- Email: [email protected]
Apparent diffusion coefficient values obtained by unenhanced MRI predicts disease-specific survival in bladder cancer
Clinical Radiology 73 (2018) 881e885
Contents lists available at ScienceDirect
Clinical Radiology journal homepage: www.clinicalradiologyonline.net
Apparent diffusion coefficient values obtained by unenhanced MRI predicts disease-specific survival in bladder cancer S. Sevcenco a, A.B. Maj-Hes b, S. Hruby c, d, L. Ponhold e, G. Heinz-Peer e, M. Rauchenwald a, M. Marszalek a, H.C. Klingler f, S. Polanec g, P.A.T. Baltzer g, * a
Department of Urology and Andrology, Donauspital, Vienna, Austria Department of Urology, Medical University of Vienna, Austria c Department of Oncology, Kaiser-Franz-Joseph Hospital, Austria d Department of Urology, Landeskrankenhaus Salzburg, Paracelsus Medical University, Salzburg, Austria e €lten, Austria Department of Radiology, University Hospital of Sankt-Po f Department of Urology, Wilhelminenspital Vienna, Austria g Department of Biomedical Imaging and Image-Guided Therapy, Medical University of Vienna, Austria b
art icl e i nformat ion Article history: Received 13 September 2017 Accepted 15 May 2018
AIM: To assess the ability of apparent diffusion coefficient (ADC) measurements obtained by MRI to predict disease-specific survival (DSS) in patients with bladder cancer and compare it with established clinico-pathological prognostic factors. MATERIAL AND METHODS: The ethical review board approved this cross-sectional study. Patients with suspected bladder cancer receiving diagnostic 3 T diffusion-weighted imaging (DWI) of the bladder before transurethral resection of the bladder (TUR-B) or radical cystectomy were evaluated prospectively. Two independent radiologists measured ADC values in bladder cancer lesions in regions of interest. Associations between ADC values and pathological features with DSS were tested statistically. A combined model was established using artificial neuronal network (ANN) methodology. RESULTS: A total of 51 patients (median age 69 years, range 41e89 years) were included. Three patients were lost to follow-up, leaving 48 patients for survival analysis. Seven patients died during the 795 months studied. ADC showed significant potential to predict DSS (p<0.05). Except for grading, all pathological features as assessed by TUR-B could predict DSS (p<0.05, respectively). The combined ANN classifier showed the highest accuracy to predict DSS (0.889, 95% confidence interval: 0.732e1, p¼0.001) compared to all single parameters. ADC was the second important predictor of the ANN. CONCLUSIONS: ADC measurements obtained by unenhanced MRI predicts DSS in bladder cancer patients. A combined classifier including ADC and clinico-pathological information showed high accuracy to identify patients at high risk for disease-related death. Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
* Guarantor and correspondent: P. A. T. Baltzer, Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Allgemeines € rtel 18-20, 1090 Vienna, Austria. Tel.: þ43 1 40400 48180. Krankenhaus Wien, W€ ahringer Gu E-mail address: [email protected] (P.A.T. Baltzer). https://doi.org/10.1016/j.crad.2018.05.022 0009-9260/Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
882
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
Introduction Bladder cancer is one of the more frequent malignant tumours of the urinary system causing high morbidity and mortality.1 Treatment options for bladder cancer depend on the extent of muscle invasion. Non-muscle-invasive bladder cancers (low and high grade) are managed by transurethral simple resection and fulguration of the tumour and/or selective intravesical chemotherapy.2 Radical cystectomy and extended lymphadenectomy is the reference standard treatment in muscle-invasive bladder cancer patients.3 Prediction of the clinical course and the choice of treatment modalities are largely based on T-stage and histological grading; however, these prognostic features (T stage and grading) are limited in the prediction of treatment outcomes.4e7 Therefore, additional prognostic markers are currently under investigation to predict the clinical course of a given cancer and allow an accurate risk-stratification for improved clinical decision-making regarding neoadjuvant chemotherapy.7,8 Imaging biomarkers in this respect are of special interest as they provide spatially resolved quantifiable information connected to functional cancer properties.9 The most promising imaging biomarker in bladder imaging is the apparent diffusion coefficient (ADC) obtained by diffusion-weighted imaging (DWI), a method probing extracellular water diffusion in vivo. Prior research has shown strong associations of low ADC values with surrogate markers of unfavourable prognosis such as grading and staging.9e13 Although conceivable by these results, a direct association of ADC with clinical outcomes such as diseasespecific survival (DSS) has not been shown. Therefore, the aim of the present study was to assess the ability of ADC measurements obtained at magnetic resonance imaging (MRI) to predict DSS and compare it to established clinico-pathological prognostic factors.
