The prognostic value of CT radiomic features from primary tumours and pathological lymph
nodes in head and neck cancer patients
Zhai, Tiantian
DOI:10.33612/diss.111448998
IMPORTANT NOTE: You are advised to consult the publisher's version (publisher's PDF) if you wish to cite from it. Please check the document version below.
Document Version
Publisher's PDF, also known as Version of record
Publication date: 2020
Link to publication in University of Groningen/UMCG research database
Citation for published version (APA):
Zhai, T. (2020). The prognostic value of CT radiomic features from primary tumours and pathological lymph nodes in head and neck cancer patients. University of Groningen. https://doi.org/10.33612/diss.111448998
Copyright
Other than for strictly personal use, it is not permitted to download or to forward/distribute the text or part of it without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license (like Creative Commons).
Take-down policy
If you believe that this document breaches copyright please contact us providing details, and we will remove access to the work immediately and investigate your claim.
Downloaded from the University of Groningen/UMCG research database (Pure): http://www.rug.nl/research/portal. For technical reasons the number of authors shown on this cover page is limited to 10 maximum.
features from primary tumours and
pathological lymph nodes in head and
neck cancer patients
Thesis, University of Groningen, Groningen, The Netherlands
Printing of this thesis was financially supported by the University of Groningen and the Graduate School of Medical Sciences, UMCG and Elekta B.V..
Cover Siqi Qiu, Tiantian Zhai & Jiani Li
Layout Siqi Qiu & Tiantian Zhai
Printed by Ridderprint I www.ridderprint.nl
ISBN: 978-94-6375-748-5
ISBN (electronic version): 978-94-6375-749-2 © 2020, T Zhai
All rights reserved. No part of this thesis may be reproduced, stored in a retrieval system, or transmitted in any form or by any means, mechanically, by photocopying, recording, or otherwise, without prior written permission of the author.
The prognostic value of CT
radiomic features from primary
tumours and pathological lymph
nodes in head and neck cancer
patients
PhD thesis
to obtain the degree of PhD at the University of Groningen
on the authority of the Rector Magnificus Prof. C. Wijmenga
and in accordance with the decision by the College of Deans. This thesis will be defended in public on Wednesday 15 January 2020 at 12.45 hours
by
Tiantian Zhai
born on 17 August 1987 in Shandong, China
Co-supervisors
Dr. R.J.H.M. Steenbakkers Dr. N.M. Sijtsema
Assessment Committee
Prof. J. Pruim
Prof. B.F.A.M. van der Laan Prof. A.L.A.J. Dekker
Chapter 1 General introduction 9
Chapter 2 Improving the prediction of overall survival for head and
neck cancer patients using image biomarkers in combination with clinical parameters
19
Chapter 3 The prognostic value of CT-based image-biomarkers for head
and neck cancer patients treated with definitive (chemo-) radiation
37
Chapter 4 Pre-treatment radiomic features predict individual lymph
node failure for head and neck cancer patients 59
Chapter 5 External validation of nodal failure prediction models
inclu-ding radiomics in head and neck cancer 91
Chapter 6 Discussion and future perspectives 113
Appendices Chinese summary (中文摘要) 141
Nederlandse samenvatting (Dutch summary) 145
Curriculum vitae 152
List of publications 153
CHAPTER 1
1.1 Introduction
Head and neck cancer (HNC) originating from the oral cavity, larynx and pharynx is responsible for about 0.83 million new cancer cases and 0.43 million cancer deaths worldwide every year. These are predominantly squamous cell carcinoma (SCC) [1,2]. Much progress has been made in the treatment of HNC in the last two decades. The introduction of chemotherapy, better understanding of human papillomavirus (HPV)-related oropharynx SCC, and multidisciplinary management, has resulted in improved overall survival (OS) rates [3–5].
Based on data from the Surveillance, Epidemiology and End Results (SEER) program and the Netherlands Cancer Registry, the 5-year OS rate for HNC patients is approximately 60%. For patients with early stage disease, the 5-year OS rate can even reach 70-90% [6]. However, 30%-50% of patients with locally advanced head and neck squamous cell carcinoma (HNSCC) still experience treatment failures, predominantly occurring at the site of the primary tumour, followed by regional failures and distant metastases [7]. The prolonged life expectancy of HNC patients has consequently increased the number of patients with acute and late toxicity following treatment. To enable more personalised treatment approaches for HNC patients, there is a rising demand for adequate prediction of treatment failure, as well as complications [5,8–11].
Normal tissue complication probability (NTCP) models and tumour control probability (TCP) models are used for the prediction of treatment outcome, this can mean treatment-related side effects and/or tumour control. For HNSCC patients, the most common treatment-related side effects are xerostomia, dysphagia and tube feeding dependence, while the main treatment failures are local failure, regional failure, distant metastasis and death. The current NTCP models of the most frequently reported side effects and TCP models of the main treatment failures of HNSCC are mainly based on classic prognostic factors such as tumour stage, performance status, age, baseline toxicity scores, dose-volume parameters, etc. [7,10,12–15]. However, patients with similar prognostic factors may still have different outcomes. In other words, these TCP and NTCP models need to be further improved before they can be used for future personalised medicine [16]. New emerging data in the form of radiomics, also called image biomarkers (IBM), reflect
1
the intensity, shape and textural heterogeneity of the region of interest derived from medical images. Such imaging data has shown significant associations with survival and complications in HNSCC patients [17,18]. An important question that has remained unanswered in previous publications is about the the extent to which the addition of these radiomic features improves the predictive power of models consisting only of classical prognostic factors, such as TNM staging, performance status and baseline toxicity scores.
The additional role of radiomic features in predicting radiation-induced toxicities for HNSCC patients has been explored in a thesis by LV van Dijk (UMC Groningen, 2018) and associated publications [18–20]. This thesis focused on improving the prediction of treatment failures.
