Enhanced CT Based Texture Analysis and Machine Learning for Differentiation between Adenolymphoma and Mixed Tumors of the Parotid Gland
-
摘要: 目的:探究基于增强CT纹理分析技术联合机器学习在腮腺腺淋巴瘤与混合瘤鉴别中的应用。方法:回顾性分析40例于本院手术并有完整病理资料的腮腺腺淋巴瘤与混合瘤患者,其中腺淋巴瘤组21例,混合瘤组19例。运用Mazda软件在增强CT静脉期图像上手动勾画病灶最大层面ROI区;应用Fisher系数、POE+ACC、MI及三者联合应用(FPM)的方法,筛选出最佳纹理参数,通过ROC曲线评估其诊断效能;最后采用RDA、PCA和LDA、NDA四种机器学习算法进行分类分析,并分析不同算法的诊断效能。结果:纹理特征参数中腺淋巴瘤组的WavEnHH_s-4、GrVariance、45dgr_Fraction低于混合瘤组,WavEnLL_s-4、GrSkewness高于混合瘤组,且均在组间有统计学意义。ROC曲线显示WavEnLL_s-4的敏感性与特异性较为平衡,AUC值、敏感性、特异性分别为0.797、84.2%、76.2%,具有良好诊断效能;RDA、PCA、LDA、NDA算法的误判率范围分别为30.0%~37.5%、30.0%~37.5%、7.5%~37.5%、5.0%~12.5%,其中误判率最低的是FPM联合NDA分类分析法,为5.0%;准确率、敏感性、特异性、阳性预测值、阴性预测值分别为95.0%、95.2%、94.7%、95.2%和94.7%,分类效能最佳。结论:增强CT纹理分析提取的最佳特征参数在腮腺腺淋巴瘤与混合瘤间具有显著差异,FPM联合NDA分类分析法误判率最低,有助于鉴别腮腺腺淋巴瘤与混合瘤。Abstract: Objective: To explore the application of enhanced computed tomography (CT)-based texture analysis combined with machine learning in the differential diagnosis of adenolymphomas and mixed tumors of the parotid gland. Methods: We retrospectively analyzed 21 and 19 cases of adenolymphomas and mixed tumors of the parotid gland, respectively. Regions of interest (ROI) were chosen on axial enhanced-CT images of the tumor’s maximum cross section using the Mazda software. The optimal texture parameters were selected using Fisher’s coefficient, probability of classification error and average correlation coefficients, mutual information, and a combination of the three. Receiver operating characteristic (ROC) curve analysis was performed to evaluate the diagnostic efficacy. Finally, the texture parameters were classified and analyzed using the following four machine-learning methods: raw data analysis, principal component analysis, linear discriminant analysis, and nonlinear discriminant analysis (NDA). The diagnostic efficiencies of these classification algorithms were analyzed. Results: WavEnHH_s-4, GrVariance, 45dgr_Fraction, WavEnLL_s-4, and GrSkewness were the statistically significant texture feature parameters for differentiating between parotid adenolymphomas and mixed tumors.. ROC curve analysis revealed that WavEnLL_s-4 had a balanced sensitivity and specificity, and the area under the curve, sensitivity, and specificity were 0.797, 84.2%, and 62.5%, respectively. The misclassification rate of NDA (5.0%–12.5%) was lower than that of the other algorithms. The NDA of FPM had the lowest misclassification rate (5.0%); its accuracy, sensitivity, specificity, positive predictive value, and negative predictive value were 95.0%, 95.2%, 94.7%, 95.2, and 94.7, respectively. Conclusion: The optimum enhanced CT-based texture features differed significantly between parotid adenolymphomas and mixed tumors. A combination of FPM and NDA had the lowest misclassification rate; it can contribute toward the identification of parotid adenolymphomas and mixed tumors.