Materials and methods Patients Patients with suspected bladder cancer that received diagnostic 3 T DWI of the bladder before transurethral resection of the bladder (TUR-B) or radical cystectomy were eligible for this prospective, ethics review board-approved study. All patients provided written informed consent for the study. The exclusion criteria were lack of MRI due to contraindications against MRI or in case the patient refused the additional MRI.
MRI protocol All imaging was performed using a whole-body MRI system at a field strength of 3 T (TIM Trio, Siemens Healthineers, Erlangen, Germany). Dedicated vendor-supplied phasedarray receiver coils were used for image acquisition. The imaging protocol included an echo planar imaging (EPI)based DWI sequence (7,500 ms repetition time [TR], 84 ms
echo time [TE], b-values of 50, 400, and 1000 s/mm2, parallel imaging using GRAPPA factor 2, 1736 Hz receiver bandwidth, echo spacing of 0.92 m/s, 1.81.55 mm spatial resolution, 6 minutes acquisition time). ADC maps were automatically calculated by the scanner software in a pixel-wise manner using a mono-exponential regression algorithm.
Data analysis Imaging data were analysed using a dedicated workstation (Siemens Leonardo MMWP, Munich, Germany) by two independent radiologists experienced in DWI and bladder cancer imaging. Solid parts of the investigated lesions were carefully identified on b¼100 s/mm2 DWI images and ADC values were measured by placing a small (5e15 pixels) region of interest (ROI) on the ADC map. In the case of multiple lesions, the largest lesion was chosen. Lesion size was measured using electronic callipers on T2-weighted MRI images that were reviewed in parallel.
Statistical analysis The association between ADC values, pathological features, and DSS were tested by receiver operating characteristics (ROC) analysis. A combined model of all imaging and pathological variables was established using multilayer perceptron artificial neuronal network (ANN) methodology. The study sample was split into training and test samples (70%, 30%) and the ANN was batch trained and optimised using scaled conjugates. The initial lambda was set to 107 and the initial sigma 105, the interval offset was 0.5. KaplaneMeier curves and log rank tests were used to further analyse the survival data.
Results A total of 51 patients (median age 69 years, range 41e89 years) were included. Three patients were lost to follow-up immediately after diagnosis and were thus excluded, leaving 48 patients for survival analysis. In the investigated cumulative time at risk of 795 months (median follow-up time 17, interquartile range [IQR] 8e27 months), seven patients died. Detailed patient and histological characteristics are presented in Table 1.
Prediction of DSS by ADC values Regarding DSS, ADC values measured by both readers showed significant differences between subgroups (Fig 1, DSS survivor: ADC1: 1101, IQR: 420; ADC2: 1098, IQR 381; DSS non-survivor: ADC1: 781, IQR 497; ADC2: 734, IQR: 428) with p¼0.024 (ADC1) and p¼0.015 (ADC2). ROC analysis revealed an area under the curve (AUC) of 0.765 (95% CI: 0.566e0.965, p¼0.026) for ADC1 and of 0.786 (95% CI: 0.580e0.991, p¼0.016) for ADC2 to predict DSS. This was in line with KaplaneMeier analysis results that demonstrated significant potential of ADC1 (p¼0.007) and ADC2 (p¼0.008) to predict DSS using the log rank test.
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
Alive
Died
32 10
5 2
2 40
3 4
7 35
6 1
compared to all single parameters to predict DSS. Notably, ADC was the second important predictor of the ANN (see Fig 2). Sensitivity was as high as the most sensitive single parameter (muscle invasion) and as specific as the most specific single parameter (vascular invasion), leading to a sensitivity and specificity of 85.7% (95% CI: 42.1e99.6%) and 90.5% (95% CI: 77.4e97.3%). All ROC curves are given in Fig 3. KaplaneMeier survival curves showed an improved distinction of survivors and non-survivors using the ANN classifier (p¼0.000027, log rank test, Harrel’s C predictive index: 0.7669, see Fig 4).
18 24
6 1
Discussion
6 36
5 2
4 38
5 2
13 29
6 1
Table 1 Patient characteristics stratified by DSS status.
Gender Male Female Lymph node status Positive Negative Muscle invasion Positive Negative Grade High-grade Low-grade Lymphovascular invasion Positive Negative Angiovascular invasion Positive Negative Apparent diffusion coefficienta Low High a
883
Cut-off: >0.93103 mm2/s.
Prediction of DSS by pathological features Except for grading (AUC 0.714, 95% CI: 0.527e0.902, p¼0.072), all pathological features as assessed by TUR-B showed all a significant diagnostic value to predict DSS using ROC analysis: vascular invasion (AUC 0.810, 95% CI: 0.603e1, p¼0.009), lymphovascular invasion (AUC 0.786, 95% CI: 0.579e0.993, p¼0.016), muscle invasion (AUC 0.845, 95% CI: 0.680e1, p¼0.004).