The overarching aim of this thesis was to evaluate the prognostic ability of radiomic features and to test whether the performance of prediction models for different treatment failures could be improved by the addition of radiomics to the classical prognostic factors for HNSCC patients primarily treated with radiotherapy.
1.2 Candidate features
Classic prognostic factors such as clinical and biological factors have been shown to influence treatment response, and new emerging radiomics (image biomarkers) have also demonstrated their prognostic values in a series of studies [21–27]. All of them are included as candidate parameters in the analyses described within thesis.
Clinical features
Currently, clinical factors are most frequently used in routine clinical practice to estimate local control (LC), regional control (RC), distant metastasis-free survival (DMFS), disease-free survival (DFS) and overall survival (OS) rates of individual patients. The clinical factors can be categorised as follows: (a) tumour-related and (b) patient-related prognostic factors.
TNM classification is the most important tumour-related feature and has been used in routine clinical practice to guide treatment decision-making [8,22,28]. Generally, systemic therapy is recommended for patients with advanced TNM stage to improve tumour control. Other similar features such as tumour volume, tumour diameter and overall
stage have also frequently been reported [21,22,29,30]. Besides TNM classification, large variations in OS and LC can be found in patients with HNSCC from different tumour subsites [28]. Despite being centimeters apart, nasopharyngeal and laryngeal cancer have a higher radio-sensitivities than oral cavity and hypopharynx cancer [7,10,25,31]. In addition to tumour-related prognostic factors, patient-related features such as high alcohol intake, active smoking status, presence of co-morbidities, advanced age, poor WHO performance score, instance of weight loss and male gender all showed significant associations with worse LC, RC, DFS and OS [11,22,28,32]. These factors are currently taken into consideration when treatment decisions are made. For example, patients with a poor WHO performance score who are older than 70 years are not considered candidates for systemic treatment, while surgery is not preferred for patients with severe co-morbidities [8].
Histopathological/molecular biological features
Many histopathological and molecular biological markers that are associated with prognosis have been previously studied and reported. Extranodal extension (ENE) and p16 are the most important biomarkers identified for HNSCC. These have therefore been included in the AJCC 8th TNM classification to guide treatment decisions [5,8,33–35]. Additionally, the over-expression of epidermal growth factor receptor (EGFR) showed an association with a poor response to radiotherapy. Also, the concurrent use of anti-EGFR monoclonal antibody cetuximab with radiotherapy has been shown superior to radiotherapy alone, with improved survival [36]. There are many other potential biological factors, such as gene mutations related to hypoxia, intrinsic radio-sensitivity and apoptosis, that are likely to play a role in determining a patient’s radiation response. However, none have yet entered routine clinical practice [22,37].
A main concern with histopathological/molecular biological features is that the tests are not always available and feasible in the clinic. For patients treated with primary non-surgical modalities (the patient population discussed in this thesis), such as (chemo) radiotherapy, only limited pathological information is available. Furthermore, tests based on the limited biopsy specimen might not be reliable and representative for the whole tumour [22,34].
1
Radiomics, or IBM, refers to the comprehensive quantification of tumour phenotypes based on medical images. A wide variety of medical images is already available for diagnostic, staging and radiotherapy treatment planning purposes. Radiomic features can be extracted from these medical imaging modalities (computed tomography (CT), magnetic resonance imaging (MRI), positron emission tomography (PET) or ultrasound) without the need for additional image acquisition [38,39]. For HNSCC patients primarily treated with radiotherapy (RT) at the University Medical Centre Groningen (UMCG), a pre-treatment planning CT scan is always acquired and saved for every patient. Therefore a large dataset is available to study CT-based radiomic features. Compared with the histopathological/molecular biological features based on the limited and invasive needle biopsy specimens, radiomics is able to capture the phenotype of the whole tumour non-invasively [38]. Moreover, radiomic features are objectively calculated by using standard formulas, providing quantitative information regarding intensity, shape and textural characteristics of a region of interest. This transforms three-dimensional morphological tumor information into multi-dimensional and mineable data [16,17,40–44]. Together, the advantages of radiomic features make them very promising candidates for the prediction of patient-specific treatment outcome.
1.3 Study cohort
It is important to study prognostic factors in a well-defined patient group treated with the same modalities [12]. In this thesis, with respect to this criterion, we focused on the HNSCC patients who were primarily treated with definitive radiotherapy, which accounts for two-thirds of new HNSCC patients [45,46]. All studies in this thesis are retrospective analyses, which was composed of 707 consecutive, prospectively collected, non-surgically treated HNSCC patients from University Medical Centre Groningen (UMCG) , the Netherlands, 113 HNSCC patients from department of radiation oncology in Maastricht (MAASTRO), the Netherlands, and 289 nasopharyngeal cancer patients from Shantou University Medical College (SUMC), China with a standardised follow-up programme [47].
The first part of this thesis (Chapter 2 and 3) focussed on improving the prediction of
different prognostic endpoints with CT-based radiomic features.
We tested the hypothesis that the prediction of overall survival (OS) could be improved by adding radiomic features to clinical features in Chapter 2. Furthermore, the ability to
generalise the prognostic value of radiomics for different tumour types was investigated by training a model on nasopharyngeal cancer patients and externally validating this model on other HNC sub-types.
With the knowledge gained in Chapter 2, we subsequently tested the performance of radiomic features in a systematic and thorough analysis using local control (LC), regional control (RC), distant metastasis-free survival (DMFS) and disease-free survival (DFS) rate as endpoints in Chapter 3.
We expected that prognostic models including radiomic and clinical features could be used to identify patients with a high risk of treatment failures prior to treatment, and consequently could support more effective personalised treatment approaches.
