|
|
|
| Application of magnetic resonance diffusion-weighted image based nomogram in the differential diagnosis of prostate cancer and prostatic hyperplasia and its feasibility in the diagnosis of prostate cancer with PI-RADS 4 |
| CEHN Li-hua1, LIU Ai-lian1, GUO Yan2, LI Xin2, GUO Dan1, SONG Qing-wei1, WEI Qiang1 |
| 1. The First Affiliated Hospital of Dalian Medical University, Dalian Liaoning 116011, China;
2. GE Healthcare, Beijing 100176, China |
|
|
|
|
Abstract Objective: To evaluate radiomics as a tool to distinguish prostate cancer(PCa) frombenign prostatic hyperplasia(BPH) based on diffusion-weighted imaging(DWI) sequence without subjective factors, and diagnosedclinical suspicious lesions (PI-RADS score 4 points) in the feasibility of PCa. Methods: This retrospective study was approved by local IRB, and written informed consent was waived. Two hundred and sixty patients with PCa or BPH who underwent MRI exams between January 2010 and October 2017 were enrolled in this study. Among them, 130 were PCa(between July 2010 and October 2017) and 130(between July 2010 and July 2016) were BPH, confirmed by pathologically. All MRI scans were performed on a 3.0T scanner with an eight-channel phased-array body-coil. T1WI, T2WI, DCE-MRI and DWI(b=0，1 000 s/mm2) were scanned. All the images were read by two radiologists with 5 years’ experience according to the PI-RADS V2. High-throughput extraction and analysis of the radiomic features based on DWI mainly included five key procedures: ①data pre-processing and segmentation were performed by two radiologists who were blinded to pathology, 2D region of interest(ROI) was sketched along the edge of the whole prostate gland at the slice with the maximum diameter of the lesion. ②Three hundred and ninety-seven radiomics features, including size and shape based-features, histogram and GLCM(Gray-Level Co-occurrence Matrix) as well as GLRLM(Gray-Level Run Length Matrix) texture features were generated automatically using Analysis-Kinetics software(GE Healthcare, China). ③Feature selection: according to the ratio of 7∶3, the samples were randomly divided into training and validation set, the training samples, the maximum correlation minimum redundancy algorithm(MRMR) and LASSO select and retain the best robustness characteristics used in modeling. ④Model construction: based on the choice of radiomic features, we established a Logistic regression model, and got the radiomic model. Based on clinical factors, including age, DWI routine diagnostic signal characteristics and TPSA index level, we built clinical model. Combined with radiomics features and clinical data, we got the joint model and nomogram. ⑤Model validation: The rest 30% data were used to validate the models. ROC curves were used to evaluate the diagnostic efficacy of the three models. The clinical suspicious lesions, evaluated as PI-RADS 4 by two radiologists, were underwent hierarchical analysis. Then we used calibration curve and decision curve to evaluate the calibration efficiency and clinical application value of nomogram. Results: The AUC of training group and the validation group were 0.95 and 0.92 respectively in this study to identify the PCa and BPH. The analysis results of calibration curve and decision curve also showed that nomogram had good clinical application value. The AUC of nomogram, radiomics model and clinical model were 0.73, 0.81 and 0.54 respectively in suspicious lesions evaluated as PI-RADS 4. Conclusion: The nomogram based on DWI can identify PCa and BPH well. For the lesions evaluated as PI-RADS 4, the clinical model’s diagnostic efficiency is lower than the radiomics model. The nomogram’s diagnosis efficiency is also lower than the radiomics model. We will make a deep exploration by increasing the sample in the future.
|
|
Received: 09 October 2019
|
|
|
|
|
|
| [1]Jemal A, Center MM, Desantis C, et al. Global patterns of cancer incidence and mortality rates and trends[J]. Cancer Epidemiol Biomarkers Prev, 2010, 19(8): 1893-1907.
[2]Felker ER, Raman SS, Shakeri S, et al. Utility of Restriction Spectrum Imaging Among Men Undergoing First-Time Biopsy for Suspected Prostate Cancer[J]. AJR, 2019, 213(2): 365-370.
[3]McPartlin A, Kershaw L, McWilliam A, et al. Changes in prostate apparent diffusion coefficient values during radiotherapy after neoadjuvant hormones[J]. Ther Adv Urol, 2018, 10(12): 359-364.
[4]Sprinkart AM, Marx C, Traber F, et al. Evaluation of Exponential ADC(eADC) and Computed DWI(cDWI) for the Detection of Prostate Cancer[J]. Rofo, 2018, 190(8): 758-766.
[5]Barentsz JO, Weinreb JC, Verma S, et al. Synopsis of the PI-RADS v2 Guidelines for Multiparametric Prostate Magnetic Resonance Imaging and Recommendations for Use[J]. Eur Urol, 2016, 69(1): 41-49．
[6]Ghabili K, Swallow M, Sherrer RL, et al. Association Between Tumor Multifocality on Multi-parametric MRI and Detection of Clinically-Significant Prostate Cancer in Lesions with Prostate Imaging Reporting and Data System(PI-RADS) Score 4[J]. Urology, 2019, 134: 173-180.