ANN predictor for DSS The ANN classifier contained one hidden layer and showed a higher AUC (0.889, 95% CI: 0.732e1, p¼0.001)
Figure 1 Boxplots of ADC measurements stratified by patient survivor status. Both ADC measurements were significantly lower in nonsurvivors as compared to survivors (p<0.05, respectively).
Local staging of bladder cancer is based on diagnostic tools such as computed tomography and MRI.4,5 Prior research demonstrated that ADC measurements are a surrogate marker for prognostic factors such as histological grade and stage in bladder cancer.12,14,15 The present study demonstrated a direct association between quantitative ADC measurements and DSS in bladder cancer. Furthermore, a ANN classifier showed the high importance of the ADC measurements when combined with established clinico-pathological factors. This finding is of importance as improved prediction of bladder cancer outcomes may be used for planning of tailored therapies. Diffusion reflects the random Brownian motion of water molecules within tissues. The ADC values obtained from DWI relate to tissue cellularity, which in the setting of neoplasia is usually the result of a more aggressive tumour phenotype. An aggressive neoplasm shows a high proliferation rate and the resulting increase in cellularity results in a relative decrease of the extracellular space that mainly contributes to the measured diffusivity in vivo. Thus, lower ADC values are measured, causing a dark appearance on quantitative parametric ADC maps. The association of ADC values with biomarkers related to cell cycle and
Figure 2 Importance of features incorporated into the ANN classifier. After angiovascular invasion, ADC values were identified as the second most important feature of the classifier.
884
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
Figure 3 ROC curves of all univariate factors and the multivariate ANN classifier.
proliferative activity such as p21, p53, and ki67 was shown in bladder cancer.16 Therefore, the pre-treatment ADC value provides additional information for risk stratification of bladder cancer patients. Treatment options for more aggressive bladder cancer phenotypes such as multifocal high-grade non-invasive bladder cancer include immediate or delayed radical cystectomy as well as intravesical Bacillus rin (BCG) therapy with close surveillance in CalmetteeGue non-invasive bladder cancer. In multifocal high-grade bladder cancer, early radical cystectomy has been shown to result in an overall survival benefit.4,17,18 In addition, improved risk stratification may result in an improved selection of patients for neoadjuvant chemotherapy. In this context, the proposed ANN that incorporates both imaging (ADC) and clinico-pathological parameters may help in patient selection for early multimodal therapy.
The association of ADC values with clinico-pathological prognostic factors such as grading and T stage in bladder cancer has been previously established. A recent metaanalysis including data from 11 studies concluded that DWI exhibited excellent diagnostic performance to distinguish muscle invasive from non-muscle invasive bladder cancer.11 The authors recommended the routine use of DWI in the examination of patients with bladder cancer; however, the authors concluded that further studies are needed to provide information on the association of ADC values with long-term survival.11 A pilot study by Funatsu et al. added to these findings by demonstrating an association between the ADC values and the occurrence of bladder cancer recurrence in 44 patients.19 The present study finally fills the research gap by demonstrating a direct association between ADC and survival in bladder cancer patients; however, there are certain
Figure 4 KaplaneMeier curves for dichotomised (cut-off >0.93103 mm2/s) univariate ADC (left side, log-rank p¼0.007) and the multivariate ANN classifier (right side, log-rank p<0.001).
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
limitations to the present study. First, this is an exploratory study that demonstrates differences between patient outcomes. Specific cut-off values that can be used in clinical practice cannot be proposed as these cannot be validated. In accordance with European Association of Urology (EAU) guidelines, ADC and other non-imaging biomarkers are still under investigation and are not recommended for routine clinical use due to a lack of standardisation and validation. Thus, further research using prospective and randomised interventional study designs should demonstrate whether the prognostic information by non-invasive ADC measurements may be used to improve therapeutic decisions resulting in a better outcome. In conclusion, a direct association was found between non-invasive ADC measurements and survival in bladder cancer patients. An exploratory ANN classifier combining the prognostic information of ADC with clinic-pathological information demonstrated high accuracy to identify those patients at high risk for disease-related death.