The second part of this thesis (Chapter 4 and Chapter 5) focused on advanced HNSCC
patients with pathological lymph nodes (N+). Instead of identifying patients who are at high risk of treatment failures, in this study we tried to estimate the failure risk of each individual pathological lymph node. If we can identify those lymph nodes with high failure risk prior to treatment, intensified radiation schedules could be applied to the specific high-failure risk node to improve nodal control without increasing the dose to normal tissue. Alternatively, pre-treatment lymph node-targeted dissection could be arranged to avoid severe post-operative complications.
Chapter 4 explored the predictive clinical and radiomic features that could be used to
identify pathological lymph nodes that have a large risk of persistence or recurrence. A pre-treatment prediction model for nodal failures was developed and internally validated
in Chapter 4. Before this model is introduced into the clinic, a TRIPOD (transparent
reporting of a multivariable prediction model for individual prognosis or diagnosis) type 4 level model validation [48] would be required. Therefore, we validated the model in an external cohort of HNSCC patients from the MAASTRO clinic in Chapter 5.
The findings of this thesis are summarised and future perspectives are discussed in
1
REFERENCES
1. Bray F, Ferlay J, Soerjomataram I, Siegel RL, Torre LA, Jemal A. Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin 2018;68:394–424.
2. Baxi S, Fury M, Ganly I, Rao S, Pfister DG. Ten years of progress in head and neck cancers. JNCCN J Natl Compr Cancer Netw 2012;10:806–10.
3. Pignon JP, Maître A le, Maillard E, Bourhis J. Meta-analysis of chemotherapy in head and neck cancer (MACH-NC): An update on 93 randomised trials and 17,346 patients. Radiother Oncol 2009;92:4–14.
4. Blanchard P, Baujat B, Holostenco V, Bourredjem A, Baey C, Bourhis J, et al. Meta-analysis of chemotherapy in head and neck cancer (MACH-NC): A comprehensive analysis by tumour site. Radiother Oncol 2011;100:33–40.
5. Ang KK, Harris J, Wheeler R, Weber R, Rosenthal DI, Nguyen-Tân PF, et al. Human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010;363:24–35.
6. Howlader N, Noone AM, Krapcho M, et al. SEER cancer statistics review, 1975-2013, based on November 2015 SEER. http://seer.cancer.gov/csr/1975_2013/.
7. Pagh A, Grau C, Overgaard J. Failure pattern and salvage treatment after radical treatment of head and neck cancer. Acta Oncol (Madr) 2016;55:625–32.
8. Pfister DG, Spencer SA, Brizel DM, Burtness BA, Busse PM, Caudell JJ, et al. Head and neck cancers, version 1.2015 featured updates to the NCCN guidelines. JNCCN J Natl Compr Cancer Netw 2015;13:847–56.
9. Liu MT, Chang TH, Lin JP, Huang CC, Wang AY. Prognostic factors affecting the outcome of nasopharyngeal carcinoma. Jpn J Clin Oncol 2003;33:501–508.
10. Regueiro CA, Aragón G, Millán L, Valcárcel FJ, de la Torre A, Magallón R. Prognostic factors for local control, regional control and survival in oropharyngeal squamous cell carcinoma. Eur J Cancer 1994;30:2060–7.
11. Langius JAE, Bakker S, Rietveld DHF, Kruizenga HM, Langendijk JA, Weijs PJM, et al. Critical weight loss is a major prognostic indicator for disease-specific survival in patients with head and neck cancer receiving radiotherapy. Br J Cancer 2013;109:1093-1099.
12. van den Brekel MWM, Bindels EMJ, Balm AJM. Prognostic factors in head and neck cancer. Eur J Cancer 2002;38(8):1041-1043.
13. Beetz I, Schilstra C, Van Der Schaaf A, Van Den Heuvel ER, Doornaert P, Van Luijk P, et al. NTCP models for patient-rated xerostomia and sticky saliva after treatment with intensity modulated radiotherapy for head and neck cancer: The role of dosimetric and clinical factors. Radiother Oncol 2012;105:101–6.
14. Christianen MEMC, Schilstra C, Beetz I, Muijs CT, Chouvalova O, Burlage FR, et al. Predictive modelling for swallowing dysfunction after primary (chemo)radiation:
Results of a prospective observational study. Radiother Oncol 2012;105:107–14. 15. Wopken K, Bijl HP, Langendijk JA. Prognostic factors for tube feeding dependence after
curative (chemo-) radiation in head and neck cancer: A systematic review of literature. Radiother Oncol 2018;126:56–67.
16. Lambin P, Leijenaar RTH, Deist TM, Peerlings J, De Jong EEC, Van Timmeren J, et al. Radiomics: The bridge between medical imaging and personalized medicine. Nat Rev Clin Oncol 2017;14:749–62.
17. Aerts HJWL, Velazquez ER, Leijenaar RTH, Parmar C, Grossmann P, Cavalho S, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 2014;5:4006.
18. van Dijk L V., Brouwer CL, van der Schaaf A, Burgerhof JGM, Beukinga RJ, Langendijk JA, et al. CT image biomarkers to improve patient-specific prediction of radiation-induced xerostomia and sticky saliva. Radiother Oncol 2017;122:185–91.
19. van Dijk L V., Noordzij W, Brouwer CL, Boellaard R, Burgerhof JGM, Langendijk JA, et al. 18F-FDG PET image biomarkers improve prediction of late radiation-induced xerostomia. Radiother Oncol 2018;126:89–95.
20. van Dijk L V., Thor M, Steenbakkers RJHM, Apte A, Zhai TT, Borra R, et al. Parotid gland fat related Magnetic Resonance image biomarkers improve prediction of late radiation-induced xerostomia. Radiother Oncol 2018;128:459–66.
21. Mehanna H, West CML, Nutting C, Paleri V. Head and neck cancer - Part 2: Treatment and prognostic factors. BMJ 2010;341:721–5.
22. Silva P, Homer JJ, Slevin NJ, Musgrove BT, Sloan P, Price P, et al. Clinical and biological factors affecting response to radiotherapy in patients with head and neck cancer: A review. Clin Otolaryngol 2007;32:337–45.
