-fold cross-validation method were used to verify the radiomic model in predicting the diagnostic efficiency of IDH-1 mutations of WHO grade Ⅱ and Ⅲ gliomas. Results There were statistical differences in the location of the disease and nodular/ring enhancement between IDH-1 mutant and IDH-1 wild-type groups (all P < 0.05), but no significant differences in other MRI morphological characteristics between the two groups. The AUC value of the area under the ROC curve from the iradiomic model was 0.794, the sensitivity was 61.4%, the specificity was 76.7%, and the accuracy was 70.9%. After the feature selection of the RF classifier, the first 30 optimal features were selected to predict IDH-1 mutation, which had the same efficiency as all the radiomic features, and significantly reduced the redundant information. Conclusion Radiomics combined with T1CE can effectively predict the IDH-1 mutation status of WHO grade Ⅱ and Ⅲ gliomas. RF classifier model has the potential to predict IDH-1 mutations, which may provide an imaging basis for early diagnosis and individualized treatment of glioma patients. -
Key words:
- Gliomas /
- Isocitrate dehydrogenase /
- Radiomics /
- Random forest /
- Magnetic resonance imaging
-
图 1 RF预测Ⅱ、Ⅲ级胶质瘤IDH-1突变状态的ROC曲线
Figure 1. ROC curves for predicting IDH-1 mutation status in grade Ⅱ and Ⅲ gliomas using RF
图 2 Gini系数评估2组胶质瘤IDH-1突变状态特征
注：最能减少Gini系数杂质的特征是那些导致最小误分类的特征。
Figure 2. Assessment of IDH-1 mutation status of glioma in 2 groups using Gini coefficient
图 3 特征选择相关分析的影像组学热图
注：蓝色表示正相关，红色表示负相关。不同的颜色深度表示不同的相关系数值。
Figure 3. The heatmaps of correlation analysis for feature selection
表 1 2组胶质瘤患者的MRI图像特征比较[例(%)]
Table 1. Comparison of MRI imaging features between two groups of glioma patient [cases(%)]
项目 总数
(n=135)IDH-1野生型
(n=51)IDH-1突变型
(n=84)χ2值 P值 部位 40.091 < 0.001 额叶 62(45.93) 11(21.57) 51(60.71) 颞叶 31(22.96) 11(21.57) 20(23.81) 顶叶 9(6.67) 6(11.76) 3(3.57) 岛叶 9(6.67) 2(3.92) 7(8.33) 枕叶 1(0.74) 1(1.96) 0 其他 23(17.04) 20(39.22) 3(3.57) 信号 0.245 0.620 均匀 24(17.78) 8(15.69) 16(19.05) 不均匀 111(82.22) 43(84.31) 68(80.95) 是否跨越中线 2.793 0.095 否 117(86.67) 41(80.39) 76(90.48) 是 18(13.33) 10(19.61) 8(9.52) 是否多灶性 0.659 0.417 否 113(83.70) 41(80.39) 72(85.71) 是 22(16.30) 10(19.61) 12(14.29) 坏死 0.257 0.612 否 62(45.93) 22(43.14) 40(47.62) 是 73(54.07) 29(56.86) 44(52.38) 囊变 0.017 0.897 否 83(61.48) 31(60.78) 52(61.90) 是 52(38.52) 20(39.22) 32(38.10) 水肿 0.058 0.810 否 36(26.67) 13(25.49) 23(27.38) 是 99(73.33) 38(74.51) 61(72.62) 边界 0.505 0.477 光滑/清晰 25(18.52) 11(21.57) 14(16.67) 模糊/不规则 110(81.48) 40(78.43) 70(83.33) 强化方式 9.406 0.002 无/轻微 57(42.22) 13(25.49) 44(52.38) 结节状/环形 78(57.78) 38(74.51) 40(47.62) 
下载: 导出CSV
表 2 RF分类器检测IDH-1野生型与IDH-1突变型组间影像组学特征
Table 2. RF Classifier-based detection of radiomic features between IDH-1 wild-type and IDH-1 mutant-type groups
特征 Gini系数 IDH-1野生型组
M(P25, P75)IDH-1突变型组
M(P25, P75)Z值 P值 Zone Percentage 5.72 2.14(0.95, 2.99) 3.92(2.98, 5.21) -5.505 < 0.001 Cluster Shade_All Direction_offset1_SD 3.78 [5.83(3.25, 8.97)]×103 [2.41(1.29, 4.01)]×103 -4.597 < 0.001 Quantile0.975 3.67 [4.47(3.29, 6.29)]×10-4 [0.13(0.12, 0.15)]×10-4 -3.681 < 0.001 High Grey Level Run Emphasis_All Direction_offset5_SD 3.01 7.77(4.01, 13.22) 4.61(2.49, 7.31) -3.749 < 0.001 Small Area Emphasis 2.83 [9.82(9.78, 9.88)]×10-1 [9.79(9.76, 9.82)]×10-1 -3.587 < 0.001 Long Run High Grey Level Emphasis_All Direction_offset7_SD 2.71 