References 1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2017. CA Cancer J Clin 2017;67:7e30. https://doi.org/10.3322/caac.21387. 2. Chou R, Selph S, Buckley DI, et al. Intravesical therapy for the treatment of non-muscle-invasive bladder cancer: a systematic review and meta-analysis. J Urol 2017 May;197(5):1189e99. https://doi.org/10.1016/j.juro.2016. 12.090. 3. Kamat AM, Hahn NM, Efstathiou JA, et al. Bladder cancer. Lancet 2016;388:2796e810. https://doi.org/10.1016/S0140-6736(16)30512-8. rat E, Cowan NC, et al. EAU guidelines on muscle-invasive 4. Witjes JA, Compe and metastatic bladder cancer: summary of the 2013 guidelines. Eur Urol 2014;65:778e92. https://doi.org/10.1016/j.eururo.2013.11.046. 5. Kamat AM, Hegarty PK, Gee JR, et al. ICUD-EAU International Consultation on Bladder Cancer 2012: screening, diagnosis, and molecular markers. Eur Urol 2013;63:4e15. https://doi.org/10.1016/j.eururo.2012.09.057. 6. Kluth LA, Black PC, Bochner BH, et al. Prognostic and prediction tools in bladder cancer: a comprehensive review of the literature. Eur Urol 2015;68:238e53. https://doi.org/10.1016/j.eururo.2015.01.032. 7. Shariat SF, Passoni N, Bagrodia A, et al. Prospective evaluation of a preoperative biomarker panel for prediction of upstaging at radical cystectomy. BJU Int 2014;113:70e6. https://doi.org/10.1111/bju.12343.
885
8. Chedgy ECP, Douglas J, Wright JL, et al. Using the neoadjuvant chemotherapy paradigm to develop precision therapy for muscle-invasive bladder cancer. Urol Oncol 2016;34:469e76. https://doi.org/10.1016/ j.urolonc.2016.05.012. 9. Padhani AR, Liu G, Koh DM, et al. Diffusion-weighted magnetic resonance imaging as a cancer biomarker: consensus and recommendations. Neoplasia N Y N 2009;11:102e25. 10. Rosenkrantz AB, Haghighi M, Horn J, et al. Utility of quantitative MRI metrics for assessment of stage and grade of urothelial carcinoma of the bladder: preliminary results. AJR Am J Roentgenol 2013;201:1254e9. https://doi.org/10.2214/AJR.12.10348. 11. Zhai N, Wang Y-H, Zhu L-M, et al. Sensitivity and specificity of diffusionweighted magnetic resonance imaging in diagnosis of bladder cancers. Clin Investig Med Med Clin Exp 2015;38:E173e84. 12. Kobayashi S, Koga F, Kajino K, et al. Apparent diffusion coefficient value reflects invasive and proliferative potential of bladder cancer. J Magn Reson Imaging 2014;39:172e8. https://doi.org/10.1002/jmri.24148. 13. Wu L-M, Chen X-X, Xu J-R, et al. Clinical value of T2-weighted imaging combined with diffusion-weighted imaging in preoperative T staging of urinary bladder cancer: a large-scale, multiobserver prospective study on 3.0-T MRI. Acad Radiol 2013;20:939e46. https://doi.org/10.1016/ j.acra.2013.02.012. 14. Rosenkrantz AB, Mussi TC, Spieler B, et al. High-grade bladder cancer: association of the apparent diffusion coefficient with metastatic disease: preliminary results. J Magn Reson Imaging 2012;35:1478e83. https:// doi.org/10.1002/jmri.23590. 15. Sevcenco S, Ponhold L, Heinz-Peer G, et al. Prospective evaluation of diffusion-weighted MRI of the bladder as a biomarker for prediction of bladder cancer aggressiveness. Urol Oncol 2014 Nov;32(8):1166e71. https://doi.org/10.1016/j.urolonc.2014.04.019. 16. Sevcenco S, Haitel A, Ponhold L, et al. Quantitative apparent diffusion coefficient measurements obtained by 3-tesla MRI are correlated with biomarkers of bladder cancer proliferative activity. PloS One 2014 Sep;9(9):e106866. https://doi.org/10.1371/journal.pone.0106866. 17. Denzinger S, Fritsche H-M, Otto W, et al. Early versus deferred cystectomy for initial high-risk pT1G3 urothelial carcinoma of the bladder: do risk factors define feasibility of bladder-sparing approach? Eur Urol 2008;53:146e52. https://doi.org/10.1016/j.eururo.2007.06.030. 18. Hautmann RE, Volkmer BG, Gust K. Quantification of the survival benefit of early versus deferred cystectomy in high-risk non-muscle invasive bladder cancer (T1 G3). World J Urol 2009;27:347e51. https://doi.org/ 10.1007/s00345-009-0402-4. 19. Funatsu H, Imamura A, Takano H, et al. Can pretreatment ADC values predict recurrence of bladder cancer after transurethral resection? Eur J Radiol 2012;81:3115e9. https://doi.org/10.1016/j.ejrad.2012.06.009.