23. Leoncini E, Vukovic V, Cadoni G, Pastorino R, Arzani D, Bosetti C, et al. Clinical features and prognostic factors in patients with head and neck cancer: Results from a multicentric study. Cancer Epidemiol 2015;39:367–74.
24. Xie K, Chen J, Zou J, Chen L, Gong L. Tumor volumes predict prognosis in head and neck cancer: a meta-analysis. Transl Cancer Res 2017;6:687–97.
25. Zhai TT, Langendijk JA, van Dijk L V., Halmos GB, Witjes MJH, Oosting SF, et al. The prognostic value of CT-based image-biomarkers for head and neck cancer patients treated with definitive (chemo-)radiation. Oral Oncol 2019;95:178–86.
26. Zhai TT, van Dijk L V., Huang BT, Lin ZX, Ribeiro CO, Brouwer CL, et al. Improving the prediction of overall survival for head and neck cancer patients using image biomarkers in combination with clinical parameters. Radiother Oncol 2017;124:256–62.
27. Zwanenburg A, Leger S, Vallières M, Löck S. Image biomarker standardisation initiative 2016; arxiv. org/abs/1612.07003
28. Avilés-Jurado FX, León X. Prognostic factors in head and neck squamous cell carcinoma: Comparison of CHAID decision trees technology and cox analysis. Head Neck
1
2013;35:877–83.
29. Folkert MR, Oh JH, Setton J, Apte AP, Thorstad WL, Schoder H, et al. Predictive Modeling of Outcomes Following Definitive Chemoradiation Therapy for Oropharyngeal Cancer Based on FDG-PET Image Characteristics. Int J Radiat Oncol 2013;87:S3.
30. van Den Broek GB, Rasch CRN, Pameijer FA, Peter E, Van Den Brekel MWM, Tan IB, et al. Pretreatment probability model for predicting outcome after intraarterial chemoradiation for advanced head and neck carcinoma. Cancer 2004;101(8):1809-1817.
31. de Jong RJB, Hermans J, Molenaar J, Briaire JJ, Cessie S Le. Prediction of survival in patients with head and neck cancer. Head Neck 2001;2:718-724.
32. Folkert MR, Setton J, Apte AP, Grkovski M, Young RJ, Schöder H, et al. Predictive modeling of outcomes following definitive chemoradiotherapy for oropharyngeal cancer based on FDG-PET image characteristics. Phys Med Biol 2017;62(13):5327-5343.
33. Mermod M, Tolstonog G, Simon C, Monnier Y. Extracapsular spread in head and neck squamous cell carcinoma: A systematic review and meta-analysis. Oral Oncol 2016;62:60–71.
34. Chai RL, Rath TJ, Johnson JT, Ferris RL, Kubicek GJ, Duvvuri U, et al. Accuracy of computed tomography in the prediction of extracapsular spread of lymph node metastases in squamous cell carcinoma of the head and neck. JAMA Otolaryngol - Head Neck Surg 2013;139:1187–94.
35. Dayyani F, Etzel CJ, Liu M, Ho CH, Lippman SM, Tsao AS. Meta-analysis of the impact of human papillomavirus (HPV) on cancer risk and overall survival in head and neck squamous cell carcinomas (HNSCC). Head Neck Oncol 2010;2:15.
36. Bentzen SM, Atasoy BM, Daley FM, Dische S, Richman PI, Saunders MI, et al. Epidermal growth factor receptor expression in pretreatment biopsies from head and neck squamous cell carcinoma as a predictive factor for a benefit from accelerated radiation therapy in a randomized controlled trial. J Clin Oncol 2005;23:5560-5567. 37. West CML. Intrinsic radiosensitivity as a predictor of patient response to radiotherapy.
Br J Radiol 1995;68:827-837.
38. Aerts HJWL. The potential of radiomic-based phenotyping in precisionmedicine a review. JAMA Oncol 2016;2:1636–42.
39. Caudell JJ, Torres-Roca JF, Gillies RJ, Enderling H, Kim S, Rishi A, et al. The future of personalised radiotherapy for head and neck cancer. Lancet Oncol 2017;18:e266– 73.
40. Grove O, Berglund AE, Schabath MB, Aerts HJWL, Dekker A, Wang H, et al. Quantitative computed tomographic descriptors associate tumor shape complexity and intratumor heterogeneity with prognosis in lung adenocarcinoma. PLoS One
2015;10: e0118261.
41. Cui Y, Song J, Pollom E, Alagappan M, Shirato H, Chang DT, et al. Quantitative Analysis of 18F-Fluorodeoxyglucose Positron Emission Tomography Identifies Novel Prognostic Imaging Biomarkers in Locally Advanced Pancreatic Cancer Patients Treated With Stereotactic Body Radiation Therapy. Int J Radiat Oncol Biol Phys2016;96:102-9 42. Haralick RM, Dinstein I, Shanmugam K. Textural Features for Image Classification. IEEE
Trans Syst Man Cybern 1973;3:610–21.
43. Tang X. Texture information in run-length matrices. IEEE Trans Image Process 1998;7:1602-1609.
44. Thibault G, Fertil B, Navarro C, Pereira S. Texture Indexes and Gray Level Size Zone Matrix Application to Cell Nuclei Classification. Patern Recognit Inf Process 2009:140-145. 45. Barton MB, Jacob S, Shafiq J, Wong K, Thompson SR, Hanna TP, et al. Estimating the
demand for radiotherapy from the evidence: A review of changes from 2003 to 2012. Radiother Oncol 2014;112:140-144.
46. Naghavi AO, Echevarria MI, Strom TJ, Abuodeh YA, Venkat PS, Ahmed KA, et al. Patient choice for high-volume center radiation impacts head and neck cancer outcome. Cancer Med 2018;7:4964–79.