-
Key words:
- enhanced CT /
- machine learning /
- texture analysis /
- parotid tumor
-
表 1 腮腺腺淋巴瘤与混合瘤间最佳纹理特征参数比较
Table 1. Comparison of the optimal texture feature parameters between parotid adenolymphomas and mixed tumors
参数 组别 统计检验 腺淋巴瘤组 混合瘤组 t/Z P WavEnHH_s-4 4.162±1.908 7.493±3.157 -4.084 <0.01 WavEnLL_s-4 21044.469±3887.164 16649.289±4309.226 3.392 0.002 GrVariance 0.185±0.046 0.236±0.033 -4.055 <0.01 GrSkewness 1.996±0.516 1.475±0.295 -3.291 0.001 45 dgr_Fraction 0.328±0.080 0.422±0.074 -3.854 <0.01 表 2 腮腺腺淋巴瘤与混合瘤间最佳纹理特征参数的诊断效能
Table 2. Diagnostic performance of the optimal texture feature parameters for parotid adenolymphomas and mixed tumors
参数 AUC 阈值 敏感性/% 特异性/% P WavEnHH_s-4 0.827 4.979 84.2 66.7 <0.01 WavEnLL_s-4 0.797 19227.148 84.2 76.2 0.001 GrVariance 0.815 0.200 89.5 66.7 0.001 GrSkewness 0.805 1.819 94.7 61.9 0.001 45 dgr_Fraction 0.802 0.384 73.7 71.4 0.001 表 3 腮腺腺淋巴瘤与混合瘤间不同机器学习算法的误判率
Table 3. False-positive rates of different machine-learning algorithms for parotid adenolymphomas and mixed tumors
组别 RDA/% PCA/% LDA/% NDA/% Fisher 37.5(15/40) 37.5(15/40) 10.0(4/40) 7.5(3/40) POE+ACC 35.0(14/40) 30.0(12/40) 22.5(9/40) 10.0(4/40) MI 30.0(12/40) 30.0(12/40) 37.5(15/40) 12.5(5/40) FPM 35.0(14/40) 32.5(13/40) 7.5(3/40) 5.0(2/40) 表 4 腮腺腺淋巴瘤与混合瘤间不同机器学习算法的效能比较
Table 4. Comparison of the performance of different machine-learning algorithms for parotid adenolymphomas and mixed tumors
分类算法 准确率/% 敏感性/% 特异性/% 阳性预测值 阴性预测值 Fisher/RDA 62.5 61.9 63.2 65.0 60.0 Fisher/PCA 62.5 61.9 63.2 65.0 60.0 Fisher/LDA 90.0 95.2 84.2 87.0 94.1 Fisher/NDA 92.5 90.5 94.7 95.0 90.0 POE+ACC/RDA 65.0 76.2 52.6 64.0 66.7 POE+ACC/PCA 70.0 76.2 63.2 69.6 70.6 POE+ACC/LDA 77.5 76.2 78.9 80.0 75.0 POE+ACC/NDA 90.0 85.7 94.7 94.7 85.7 MI/RDA 70.0 76.2 63.2 69.6 70.6 MI/PCA 70.0 76.2 63.2 69.6 70.6 MI/LDA 62.5 66.7 57.9 63.6 61.1 MI/NDA 87.5 81.0 94.7 94.4 81.8 FPM/RDA 65.0 66.7 63.2 66.7 63.2 FPM/PCA 67.5 71.4 63.2 68.2 66.7 FPM/LDA 92.5 95.2 89.5 90.9 94.4 FPM/NDA 95.0 95.2 94.7 95.2 94.7 -
[1] YAMAMOTO T, KIMURA H, HAYASHI K, et al. Pseudo-continuous arterial spin labeling MR images in Warthin tumors and pleomorphic adenomas of the parotid gland: Qualitative and quantitative analyses and their correlation with histopathologic and DWI and dynamic contrast enhanced MRI findings[J]. Neuroradiology, 2018, 60(8): 803−812. doi: 10.1007/s00234-018-2046-9 [2] ZHENG C Y, CAO R, GAO M H, et al. Comparison of surgical techniques for benign parotid tumours: A multicentre retrospective study[J]. International Journal of Oral & Maxillofacial Surgery, 2018, 48(2): 187−192. [3] 胡涛, 刘琼, 邹玉坚, 等. 扩散峰度成像及动态增强MRI鉴别腮腺多形性腺瘤与Warthin瘤[J]. 放射学实践, 2021,36(9): 1089−1094.HU T, LIU Q, ZOU Y J, et al. Application value of DKI and DEC-MRI in the differential diagnosis of parotid pleomorphic adenoma and Warthin tumor[J]. Journal of Radiology Practice, 2021, 36(9): 1089−1094. (in Chinese). [4] PARK H J, LEE S M, SONG J W, et al. Texture-based automated quantitative assessment of regional patterns on initial CT in patients with idiopathic pulmonary fibrosis: Relationship to decline in forced vital capacity[J]. American Journal of Roentgenology, 2016, 207(5): 976−983. doi: 10.2214/AJR.16.16054 [5] AHN S J, KIM J H, PARK S J, et al. Prediction of the therapeutic response after FOLFOX and FOLFIRI treatment for patients with liver metastasis from colorectal cancer using computerized CT texture analysis[J]. European Journal of Radiology, 2016, 85: 1867−1874. doi: 10.1016/j.ejrad.2016.08.014 [6] 刘文华, 张衡, 李敏, 等. CT图像纹理分析鉴别诊断腮腺混合瘤与腺淋巴瘤[J]. 