Contents lists available at ScienceDirect
Clinical Radiology journal homepage: www.clinicalradiologyonline.net
Apparent diffusion coefficient values obtained by unenhanced MRI predicts disease-specific survival in bladder cancer S. Sevcenco a, A.B. Maj-Hes b, S. Hruby c, d, L. Ponhold e, G. Heinz-Peer e, M. Rauchenwald a, M. Marszalek a, H.C. Klingler f, S. Polanec g, P.A.T. Baltzer g, * a
Department of Urology and Andrology, Donauspital, Vienna, Austria Department of Urology, Medical University of Vienna, Austria c Department of Oncology, Kaiser-Franz-Joseph Hospital, Austria d Department of Urology, Landeskrankenhaus Salzburg, Paracelsus Medical University, Salzburg, Austria e €lten, Austria Department of Radiology, University Hospital of Sankt-Po f Department of Urology, Wilhelminenspital Vienna, Austria g Department of Biomedical Imaging and Image-Guided Therapy, Medical University of Vienna, Austria b
art icl e i nformat ion Article history: Received 13 September 2017 Accepted 15 May 2018
AIM: To assess the ability of apparent diffusion coefficient (ADC) measurements obtained by MRI to predict disease-specific survival (DSS) in patients with bladder cancer and compare it with established clinico-pathological prognostic factors. MATERIAL AND METHODS: The ethical review board approved this cross-sectional study. Patients with suspected bladder cancer receiving diagnostic 3 T diffusion-weighted imaging (DWI) of the bladder before transurethral resection of the bladder (TUR-B) or radical cystectomy were evaluated prospectively. Two independent radiologists measured ADC values in bladder cancer lesions in regions of interest. Associations between ADC values and pathological features with DSS were tested statistically. A combined model was established using artificial neuronal network (ANN) methodology. RESULTS: A total of 51 patients (median age 69 years, range 41e89 years) were included. Three patients were lost to follow-up, leaving 48 patients for survival analysis. Seven patients died during the 795 months studied. ADC showed significant potential to predict DSS (p<0.05). Except for grading, all pathological features as assessed by TUR-B could predict DSS (p<0.05, respectively). The combined ANN classifier showed the highest accuracy to predict DSS (0.889, 95% confidence interval: 0.732e1, p¼0.001) compared to all single parameters. ADC was the second important predictor of the ANN. CONCLUSIONS: ADC measurements obtained by unenhanced MRI predicts DSS in bladder cancer patients. A combined classifier including ADC and clinico-pathological information showed high accuracy to identify patients at high risk for disease-related death. Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
* Guarantor and correspondent: P. A. T. Baltzer, Department of Biomedical Imaging and Image-guided Therapy, Medical University of Vienna, Allgemeines € rtel 18-20, 1090 Vienna, Austria. Tel.: þ43 1 40400 48180. Krankenhaus Wien, W€ ahringer Gu E-mail address: [email protected] (P.A.T. Baltzer). https://doi.org/10.1016/j.crad.2018.05.022 0009-9260/Ó 2018 The Royal College of Radiologists. Published by Elsevier Ltd. All rights reserved.
882
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
Introduction Bladder cancer is one of the more frequent malignant tumours of the urinary system causing high morbidity and mortality.1 Treatment options for bladder cancer depend on the extent of muscle invasion. Non-muscle-invasive bladder cancers (low and high grade) are managed by transurethral simple resection and fulguration of the tumour and/or selective intravesical chemotherapy.2 Radical cystectomy and extended lymphadenectomy is the reference standard treatment in muscle-invasive bladder cancer patients.3 Prediction of the clinical course and the choice of treatment modalities are largely based on T-stage and histological grading; however, these prognostic features (T stage and grading) are limited in the prediction of treatment outcomes.4e7 Therefore, additional prognostic markers are currently under investigation to predict the clinical course of a given cancer and allow an accurate risk-stratification for improved clinical decision-making regarding neoadjuvant chemotherapy.7,8 Imaging biomarkers in this respect are of special interest as they provide spatially resolved quantifiable information connected to functional cancer properties.9 The most promising imaging biomarker in bladder imaging is the apparent diffusion coefficient (ADC) obtained by diffusion-weighted imaging (DWI), a method probing extracellular water diffusion in vivo. Prior research has shown strong associations of low ADC values with surrogate markers of unfavourable prognosis such as grading and staging.9e13 Although conceivable by these results, a direct association of ADC with clinical outcomes such as diseasespecific survival (DSS) has not been shown. Therefore, the aim of the present study was to assess the ability of ADC measurements obtained at magnetic resonance imaging (MRI) to predict DSS and compare it to established clinico-pathological prognostic factors.