47. Vergeer MR, Doornaert PAH, Rietveld DHF, Leemans CR, Slotman BJ, Langendijk JA. Intensity-Modulated Radiotherapy Reduces Radiation-Induced Morbidity and Improves Health-Related Quality of Life: Results of a Nonrandomized Prospective Study Using a Standardized Follow-Up Program. Int J Radiat Oncol Biol Phys 2009;74:1-8.
48. Collins GS, Reitsma JB, Altman DG, Moons KGM. Transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD): The TRIPOD Statement. Eur Radiol 2015;67:1142–51.
CHAPTER 2
Improving the prediction of overall survival
for head and neck cancer patients using
image biomarkers in combination with clinical
parameters
Published in: Radiotherapy and Oncology 2017 August;124(2):256-262.
Tian-Tian Zhaia,b,*, Lisanne V. van Dijka, Bao-Tian Huangb, Zhi-Xiong Linb,*, Cássia O.
Ribeiroa, Charlotte L. Brouwera, Sjoukje F. Oostingc, Gyorgy B. Halmosd, Max J.H. Witjese,
Johannes A. Langendijka, Roel J.H.M. Steenbakkersa, Nanna M. Sijtsemaa. a Department of Radiation Oncology, University Medical Center Groningen, University
of Groningen, Groningen, The Netherlands
b Department of Radiation Oncology, Cancer Hospital of Shantou University Medical
College, Shantou, China
c Department of Medical Oncology, University Medical Center Groningen, University of
Groningen, Groningen, The Netherlands
d Department of Otolaryngology, University Medical Center Groningen, University of
Groningen, Groningen, The Netherlands
e Department of Maxillofacial Surgery, University Medical Center Groningen, University
of Groningen, Groningen, The Netherlands *Corresponding author
Abstract
Purpose
To develop and validate prediction models of overall survival (OS) for head and neck cancer (HNC) patients based on image biomarkers (IBMs) of the primary tumor and positive lymph nodes (Ln) in combination with clinical parameters.
Material and methods
The study cohort was composed of 289 nasopharyngeal cancer (NPC) patients from China and 298 HNC patients from the Netherlands. Multivariable Cox-regression analysis was performed to select clinical parameters from the NPC and HNC datasets, and IBMs from the NPC dataset. Final prediction models were based on both IBMs and clinical parameters.
Results
Multivariable Cox-regression analysis identified three independent IBMs (tumor Volume-density, Run Length Non-uniformity and Ln Major-axis-length). This IBM model showed a concordance(c)-index of 0.72 (95%CI: 0.65-0.79) for the NPC dataset, which performed reasonable with a c-index of 0.67 (95%CI: 0.62-0.72) in the external validation HNC dataset. When IBMs were added in clinical models, the c-index of the NPC and HNC datasets improved to 0.75 (95%CI: 0.68-0.82; p=0.019) and 0.75 (95%CI: 0.70-0.81; p<0.001), respectively.
Conclusion
The addition of IBMs from the primary tumor and Ln improved the prognostic performance of the models containing clinical factors only. These combined models may improve pre-treatment individualized prediction of OS for HNC patients.
2
Introduction
Head and neck cancer (HNC) accounts for about 0.65 million new cancer cases and 0.35 million cancer deaths worldwide every year [1]. Based on the Surveillance, Epidemiology, and End Results (SEER) data, the 5-year overall survival (OS) for HNC patients is approximately 60% [2]. The introduction of more intensified treatment regimens has resulted in improved OS rates, however the number of patients developing locoregional failure or distant metastases remains substantial [3,4]. To enable more personalized treatment approaches, risk stratification is becoming increasingly important [5]. Risk stratification in HNC requires new, robust and prognostic parameters to identify patients with different risk profiles for locoregional recurrence, distant metastasis and death [6-8].
In routine clinical practice, the TNM staging system is used to guide treatment decision-making often in combination with other classical prognostic factors such as performance status, tumor characteristics and age [9,10]. However, patients with similar prognostic factors may have different outcome [6,7] and thus new prognostic factors are needed to improve outcome prediction accuracy when added to prediction models based on classical prognostic factors only.
Recent studies have demonstrated the potential value of image biomarkers (IBMs), which are significantly associated with OS and complications in HNC, thoracic, pancreatic and colorectal cancer [11-13]. IBMs can be extracted from medical images and provide quantitative information regarding intensity, shape and textural characteristics of the region of interest [14-17]. By extracting IBMs, the three-dimensional morphological tumor information can be transformed into multi-dimensional and mineable data [5,18]. Furthermore, IBMs enable decoding of a general prognostic phenotype existing in different cancer types, which may widen the scope of application [11].
Although many IBMs are significantly associated with outcome, it remains unclear to what extent the addition of IBMs improves the predictive power of models only consisting of classical prognostic factors, such as TNM staging and performance status. The aim of this study was to test whether the performance of prediction models for OS could be improved by the addition of IBMs compared to models based on solely classical prognostic factors for nasopharyngeal cancer (NPC) patients. Furthermore, the ability
to generalize the prognostic value of IBMs for different tumor types was determined by externally validating this value for other HNC subtypes.
Materials and Methods
Patient demographics and treatment
This retrospective study composed of 289 consecutive NPC patients. Patients were treated with (chemo-)radiotherapy between January 2010 and June 2011 at the Cancer Hospital of Shantou University Medical College. All patients received a pre-treatment computed tomography (CT) scan (Philips Brilliance CT Big Bore Oncology Configuration, Cleveland, OH, USA; voxel size: 1.0×1.0×3.0 mm; scan voltage: 120kV; convolution kernel: Philips Healthcare’s B) for radiotherapy planning. Patients were primarily treated with intensity-modulated radiotherapy (IMRT) and received a total dose of 70.4 Gy with fractions of 2.2 Gy in 6.5 weeks (5 fractions per week).