[7]Ullrich T, Arsov C, Quentin M, et al. Analysis of PI-RADS 4 cases: Management recommendations for negatively biopsied patients[J]. Eur J Radiol, 2019, 113: 1-6.
[8]Lambin P, Rios-velazquez E, Leijenaar R, et al. Radiomics: extracting more information from medical images using advanced feature analysis[J]. Eur J Cancer, 2012, 48(4): 441-446.
[9]Lee SL, Ravi A, Morton G, et al. Changes in ADC and T2-weighted MRI-derived radiomic features in patients treated with focal salvage HDR prostate brachytherapy for local recurrence after previous external-beam radiotherapy[J]. Brachytherapy, 2019, 18(5): 567-573.
[10]Wang J, Wu CJ, Bao ML, et al. Machine learning-based analysis of MR radiomics can help to improve the diagnostic performance of PI-RADS v2 in clinically relevant prostate cancer[J]. Eur Radiol, 2017, 27(10): 4082-4090.
[11]Cuocolo R, Stanzione A, Ponsiglione A, et al. Clinically significant prostate cancer detection on MRI: A radiomic shape features study[J]. Eur J Radiol, 2019, 116: 144-149.
[12]韩苏军，张思维，陈万青，等. 中国前列腺癌发病现状和流行趋势分析[J]. 临床肿瘤学杂志，2013，18（4）：330-334.
[13]Katahira K, Takahara T, Kwee TC, et al. Ultra-high-b-value diffusionweighted MR imaging for the detection of prostate cancer: evaluation in 201 cases with histopathological correlation[J]. Eur Radiol, 2011, 21(1): 188-196.
[14]Tamada T, Prabhu V, Li J, et al. Prostate Cancer: Diffusion-weighted MR Imaging for Detection and Assessment of Aggressiveness-Comparison between Conventional and Kurtosis Models[J]. Radiology, 2017, 284(1): 100-108.
[15]Henderson DR, De Souza NM, Thomas K, et al. Nine-year followupfor a study of diffusion-weighted magnetic resonance imaging in aprospective prostate cancer active surveillance cohort[J]. Eur Urol, 2016, 69(6): 1028-1033.
[16]Merisaari H, Toivonen J, Pesola M, et al. Diffusion-weighted imagingof prostate cancer: effect of b-value distribution on repeatabilityand cancer characterization[J]. Magn Reson Imaging, 2015, 33(10): 1212-1218.
[17]Litjens GJ, Hambrock T, Hulsbergen-Van De Kaa C, et al. Interpatient variation in normal peripheral zone apparent diffusion coefficient: effect on the prediction of prostate cancer aggressiveness[J]. Radiology, 2012, 265(1): 260-266.
[18]Rosenkrantz AB, Kim S, Lim RP, et al． Prostate cancer localization using multiparametric MR imaging: comparison of Prostate Imaging Reporting and Data System(PI-RADS) and Likert scales[J]. Radiology, 2013, 269(2): 482-492．
[19]Lorenz JW, Diane S, Lisa R, et al. Serial T2-Weighted Magnetic Resonance Images Acquired on a 1.5 Tesla Magnetic Resonance Linear Accelerator Reveal Radiomic Feature Variation in Organs at Risk: An Exploratory Analysis of Novel Metrics of Tissue Response in Prostate Cancer[J]. Cureus, 2019, 11(4): e4510.
[20]David B, Simon K, Manuel W, et al. Radiomic Machine Learning for Characterization of Prostate Lesions with MRI: Comparison to ADC Values[J]. Radiology, 2018, 289(1): 128-137.
[21]Smith CP, Czarniecki M, Mehralivand S, et al. Radiomics and radiogenomics of prostate cancer[J]. Abdom Radiol(NY), 2019, 44(6): 2021-2029.
[22]Ginsburg SB, AlgoharyA, Pahwa S, et al. Radiomic features for prostate cancer detection on MRI differ between the transition and peripheral zones: Preliminary findings from a multi-institutional study[J]. J Magn Reson Imaging, 2017, 46(1): 184-193.
[23]Lin YC, Lin G, Hong JH, et al. Diffusion radiomics analysis of intratumoral heterogeneity in a murine prostate cancer model following radiotherapy: Pixelwise correlation with histology[J]. J Magn Reson Imaging, 2017, 46(2): 483-489.
[24]Chen T, Li M, Gu Y, et al. Prostate Cancer Differentiation and Aggressiveness: Assessment With a Radiomic-Based Model vs. PI-RADS v2[J]. J Magn Reson Imaging, 2019, 49(3): 875-884.
[25]Wang J, Wu CJ, Bao ML, et al. Machine learning-based analysis of MR radiomics can help to improve the diagnostic performance of PI-RADS v2 in clinically relevant prostate cancer[J]. Eur Radiol, 2017, 27(10): 4082-4090. |
|
|
|