Materials and methods Patients Patients with suspected bladder cancer that received diagnostic 3 T DWI of the bladder before transurethral resection of the bladder (TUR-B) or radical cystectomy were eligible for this prospective, ethics review board-approved study. All patients provided written informed consent for the study. The exclusion criteria were lack of MRI due to contraindications against MRI or in case the patient refused the additional MRI.
MRI protocol All imaging was performed using a whole-body MRI system at a field strength of 3 T (TIM Trio, Siemens Healthineers, Erlangen, Germany). Dedicated vendor-supplied phasedarray receiver coils were used for image acquisition. The imaging protocol included an echo planar imaging (EPI)based DWI sequence (7,500 ms repetition time [TR], 84 ms
echo time [TE], b-values of 50, 400, and 1000 s/mm2, parallel imaging using GRAPPA factor 2, 1736 Hz receiver bandwidth, echo spacing of 0.92 m/s, 1.81.55 mm spatial resolution, 6 minutes acquisition time). ADC maps were automatically calculated by the scanner software in a pixel-wise manner using a mono-exponential regression algorithm.
Data analysis Imaging data were analysed using a dedicated workstation (Siemens Leonardo MMWP, Munich, Germany) by two independent radiologists experienced in DWI and bladder cancer imaging. Solid parts of the investigated lesions were carefully identified on b¼100 s/mm2 DWI images and ADC values were measured by placing a small (5e15 pixels) region of interest (ROI) on the ADC map. In the case of multiple lesions, the largest lesion was chosen. Lesion size was measured using electronic callipers on T2-weighted MRI images that were reviewed in parallel.
Statistical analysis The association between ADC values, pathological features, and DSS were tested by receiver operating characteristics (ROC) analysis. A combined model of all imaging and pathological variables was established using multilayer perceptron artificial neuronal network (ANN) methodology. The study sample was split into training and test samples (70%, 30%) and the ANN was batch trained and optimised using scaled conjugates. The initial lambda was set to 107 and the initial sigma 105, the interval offset was 0.5. KaplaneMeier curves and log rank tests were used to further analyse the survival data.
Results A total of 51 patients (median age 69 years, range 41e89 years) were included. Three patients were lost to follow-up immediately after diagnosis and were thus excluded, leaving 48 patients for survival analysis. In the investigated cumulative time at risk of 795 months (median follow-up time 17, interquartile range [IQR] 8e27 months), seven patients died. Detailed patient and histological characteristics are presented in Table 1.
Prediction of DSS by ADC values Regarding DSS, ADC values measured by both readers showed significant differences between subgroups (Fig 1, DSS survivor: ADC1: 1101, IQR: 420; ADC2: 1098, IQR 381; DSS non-survivor: ADC1: 781, IQR 497; ADC2: 734, IQR: 428) with p¼0.024 (ADC1) and p¼0.015 (ADC2). ROC analysis revealed an area under the curve (AUC) of 0.765 (95% CI: 0.566e0.965, p¼0.026) for ADC1 and of 0.786 (95% CI: 0.580e0.991, p¼0.016) for ADC2 to predict DSS. This was in line with KaplaneMeier analysis results that demonstrated significant potential of ADC1 (p¼0.007) and ADC2 (p¼0.008) to predict DSS using the log rank test.
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
Alive
Died
32 10
5 2
2 40
3 4
7 35
6 1
compared to all single parameters to predict DSS. Notably, ADC was the second important predictor of the ANN (see Fig 2). Sensitivity was as high as the most sensitive single parameter (muscle invasion) and as specific as the most specific single parameter (vascular invasion), leading to a sensitivity and specificity of 85.7% (95% CI: 42.1e99.6%) and 90.5% (95% CI: 77.4e97.3%). All ROC curves are given in Fig 3. KaplaneMeier survival curves showed an improved distinction of survivors and non-survivors using the ANN classifier (p¼0.000027, log rank test, Harrel’s C predictive index: 0.7669, see Fig 4).
18 24
6 1
Discussion
6 36
5 2
4 38
5 2
13 29
6 1
Table 1 Patient characteristics stratified by DSS status.
Gender Male Female Lymph node status Positive Negative Muscle invasion Positive Negative Grade High-grade Low-grade Lymphovascular invasion Positive Negative Angiovascular invasion Positive Negative Apparent diffusion coefficienta Low High a
883
Cut-off: >0.93103 mm2/s.
Prediction of DSS by pathological features Except for grading (AUC 0.714, 95% CI: 0.527e0.902, p¼0.072), all pathological features as assessed by TUR-B showed all a significant diagnostic value to predict DSS using ROC analysis: vascular invasion (AUC 0.810, 95% CI: 0.603e1, p¼0.009), lymphovascular invasion (AUC 0.786, 95% CI: 0.579e0.993, p¼0.016), muscle invasion (AUC 0.845, 95% CI: 0.680e1, p¼0.004).