An additional set of 298 consecutive HNC patients (including 4.4% NPC patients) was treated with definitive radiotherapy, either combined or not, with chemotherapy or cetuximab at the University Medical Center Groningen between November 2007 and May 2013. For all patients, a pre-treatment CT-scan (Somatom Sensation Open, Siemens, Forchheim, Germany; voxel size: 1.0×1.0×2.0 mm; scan voltage: 120kV; convolution kernel: B30) was acquired for radiotherapy planning. Radiotherapy consisted of primarily three-dimensional conformal radiotherapy or IMRT to a total dose of 70 Gy with fractions of 2 Gy in 6-7 weeks (6 or 5 fractions per week).
Inclusion criteria were as follows: confirmed primary tumor with pathological diagnosis, standard contrast-enhanced planning CT-scan, treatment with curative intent, and OS data available.
Clinical parameters
All clinical parameters including age, gender, tumor location, treatment modality, human papilloma virus (HPV) status (only for oropharyngeal cancer (OPC)) and World Health Organization performance status (WHO PS) [19] were derived from medical records. Dose-volume information of the primary tumor (PT) and positive lymph nodes (pLN) was derived from the radiotherapy planning system(mean dose, V50, V60, V70, V80, D90%, D95% and D98%). Tumor (T) and positive lymph node (N) stage were defined according
2
to the 6th edition of the American Joint Committee on Cancer Staging Manual [10].
CT Image biomarkers
The PT and pLN were delineated for the NPC and HNC datasets on the planning CT-scan by experienced head and neck radiation-oncologists. In-house software was used to extract the IBMs, developed using common formulas in Matlab R2014a (Mathworks, Natick, USA). Twenty-four CT intensity and 20 geometric IBMs were directly derived from every delineated structure (the PT, all pLN and the pLN with the largest volume). The intensity IBMs were obtained from the histogram of all voxel values, such as median of the voxels and entropy of the voxels. Geometric IBMs, such as volume, compactness and major axis length, were calculated from the three-dimensional shape and size of the contoured structures. Ninety textural CT IBMs from both the PT and the pLN with the largest volume were defined to quantify the heterogeneity of tissue. They were derived from three different matrices: the grey level co-occurrence matrix (GLCM) [15], grey level run-length matrix (GLRLM) [16] and grey level size-zone matrix (GLSZM) [17]. GLCM describes the grey level transition, GLRLM and GLSZM describe the directional and volumetric grey level repetition. They were calculated from the three-dimensional contoured structures. More details on feature extraction and used algorithms are described in our previous publication [20]. The lymph node IBMs from patients without lymph node metastasis were defined as 0.
Data analysis
The endpoint of this study was OS, defined as the time from the first day of radiotherapy to the date of death from any cause. Patients alive were censored at the date of last follow up. An overview of the analysis design is shown in Figure 1.
Step 1 : Clinical models
Potential clinical parameters that were considered for their prognostic ability in the NPC and HNC datasets included age (>median vs. ≤median), gender (female vs. male), T-stage (T3-T4 vs. T1-T2), N-stage (N2-N3 vs. N0-N1), treatment modality (RT with systemic treatment vs. RT only), WHO PS (1-3 vs. 0) and dose parameters (>median vs. ≤median). HPV status assessed by p16 immunohistochemistry and DNA polymerase chain reaction (OPC positive vs. others) was included in the analysis for the HNC dataset, as this is a strong risk factor for oropharyngeal cancer [10, 21-22].
Due to the known difference in etiology between NPC and HNC, two multivariable clinical prediction models were created: one based on the NPC and the other on the HNC dataset.
Fig. 1. Analysis workflow. Step 1 Training: Two different clinical prediction models were created: for each of
the NPC and HNC datasets. Step 2 External validation: an IBM model was created based on the NPC dataset and externally validated using the HNC dataset. Step 3 Update: The combined models were created by combining the IBMs with the clinical parameters from NPC and HNC datasets separately. Abbreviations: NPC = nasopharyngeal cancer; c-index = concordance index; HNC = head and neck cancer; IBM = image biomarker.
Step 2: IBM model
IBM variables were pre-selected to reduce the probability of over-fitting. If the Pearson correlation between pairs of IBMs was larger than 0.80, then the IBM with the lower univariable association with OS was omitted from further analysis [23-24]. All pre-selected potential IBMs were analysed for their prognostic power, using their median value (> median vs. ≤ median) in the NPC dataset as the threshold value in the univariable analysis. After selection of the independent prognostic factors, the threshold values were optimized by testing the values around the median. A multivariable IBM model was developed based on the NPC dataset only. Finally, the thresholds of IBMs for the NPC dataset were used for the HNC dataset to externally validate the IBM model.
Step 3: Combined models
All clinical parameters from the NPC and HNC datasets and pre-selected IBMs from the NPC dataset were merged into multivariable analysis and the coefficients (β) of the features were refitted to the NPC dataset and HNC dataset respectively, to generate the combined IBM-NPC and IBM-HNC models.
2
test were used to test the normality of all potential clinical parameters and IBMs. The chi-square test was used to compare the rates and an independent sample t-test was used to compare normally distributed variables between different groups. Univariable Cox regression analysis was performed to assess the risk factors for OS, and multivariable Cox proportional hazards regression analysis (forward selection: Likelihood ratio test, p-value < 0.05) was used for the development of the multivariable model. The concordance index (c-index) was determined to assess the models discriminative power. The z-score test (Package “compareC” in R) was used to test the difference between two c-indices. Internal validation was performed by using bootstrap validation [25]. The differences in c-indices of the bootstrap model on the bootstrap sample and the original sample were calculated 1000 times. The optimism-corrected c-index was obtained by subtracting the average c-indices difference. The median of the linear predictor was defined as the threshold to separate the Kaplan-Meier survival curves: one curve showing patients with high hazard values (high risk) and the other with low hazard values (low risk). The Kaplan-Meier survival curves were compared using a log-rank test. Two tailed p-values < 0.05 were considered statistically significant. Statistical analysis was performed using the R software (version 3.2.1).
