目的 探讨双参数MRI瘤周影像组学联合机器学习对前列腺癌术后切缘阳性的预测价值。方法 回顾性分析解放军总医院第一医学中心(一中心) 、第三医学中心(三中心)行前列腺癌根治性切除术的274例患者的临床资料。将一中心的患者按照随机种子数以7∶3随机分为训练集和内部测试集,三中心的数据资料作为外部验证集。利用影像组学软件分别提取T2WI和ADC图肿瘤周围区域的107个影像组学特征,使用最大相关性和最小冗余(mRMR)算法去除相关性较高的特征,利用最小绝对收缩和选择算子(LASSO)算法进行特征筛选,构建组学模型。比较组学模型在Extra Trees、XGBoost和RandomForest三种机器学习算法的性能,采用DeLong检验比较三种机器学习算法AUC的差异。结果 在组学模型中,Extra Trees的训练集AUC=0.771,内部测试集AUC=0.743,外部验证集AUC=0.726,优于XGBoost及RandomForest。在联合模型中,Extra Trees的训练集AUC=0.778,内部测试集AUC=0.753,外部验证集AUC=0.777,模型的性能优于XGBoost及RandomForest。DeLong检验显示:在组学模型中,三种机器学习算法的AUC差异无统计学意义;在联合模型的内部测试集中,三种机器学习算法的AUC差异无统计学意义;在外部验证集中,RandomForest与Extra Trees相比,两者AUC差异无统计学意义(P=0.160),而RandomForest和XGBoost相比(P=0.006),Extra Trees与XGBoost相比(P=0.001),差异均具有统计学意义。结论 Extra Trees在内部测试集和外部验证集均能达到满意的效果,模型整体表现最好。利用Extra Trees构建的联合模型对前列腺癌术后切缘阳性具有良好的预测效能。
Abstract
Objective To explore the predictive value of dual-parameter MRI peritumoral radiomics combined with machine learning for positive surgical margins after prostate cancer radical resection. Methods A retrospective analysis of clinical data of 274 patients who underwent radical prostatectomy at the First Medical Center (the First Center) and the Third Medical Center (the Third Center) of PLA General Hospital. The patients from the First Center were randomly divided into training set and internal set at a ratio of 7∶3, and the data from the Third Center served as the external validation set. Radiomics features of the peritumoral regions of T2WI and ADC images were extracted using radiomics software, and the maximal relevance and minimal redundancy (mRMR) algorithm was applied to remove highly correlated features, followed by feature selection using the least absolute shrinkage and selection operator (LASSO) algorithm to construct the radiomics model. The performance of the omics models in three machine learning algorithms (Extra Trees, XGBoost, and Random Forest) was compared, and the differences in AUC of the three machine learning algorithms were compared using DeLong test. Results In the omics model, Extra Trees achieved a training set AUC of 0.771, an internal test set AUC of 0.743, and an external validation set AUC of 0.726, outperforming both XGBoost and RandomForest. In the ensemble model, Extra Trees achieved a training set AUC of 0.778, an internal test set AUC of 0.753, and an external validation set AUC of 0.777. The performance of this model surpassed that of XGBoost and Random Forest. The DeLong test revealed no significant difference in AUC among the three machine learning algorithms in the omics model. No significant difference in AUC was observed among the three machine learning algorithms in the internal test set of the ensemble model. In the external validation set, the AUC difference between RandomForest and Extra Trees was not statistically significant (P=0.160). However, the differences between RandomForest and XGBoost (P=0.006) and between Extra Trees and XGBoost (P=0.001) were both statistically significant. Conclusions Extra Trees can achieve satisfactory results in both the internal test set and the external validation set, and the overall performance of the model is the best. The combined model constructed using Extra Trees is characterized by good predictive efficacy for positive surgical margins after prostate cancer radical resection.
关键词
双参数MRI /
肿瘤周围 /
机器学习 /
影像组学 /
前列腺癌 /
切缘阳性
Key words
dual-parameter MRI /
peritumoral /
machine learning /
radiomics /
prostate cancer /
positive surgical margin
{{custom_sec.title}}
{{custom_sec.title}}
{{custom_sec.content}}
参考文献
[1] 许梨梨, 孙 昊, 金征宇. 前列腺癌包膜外侵犯的MRI相关影像及临床预测方法研究进展[J]. 放射学实践,2022,37(1):110-114.
[2] Diamand R, Roche J B, Lievore E,et al. External validation of models for prediction of side-specific extracapsular extension in prostate cancer patients undergoing radical prostatectomy[J]. Eur Urol Focus, 2023, 9(2): 309-316.
[3] Loeb S, Catalona W J. What is the prognostic impact of positive surgical margins in surgically treated prostate cancer?[J]. Nat Clin Pract Urol, 2006, 3(8): 418-419.
[4] He D, Wang X, Fu C, et al. Mri-based radiomics models to assess prostate cancer, extracapsular extension and positive surgical margins[J]. Cancer Imaging, 2021, 21(1): 46.