ANN predictor for DSS The ANN classifier contained one hidden layer and showed a higher AUC (0.889, 95% CI: 0.732e1, p¼0.001)
Figure 1 Boxplots of ADC measurements stratified by patient survivor status. Both ADC measurements were significantly lower in nonsurvivors as compared to survivors (p<0.05, respectively).
Local staging of bladder cancer is based on diagnostic tools such as computed tomography and MRI.4,5 Prior research demonstrated that ADC measurements are a surrogate marker for prognostic factors such as histological grade and stage in bladder cancer.12,14,15 The present study demonstrated a direct association between quantitative ADC measurements and DSS in bladder cancer. Furthermore, a ANN classifier showed the high importance of the ADC measurements when combined with established clinico-pathological factors. This finding is of importance as improved prediction of bladder cancer outcomes may be used for planning of tailored therapies. Diffusion reflects the random Brownian motion of water molecules within tissues. The ADC values obtained from DWI relate to tissue cellularity, which in the setting of neoplasia is usually the result of a more aggressive tumour phenotype. An aggressive neoplasm shows a high proliferation rate and the resulting increase in cellularity results in a relative decrease of the extracellular space that mainly contributes to the measured diffusivity in vivo. Thus, lower ADC values are measured, causing a dark appearance on quantitative parametric ADC maps. The association of ADC values with biomarkers related to cell cycle and
Figure 2 Importance of features incorporated into the ANN classifier. After angiovascular invasion, ADC values were identified as the second most important feature of the classifier.
884
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
Figure 3 ROC curves of all univariate factors and the multivariate ANN classifier.
proliferative activity such as p21, p53, and ki67 was shown in bladder cancer.16 Therefore, the pre-treatment ADC value provides additional information for risk stratification of bladder cancer patients. Treatment options for more aggressive bladder cancer phenotypes such as multifocal high-grade non-invasive bladder cancer include immediate or delayed radical cystectomy as well as intravesical Bacillus rin (BCG) therapy with close surveillance in CalmetteeGue non-invasive bladder cancer. In multifocal high-grade bladder cancer, early radical cystectomy has been shown to result in an overall survival benefit.4,17,18 In addition, improved risk stratification may result in an improved selection of patients for neoadjuvant chemotherapy. In this context, the proposed ANN that incorporates both imaging (ADC) and clinico-pathological parameters may help in patient selection for early multimodal therapy.
The association of ADC values with clinico-pathological prognostic factors such as grading and T stage in bladder cancer has been previously established. A recent metaanalysis including data from 11 studies concluded that DWI exhibited excellent diagnostic performance to distinguish muscle invasive from non-muscle invasive bladder cancer.11 The authors recommended the routine use of DWI in the examination of patients with bladder cancer; however, the authors concluded that further studies are needed to provide information on the association of ADC values with long-term survival.11 A pilot study by Funatsu et al. added to these findings by demonstrating an association between the ADC values and the occurrence of bladder cancer recurrence in 44 patients.19 The present study finally fills the research gap by demonstrating a direct association between ADC and survival in bladder cancer patients; however, there are certain
Figure 4 KaplaneMeier curves for dichotomised (cut-off >0.93103 mm2/s) univariate ADC (left side, log-rank p¼0.007) and the multivariate ANN classifier (right side, log-rank p<0.001).
S. Sevcenco et al. / Clinical Radiology 73 (2018) 881e885
limitations to the present study. First, this is an exploratory study that demonstrates differences between patient outcomes. Specific cut-off values that can be used in clinical practice cannot be proposed as these cannot be validated. In accordance with European Association of Urology (EAU) guidelines, ADC and other non-imaging biomarkers are still under investigation and are not recommended for routine clinical use due to a lack of standardisation and validation. Thus, further research using prospective and randomised interventional study designs should demonstrate whether the prognostic information by non-invasive ADC measurements may be used to improve therapeutic decisions resulting in a better outcome. In conclusion, a direct association was found between non-invasive ADC measurements and survival in bladder cancer patients. An exploratory ANN classifier combining the prognostic information of ADC with clinic-pathological information demonstrated high accuracy to identify those patients at high risk for disease-related death.