200.31(146.63, 274.82) 345.20(195.19, 502.44) -3.531 < 0.001 Voxel Value Sum 2.35 [16.00(6.27, 39.55)]×106 [32.35(18.00, 51.65)]×106 -2.778 0.005 Low Intensity Large Area Emphasis 2.01 [2.21(1.01, 4.76)]×10-6 [1.94(1.43, 5.68)]×10-6 -0.855 0.392 Haralick Correlation_All Direction_offset3_SD 1.85 [7.11(2.44, 24.45)]×106 [2.97(1.48, 7.95)]×106 -2.748 0.006 Quantile0.025 1.67 [3.68(2.35, 6.55)]×102 [4.36(2.76, 5.43)]×102 -6.440 < 0.001 Long Run Low Grey Level Emphasis_All Direction_offset1_SD 1.58 [20.10(4.74, 65.70)]×10-6 [7.11(2.43, 23.40)]×10-6 -3.336 0.001 Short Run Emphasis_All Direction_offset3_SD 1.53 [2.02(1.25, 2.79)]×10-4 [1.58(1.21, 2.14)]×10-4 -1.870 0.062 age 1.49 41(28, 51) 43(33, 52) -1.090 0.276 uniformity 1.49 0.64(0.54, 0.72) 0.72(0.61, 0.78) -2.269 0.023 Long Run High Grey Level Emphasis_All Direction_offset2_SD 1.42 12.05(8.43, 17.83) 14.24(9.61, 22.44) -1.121 0.262 GLCM Entropy_angle90_offset8 1.37 7.88(6.59, 9.40) 8.08(6.83, 8.83) -0.054 0.957 Haralick Correlation_All Direction _offset9_SD 1.34 [13.80(4.36, 49.95)]×106 [5.56(2.49, 17.05)]×106 -3.048 0.002 Low Grey Level Run Emphasis_All Direction _offset8_SD 1.33 [7.49(4.17, 24.95)]×10-7 [4.19(1.78, 15.33)]×10-7 -2.909 0.004 High Grey Level Run Emphasis_All Direction_offset9_SD 1.30 5.94(3.72, 9.66) 3.78(2.43, 5.98) -3.408 0.001 Correlation_All Direction_offset9_SD 1.29 [5.59(3.29, 8.81)]×10-5 [8.05(5.06, 10.20)]×10-5 -0.254 0.799 GLCM Entropy_All Direction_offset4_SD 1.29 0.17(0.08, 0.28) 0.18(0.09, 0.29) -0.436 0.663 Min Intensity 1.26 63.00(13.00, 281.50) 61.00(10.50, 114.50) -1.405 1.600 Short Run Emphasis_All Direction _offset7_SD 1.24 [1.13(0.80, 1.53)]×10-3 [0.96(0.70, 1.28)]×10-3 -2.069 0.039 GLCM Entropy_All Direction _offset2_SD 1.22 0.12(0.06, 0.29) 0.16(0.08, 0.26) -0.449 0.653 Cluster Prominence_angle45_offset9 1.18 [5.12(2.27, 9.14)]×107 [2, 21(1.09, 5.47)]×107 -2.737 0.006 Haralick Correlation_angle45_offset9 1.15 [3.93(1.70, 7.68)]×108 [2.75(1.76, 7.86)]×108 -0.783 0.434 Cluster Shade_angle0_offset9 1.13 [69.78(-7.13, 236.04)]×103 [30.44(8.51, 108.40)]×103 -1.112 0.266 Long Run High Grey Level Emphasis_All Direction _offset3_SD 1.07 10.92(6.35, 17.19) 8.80(5.69, 12.44) -1.693 0.090 Inverse Difference Moment_All Direction _offset9_SD 1.04 [5.07(3.63, 6.72)]×10-3 [4.88(3.96, 5.99)]×10-3 -0.454 0.650 skewness 1.04 0.63(0.08, 1.07) 0.40(-0.08, 0.98) -1.085 0.278
下载: 导出CSV
-
[1] OSTROM Q T, PRICE M, NEFF C, et al. CBTRUS statistical report: primary brain and other central nervous system tumors diagnosed in the United States in 2015-2019[J]. Neuro Oncol, 2022, 24(Suppl 5): v1-v95. [2] 汪潮潮, 程哲, 常雪莲, 等. 肿瘤坏死因子α诱导蛋白3在胶质瘤中的表达特性及其对胶质瘤细胞侵袭和迁移的影响[J]. 中华全科医学, 2022, 20(5): 756-760. doi: 10.16766/j.cnki.issn.1674-4152.002447WANG C C, CHENG Z, CHANG X L, et al. Expression characteristics of tumor necrosis factor alpha inducible protein 3 in glioma and its effect on invasion and migration of glioma cell[J]. Chinese Journal of General Practice, 2022, 20(5): 756-760. doi: 10.16766/j.cnki.issn.1674-4152.002447 [3] 高丽萍, 曹风军. 大分割立体定向放疗联合重组人血管内皮抑制素治疗复发性脑胶质瘤的疗效分析[J]. 