Results
The median follow up for the NPC patients was 37.6 months (range 2.4-58.6) and overall 64 deaths (22.1%) were observed. The median follow up for the HNC patients was 32.8 months (range: 1.6-89.7) and overall 126 deaths (42.3%) were observed.
Step 1: Clinical models
The clinical characteristics of the patients in the NPC and HNC datasets are listed in Table 1. All characteristics except gender were significantly different between the two datasets.
In the NPC dataset, the univariable analysis revealed significant associations between the minimum dose value of 98% of the primary tumor volume (D98% of GTV-PT), age, N-stage and T-stage with OS (Table S1). In the multivariable analysis (Table S2), lower values of D98% of GTV-PT (hazard ratio (HR): 0.37; 95% confidence interval (CI): 0.21-0.63), increasing age (HR: 2.14; 95%CI: 1.30-3.53) and N-stage (HR: 2.37; 95%CI:
1.26-4.45) were associated with worse OS. The multivariable clinical NPC model based on these three clinical features resulted in a c-index of 0.69 (95%CI: 0.61-0.76).
Univariable analysis showed that D98% of GTV-PT, WHO PS, N-stage, T-stage and HPV-status were significantly associated with OS (Table S1) in the HNC dataset. The multivariable clinical HNC model (Table S2) including WHO PS (HR: 3.46; 95%CI: 2.41-4.97), D98% of GTV-PT (HR: 0.51; 95%CI: 0.35-0.74) and T-stage (HR: 1.93; 95%CI: 1.28-2.91) as independent prognostic factors for OS, had a c-index of 0.72 (95%CI: 0.67-0.78).
Table 1 Baseline characteristics of the patients in the NPC and HNC datasets. Characteristic NPC dataset (n=289)(%) HNC dataset (n=298)(%) p-Value Gender 0.489b Male 203 (70.2) 217 (72.8) Female 86 (29.8) 81 (27.2) Age at diagnosis (median ± SD)(year) 51±12 62±11 <0.001C T stagea <0.001b T1 22 (7.6) 22 (7.4) T2 49 (17.0 ) 103 (34.6) T3 104 (36.0 ) 90 (30.2) T4 114 (39.4 ) 83 (27.9) N stagea <0.001b N0 28 (9.7) 117 (39.3) N1 60 (20.8) 24 (8.1) N2 158 (54.7) 142 (47.7) N3 43 (14.9) 15 (5.0) Treatment modality <0.001b RT only 48 (16.6) 161 (54.0) RT with systemic Treatment 241 (83.4) 137 (46.0) D98% of GTV-PT (median ± SD)(Gy) 69.60 ± 2.30 68.22±3.10 <0.001C WHO PS 0 241(83.4) 192 (64.4) <0.001b 1-3 48 (16.6) 106 (35.6)
2
The Kaplan-Meier curves of clinical NPC and HNC models are shown in Figure 2A and 2B. The probability of OS after 2 and 3 years was calculated for the two datasets, and shown in Table S3.
Step 2: IBM model
Forty-eight of the 312 IBMs was pre-selected from the NPC dataset by testing the inter-variable correlation. In the univariable analysis, nine of the pre-selected IBMs were significantly associated with OS (Table S4). Multivariable analysis revealed three independent features: Volume-density (HR, 0.44; 95%CI: 0.26-0.74), Run Length Non-uniformity (RLN) (HR, 2.98; 95%CI, 1.73-5.14) from the PT and Major-axis-length from all pLN (threshold: 55 mm; HR: 2.11; 95% CI, 1.29-3.46) (Table S5). The optimal threshold was the median unless stated otherwise. A multivariable IBM model was developed based on the three IBMs and resulted in a c-index of 0.72 (95%CI: 0.65-0.79). External validation using the HNC dataset resulted in a c-index of 0.67 (95%CI: 0.62-0.72) (Figure 2C and 2D). The subgroup multivariable analysis using patient data with positive lymph
Table 1 Baseline characteristics of the patients in the NPC and HNC datasets-continued. Characteristic NPC dataset (n=289)(%) HNC dataset(n=298)(%) p-Value HPV OPC negative - 68 (22.8) OPC positive - 29 (9.7) Not OPC - 201 (67.4) Tumor site Nasopharynx 289 (100.0) 13 (4.4) Oropharynx - 97(32.6) Hypopharynx - 26(8.7) Larynx - 120(40.3) Oral cavity - 29(9.7) Others - 13(4.4)
Abbreviations: NPC = nasopharyngeal caner; HNC = head and neck cancer; T = tumor; N = lymph node; RT = radiotherapy; D98 of GTV-PT = minimum dose value of 98% volume of primary tumor; WHO PS = World Health Organization performance status; HPV = human papilloma virus; OPC = oropharyngeal cancer.
a According to the 6th edition of the AJCC/UICC staging system. b p-Value was calculated using the chi-square test.
nodes only revealed the same three significant IBMs with similar model performance.
Fig. 2. Overall survival stratified by risk groups for the NPC and HNC datasets. The survival curve separation
and hazard ratio (>median vs. ≤median) between different risk groups are shown based on the clinical model of NPC dataset (A) and HNC dataset (B), IBM model in NPC dataset (C) and HNC dataset (D), the combined IBM-NPC model (E) in the IBM-NPC dataset and the combined IBM-HNC model in the HNC dataset (F). Abbreviations: NPC = nasopharyngeal cancer; HNC = head and neck cancer; IBM = image biomarker; HR = hazard ratio; CI = confidence interval.