References 1. Siegel RL, Miller KD, Jemal A. Cancer statistics, 2017. CA Cancer J Clin 2017;67:7e30. https://doi.org/10.3322/caac.21387. 2. Chou R, Selph S, Buckley DI, et al. Intravesical therapy for the treatment of non-muscle-invasive bladder cancer: a systematic review and meta-analysis. J Urol 2017 May;197(5):1189e99. https://doi.org/10.1016/j.juro.2016. 12.090. 3. Kamat AM, Hahn NM, Efstathiou JA, et al. Bladder cancer. Lancet 2016;388:2796e810. https://doi.org/10.1016/S0140-6736(16)30512-8. rat E, Cowan NC, et al. EAU guidelines on muscle-invasive 4. Witjes JA, Compe and metastatic bladder cancer: summary of the 2013 guidelines. Eur Urol 2014;65:778e92. https://doi.org/10.1016/j.eururo.2013.11.046. 5. Kamat AM, Hegarty PK, Gee JR, et al. ICUD-EAU International Consultation on Bladder Cancer 2012: screening, diagnosis, and molecular markers. Eur Urol 2013;63:4e15. https://doi.org/10.1016/j.eururo.2012.09.057. 6. Kluth LA, Black PC, Bochner BH, et al. Prognostic and prediction tools in bladder cancer: a comprehensive review of the literature. Eur Urol 2015;68:238e53. https://doi.org/10.1016/j.eururo.2015.01.032. 7. Shariat SF, Passoni N, Bagrodia A, et al. Prospective evaluation of a preoperative biomarker panel for prediction of upstaging at radical cystectomy. BJU Int 2014;113:70e6. https://doi.org/10.1111/bju.12343.
885
8. Chedgy ECP, Douglas J, Wright JL, et al. Using the neoadjuvant chemotherapy paradigm to develop precision therapy for muscle-invasive bladder cancer. Urol Oncol 2016;34:469e76. https://doi.org/10.1016/ j.urolonc.2016.05.012. 9. Padhani AR, Liu G, Koh DM, et al. Diffusion-weighted magnetic resonance imaging as a cancer biomarker: consensus and recommendations. Neoplasia N Y N 2009;11:102e25. 10. Rosenkrantz AB, Haghighi M, Horn J, et al. Utility of quantitative MRI metrics for assessment of stage and grade of urothelial carcinoma of the bladder: preliminary results. AJR Am J Roentgenol 2013;201:1254e9. https://doi.org/10.2214/AJR.12.10348. 11. Zhai N, Wang Y-H, Zhu L-M, et al. Sensitivity and specificity of diffusionweighted magnetic resonance imaging in diagnosis of bladder cancers. Clin Investig Med Med Clin Exp 2015;38:E173e84. 12. Kobayashi S, Koga F, Kajino K, et al. Apparent diffusion coefficient value reflects invasive and proliferative potential of bladder cancer. J Magn Reson Imaging 2014;39:172e8. https://doi.org/10.1002/jmri.24148. 13. Wu L-M, Chen X-X, Xu J-R, et al. Clinical value of T2-weighted imaging combined with diffusion-weighted imaging in preoperative T staging of urinary bladder cancer: a large-scale, multiobserver prospective study on 3.0-T MRI. Acad Radiol 2013;20:939e46. https://doi.org/10.1016/ j.acra.2013.02.012. 14. Rosenkrantz AB, Mussi TC, Spieler B, et al. High-grade bladder cancer: association of the apparent diffusion coefficient with metastatic disease: preliminary results. J Magn Reson Imaging 2012;35:1478e83. https:// doi.org/10.1002/jmri.23590. 15. Sevcenco S, Ponhold L, Heinz-Peer G, et al. Prospective evaluation of diffusion-weighted MRI of the bladder as a biomarker for prediction of bladder cancer aggressiveness. Urol Oncol 2014 Nov;32(8):1166e71. https://doi.org/10.1016/j.urolonc.2014.04.019. 16. Sevcenco S, Haitel A, Ponhold L, et al. Quantitative apparent diffusion coefficient measurements obtained by 3-tesla MRI are correlated with biomarkers of bladder cancer proliferative activity. PloS One 2014 Sep;9(9):e106866. https://doi.org/10.1371/journal.pone.0106866. 17. Denzinger S, Fritsche H-M, Otto W, et al. Early versus deferred cystectomy for initial high-risk pT1G3 urothelial carcinoma of the bladder: do risk factors define feasibility of bladder-sparing approach? Eur Urol 2008;53:146e52. https://doi.org/10.1016/j.eururo.2007.06.030. 18. Hautmann RE, Volkmer BG, Gust K. Quantification of the survival benefit of early versus deferred cystectomy in high-risk non-muscle invasive bladder cancer (T1 G3). World J Urol 2009;27:347e51. https://doi.org/ 10.1007/s00345-009-0402-4. 19. Funatsu H, Imamura A, Takano H, et al. Can pretreatment ADC values predict recurrence of bladder cancer after transurethral resection? Eur J Radiol 2012;81:3115e9. https://doi.org/10.1016/j.ejrad.2012.06.009.