临床内科杂志, 2021, 38(07): 480-482.GAO L P, CAO F J. Efficacy analysis of large segmentation stereotactic radiotherapy combined with recombinant human endostatin in the treatment of recurrent brain glioma[J]. Journal of Clinical Internal Medicine, 2021, 38(07): 480-482. [4] LASOCKI A, ANJARI M, ORS KOKURCAN S, et al. Conventional MRI features of adult diffuse glioma molecular subtypes: a systematic review[J]. Neuroradiology, 2021, 63(3): 353-362. doi: 10.1007/s00234-020-02532-7 [5] PARK S I, SUH C H, GUENETTE J P, et al. The T2-FLAIR mismatch sign as a predictor of IDH-mutant, 1p/19q-noncodeleted lower-grade gliomas: a systematic review and diagnostic meta-analysis[J]. Eur Radiol, 2021, 31(7): 5289-5299. doi: 10.1007/s00330-020-07467-4 [6] WELLER M, VAN DEN BENT M, PREUSSER M, et al. EANO guidelines on the diagnosis and treatment of diffuse gliomas of adulthood[J]. Nat Rev Clin Oncol, 2021, 18(3): 170-186. doi: 10.1038/s41571-020-00447-z [7] 李欣, 谢继承, 王静, 等. 磁共振MRS、DWI及SWI序列在脑胶质瘤分级诊断中的应用价值[J]. 中华全科医学, 2022, 20(9): 1541-1544. doi: 10.16766/j.cnki.issn.1674-4152.002644LI X, XIE J C, WANG J, et al. The application value of magnetie resonance MRS, DWI and SWI sequences in the grading diagnosis of glioma[J]. Chinese Journal of General Practice, 2022, 20(9): 1541-1544. doi: 10.16766/j.cnki.issn.1674-4152.002644 [8] 侯仕强, 金春景, 石碑田, 等. 术前NLR、PLR和MLR在胶质瘤患者预后中的应用研究[J]. 中华全科医学, 2020, 18(7): 1118-1121. doi: 10.16766/j.cnki.issn.1674-4152.001443HOU S Q, JIN C J, SHI B T, et al. Application of preoperative NLR, PLR and MLR in the prognosis of patients with glioma[J]. Chinese Journal of General Practice, 2020, 18(7): 1118-1121. doi: 10.16766/j.cnki.issn.1674-4152.001443 [9] ABD-ELLAH M K, AWAD A I, KHALAF A A M, et al. A review on brain tumor diagnosis from MRI images: practical implications, key achievements, and lessons learned[J]. Magn Reson Imaging, 2019, 61: 300-318. doi: 10.1016/j.mri.2019.05.028 [10] ZHAO S S, FENG X L, HU Y C, et al. Better efficacy in differentiating WHO grade Ⅱ from Ⅲ oligodendrogliomas with machine-learning than radiologist' s reading from conventional T1 contrast-enhanced and fluid attenuated inversion recovery images[J]. BMC Neurology, 2020, 20(1): 48. doi: 10.1186/s12883-020-1613-y [11] IWAHASHI H, NAGASHIMA H, TANAKA K, et al. 2-Hydroxyglutarate magnetic resonance spectroscopy in adult brainstem glioma[J]. J Neurosurg, 2023, 139(2): 355-362. [12] DI STEFANO A L, NICHELLI L, BERZERO G, et al. In vivo 2-Hydroxyglutarate monitoring with edited MR spectroscopy for the follow-up of IDH-mutant diffuse gliomas: the IDASPE prospective study[J]. Neurology, 2023, 100(1): e94-e106. [13] CLÉMENT A, DOYEN M, FAUVELLE F, et al. In vivo characterization of physiological and metabolic changes related to isocitrate dehydrogenase 1 mutation expression by multiparametric MRI and MRS in a rat model with orthotopically grafted human-derived glioblastoma cell lines[J]. NMR Biomed, 2021, 34(6): e4490. DOI: 10.1002/nbm.4490. [14] CAO M Q, SUO S T, ZHANG X, et al. Qualitative and quantitative MRI analysis in IDH1 genotype prediction of lower-grade gliomas: a machine learning approach[J]. Biomed Res Int, 2021: 1235314. DOI: 10.1155/2021/1235314. [15] ZHANG Z, WEI Z Y, ZHAO L Y, et al. Assessing the clinical utility of multi-omics data for predicting serous ovarian cancer prognosis[J]. J Obstet Gynaecol, 2023, 43(1): 2171778. DOI: 10.1080/01443615.2023.2171778. [16] GALLO D M, FITZGERALD W, OMERO R, et al. Proteomic profile of extracellular vesicles in maternal plasma of women with fetal death[J]. J Matern Fetal Neonatal Med, 2023, 36(1): 2177529. DOI: 10.1080/14767058.2023.2177529. -
点击查看大图
图(3) / 表(2)
计量
- 文章访问数: 101
- HTML全文浏览量: 42
- PDF下载量: 7
- 被引次数: 0