临床放射学杂志, 2019,38(12): 2271−2274.LIU W H, ZHANG H, LI M, et al. CT texture analysis in the differential diagnosis of mixed tumor of parotid gland and adenolymphoma[J]. Journal of Clinical Radiology, 2019, 38(12): 2271−2274. (in Chinese). [7] 任思桐, 李小虎, 刘斌, 等. CT平扫图像纹理分析鉴别腮腺多形性腺瘤与恶性肿瘤的初步研究[J]. CT理论与应用研究, 2019,28(6): 685−691. DOI: 10.15953/j.1004-4140.2019.28.06.06.REN S T, LI X H, LIU B, et al. Preliminary study on differentiating pleomorphic adenoma and malignant tumors of the parotid gland by texture analysis of non-enhanced CT images[J]. CT Theory and Applications, 2019, 28(6): 685−691. DOI: 10.15953/j.1004-4140.2019.28.06.06. (in Chinese). [8] 茂盛, 王嗣伟, 晋丹丹, 等. 腮腺多形性腺瘤与腺淋巴瘤的CT影像特征及对比分析[J]. 实用放射学杂志, 2019,33(1): 28−46.MAO S, WANG S W, JIN D D, et al. CT imaging characteristics and comparative analysis of parotid pleomorphic adeonoma and adenolymphoma[J]. Journal of Practical Radiology, 2019, 33(1): 28−46. (in Chinese). [9] 茅枭骁, 征锦. CT纹理分析技术在甲状腺结节影像研究中的应用进展[J]. 医疗卫生装备, 2020,41(12): 97−100.MAO X X, ZHENG J. Research progress of CT texture analysis for imaging studies of thyroid nodules[J]. Medical and Health Equipment, 2020, 41(12): 97−100. (in Chinese). [10] 任继亮, 吴颖为, 陶晓峰, 等. 常规MRI纹理分析鉴别诊断眼眶淋巴瘤与炎性假瘤[J]. 中国医学影像技术, 2017,33: 980−984.REN J L, WU Y W, TAO X F, et al. MRI texture analysis in differential diagnosis of orbital lymphoma and inflammatory pseudotumor[J]. Chinese Journal of Medical Imaging Technology, 2017, 33: 980−984. (in Chinese). [11] DODGSON T, MEDICINES M D, SCHEDAR N, et al. Can quantitative CT texture analysis be used to differentiate fat-poor renal policewoman from renal cell carcinoma on enhancement CT images[J]. Radiology, 2015, 276(3): 787−796. doi: 10.1148/radiol.2015142215 [12] SUDARSHAN V K, MOOKIAH M R, ACHARYA U R, et al. Application of wavelet techniques for cancer diagnosis using ultrasound images: A review[J]. Computers in Biology & Medicine, 2016, 69: 97−111. [13] 徐圆, 段钰, 吴晶涛, 等. 基于CT增强扫描的纹理分析技术鉴别肾脏透明细胞癌恶性程度的可行性研究[J]. 临床放射学杂志, 2019,38: 1693−1697.XU Y, DUAN Y, WU J T, et al. Preoperative assessment of pathological grade of clear cell renal cell carcinoma by texture analysis based on CT enhanced images[J]. Journal of Clinical Radiology, 2019, 38: 1693−1697. (in Chinese). [14] 周明, 钱斌, 翟晓东. 腮腺腺淋巴瘤与多形性腺瘤的双期增强CT表现与鉴别[J]. 临床放射学杂志, 2012,31: 1243−1246.ZHOU M, QIAN B, ZHAI X D. CT differential diagnosis between parotid gland lymphoma and pleomorphic adenoma[J]. Journal of Clinical Radiology, 2012, 31: 1243−1246. (in Chinese). [15] 余先超, 孙宇凤, 李鹏, 等. 影像组学在腮腺多形性腺瘤与腺淋巴瘤鉴别诊断中的应用[J]. 现代肿瘤医学, 2021,29(5): 837−840. doi: 10.3969/j.issn.1672-4992.2021.05.025YU X C, SUN Y F, LI P, et al. Application of radiomics in the differential diagnosis of parotid pleomorphic adenoma and adenolymphoma of the parotid gland[J]. Modern Oncology, 2021, 29(5): 837−840. (in Chinese). doi: 10.3969/j.issn.1672-4992.2021.05.025 [16] 尹进学, 汤日杰, 钟熹, 等. 常规T2WI纹理分析预测早期宫颈鳞癌盆腔淋巴结转移的价值[J]. 临床放射学杂志, 2020,39: 358−362.YIN J X, TANG R J, ZHONG X, et al. Value of conventional T2-weighted images texture analysis in predicting pelvic lymph node metastasis in early-stage cervical cancer[J]. Journal of Clinical Radiology, 2020, 39: 358−362. (in Chinese). [17] 徐圆, 段钰, 曹正业, 等. CT纹理组学联合机器学习预测肺腺癌淋巴结转移[J]. 临床放射学杂志, 2020,39: 691−695.XU Y, DUAN Y, CAO Z Y, et al. Value of texture analysis combined with machine learning based on enhanced CT of lung adenocarcinoma in prediction of lymph node metastasis[J]. Journal of Clinical Radiology, 2020, 39: 691−695. (in Chinese). -