[5] Ma S, Xie H, Wang H, et al. Preoperative prediction of extracapsular extension: radiomics signature based on magnetic resonance imaging to stage prostate cancer[J]. Mol Imaging Biol, 2020, 22(3): 711-721.
[6] Braman N, Prasanna P, Whitney J, et al. Association of peritumoral radiomics with tumor biology and pathologic response to preoperative targeted therapy for her2(erbb2)-positive breast cancer[J]. JAMA Netw Open, 2019,2(4): e192561.
[7] Bulten W, Bándi P, Hoven J, et al. Epithelium segmentation using deep learning in h&e-stained prostate specimens with immunohistochemistry as reference standard[J]. Sci Rep, 2019, 9(1): 864.
[8] 黄志强,钟士江. 机器学习在抑郁症辅助诊断中的应用研究进展[J]. 武警医学, 2024, 35(9): 806-812.
[9] Chan H P,Samala R K,Hadjiiski L M, et al. Deep learning in medical image analysis[J]. Adv Exp Med Biol, 2020, 12(13): 3-21.
[10] Dvan der Velden B H M, Kuijf H J, Gilhuijs K G A, et al. Explainable artificial intelligence(xai) in deep learning-based medical image analysis[J]. Med Image Anal, 2022,7(9): 102470.
[11] 牛永超,周 芳,赵丹丹, 等.基于DWI深度学习特征的预测模型评估子宫内膜癌微卫星不稳定状态的价值[J]. 中国医学影像学杂志,2024,32(9):922-927.
[12] Bai H, Xia W, Ji X, et al. Multiparametric magnetic resonance imaging-based peritumoral radiomics for preoperative prediction of the presence of extracapsular extension with prostate cancer[J]. J Magn Reson Imaging, 2021, 54(4): 1222-1230.
[13] Wang H, Wang K, Zhang Y, et al. Deep learning-based radiomics model from pretreatment adc to predict biochemical recurrence in advanced prostate cancer[J]. Front Oncol, 2024, 14: 1342104.
[14] Wang W, Kang Y, Liu G, et al. Scu-net: semantic segmentation network for Learning channel information on remote sensing images[J]. Comput Intell Neurosci, 2022: 8469415.
[15] Guo X, O’Neill W C, Vey B, et al. Scu-net: a deep learning method for segmentation and quantification of breast arterial calcifications on mammograms[J]. Med Phys, 2021, 48(10): 5851-5861.
[16] Chen T, Zhang Z, Tan S, et al. Mri based radiomics compared with the pi-rads v2.1 in the prediction of clinically significant prostate cancer: biparametric vs multiparametric mri[J]. Front Oncol, 2022, 11: 792456.
[17] Huang F, Huang Q, Liao X, et al. Prediction of high-risk prostate cancer based on the habitat features of biparametric magnetic resonance and the omics features of contrast-enhanced ultrasound[J]. Heliyon, 2024, 10(18): e37955-37956.
[18] Chen S, Jian T, Chi C, et al. Machine learning-based models enhance the prediction of prostate cancer[J]. Front Oncol, 2022, 12: 941349.
[19] 郭 盛,周 川,王 超, 等. Bp-mri影像组学对前列腺癌根治性切除术后切缘阳性的预测价值[J]. 磁共振成像,2023,14(12):54-59.
[20] Qiu Y, Liu Y F, Shu X, et al. Peritumoral radiomics strategy based on ensemble learning for the prediction of gleason grade group of prostate cancer[J]. Acad Radiol, 2023, 30(Suppl 1): S1-S13.
[21] i Trani M G, Nezzo M, Caporale A S, et al. Performance of diffusion kurtosis imaging versus diffusion tensor imaging in discriminating between benign tissue, low and high gleason grade prostate cancer[J]. Acad Radiol, 2019, 26(10): 1328-1337.
[22] Zhang H, Li X, Zhang Y, et al. Diagnostic nomogram based on intralesional and perilesional radiomics features and clinical factors of clinically significant prostate cancer[J]. J Magn Reson Imaging, 2021, 53(5): 1550-1558.
[23] 胡晓渝,王树杰,杨 康,等. 瘤内及瘤周表观弥散系数影像组学特征预测髓母细胞瘤患儿预后[J]. 中国医学影像技术,2024,40(2):198-202.
[24] 杨 静,黄豆豆,陈峻帆,等. 基于不同感兴趣区的多参数MRI影像组学在前列腺癌侵袭性评估中的应用研究[J]. 临床放射学杂志,2023,42(9):1465-1470.
[25] Gong L, Xu M, Fang M, et al. The potential of prostate gland radiomic features in identifying the gleason score[J]. Comput Biol Med, 2022, 144: 105318.
[26] 侯 娟,刘文亚. 肿瘤瘤周的影像组学研究进展[J]. 磁共振成像,2024,15(3):230-234.
PDF(2463 KB)