2
Step 3: Combined models
For the NPC dataset, RLN, Volume-density, lymph nodes’ Major-axis- length, D98% of GTV-PT and age were identified as independent prognostic factors. However, N-stage was no longer significantly
related to OS. Compared to the clinical prediction model, the combined model showed a significantly improved c-index of 0.75 (95%CI: 0.68-0.82; p=0.019)(Figure 2E). For the HNC dataset, WHO PS, RLN, lymph nodes’ Major-axis-length, T-stage, Volume-density and D98% of GTV-PT were identified as independent prognostic factors. Compared to the clinical HNC model, the combined IBM-HNC model showed better performance with an increased c-index of 0.75 (95%CI: 0.70-0.81; p<0.001) (Figure 2F and Table S6).The coefficients and performance of the two combined models are depicted in Table 2.
Discussion
Non-invasive tools to predict treatment outcome could add value in the guidance of individual therapeutic strategies for HNC patients [26]. In this study, two PT IBMs
Table 2 Estimated coefficients of the combined models.
Combined IBM-NPC model Combined IBM-HNC model β Correct-ed β p-Value (95%CI)HR β rected
Cor-β p-Value HR (95%CI) Run Length Non-unifor-mity 0.69 0.47 0.028 2.00 (1.08-3.69) 0.37 0.24 0.099 (0.93-2.25)1.45 Volume-density 0.88 -0.60 0.001 (0.24-0.71)0.42 -0.41 -0.27 0.023 (0.46-0.94)0.66 Major-ax-is-length 0.77 0.52 0.003 (1.31-3.57)2.16 1.33 0.86 <0.001 3.79 (2.00-7.19) D98% of GTV-PT 0.84 -0.57 0.008 0.43 (0.23-0.81) -0.44 -0.29 0.032 (0.43-0.96)0.65 Age 0.56 0.38 0.029 (1.06-2.91)1.76 - - - -T stage - - - - 0.55 0.36 0.020 (1.09-2.76)1.74 WHO PS - - - - 1.36 0.88 <0.001 (2.66-5.66)3.88 Abbreviations: IBM = image biomarker; NPC = nasopharyngeal caner; HNC = head and neck cancer; HR = Hazard ratio; CI = confidence interval; D98 of GTV-PT = minimum dose value of 98% volume of primary
(Volume-density and RLN) and one pLN IBM (Major-axis-length) were extracted from CT images of an NPC dataset. These IBMs performed well for the external HNC dataset and significantly improved the prognostic performance of the models based on the clinical parameters only, both in NPC and HNC populations. This study is the first to investigate not only IBMs of the primary tumor site but IBMs of the pathologic lymph nodes as potential prognostic factors.
Although NPC falls under the category of HNC, the prognosis of patients with NPC is more favourable than HNC of other sites [2,7]. Furthermore, NPC has a distinct etiology compared to other HNC sites. The risk factors for NPC (Epstein-Barr virus, salted fish, preserved foods and genetic components) are also different from those of HNC (HPV, smoking and alcohol) [27]. Moreover, the geographical distribution and incidence of NPC are different from other HNC [28-31]. Therefore, two separate clinical models for both the NPC and the HNC datasets were developed and the combined models were fitted to the respective cohort without external validation.
Volume-density refers to the tumor volume bounded by the smallest cube containing the tumor (Figure 3A and 3B). The smaller the Volume-density, the more irregular the tumor shape, indicating tumor growth patterns[25,32-33]. A more invasive, irregular-shaped tumor requires a larger bounding cube. In figure 3A, the tumor with a smaller Volume-density is more irregular shape. Irregularly shaped tumors were associated with worse OS, which may reflect the increase of their aggressive behavior[1,25,32-33]. The RLN is derived from the GLRLM, which describes the frequency of consecutive voxels with the same grey level value. RLN is low when runs are equally distributed along the run length in contoured structures. A higher RLN value indicates PT heterogeneity, our results showed it was associated with worse survival for the NPC and HNC datasets (Figure 3C and 3D). This observation is supported by other studies [8,34-35]. Aerts et al. quantified the intratumor heterogeneity by using another feature “Grey level Non-Uniformity”, which had a high correlation (0.79) with RLN [8] in our study. Multiple subclonal populations coexist in the tumor and their evolution causes the intratumor heterogeneity. This process has been shown at cellular and genetic levels [36-39]. However, intratumor heterogeneity, can be underestimated if determined from a single or limited tumor-biopsy sample [36,40]. Medical imaging could be a better tool for
2
detecting the spatial dimension of tumor features. We quantified IBMs by extracting features from the complete tumor volume, in other words IBMs reflect the overall tumor feature. It would be even possible to distinguish different regions within the tumor based on the IBMs. The association between IBMs and cellular and genetic information requires further investigation [11,38,41].
Fig. 3. Examples of patients with low (A) and high (B) values of Volume-density of the tumor. Examples of patients with low (C) and high (D) values of Run Length Non-uniformity of the tumor. Examples of Major-axis-length in patients with two unilateral (E) and two bilateral (F) positive lymph nodes.
The Major-axis-length of pLN is defined as the longest distance between any two voxels of the all pLN. This could either be for one pLN, or any two unilateral or bilateral pLNs (Figure 2E and 2F). Our results demonstrated that the longer length (> 55 mm) was associated with an increased risk of death. N-stage did not increase prediction performance for the combined IBM-NPC model. N1 stage represents patients with only unilateral metastasis. In this situation, patients with N1 would have much shorter Major-axis-length compared to patients with N2 and N3. In fact, we found that Major-axis-length from the entire pLN had a linear correlation of 0.67 with the clinical variable N-stage. However, in the univariable analysis, the c-index of Major-axis-length was larger than that of the N-stage (0.60 vs 0.57). Therefore, we expect that Major-axis-length would be a good substitute for N-stage to improve survival prediction in the future.
To explore the general prognostic cancer phenotype and enlarge the scope of the IBM model application, Aerts et al. developed prognostic models based on IBMs from a lung cancer dataset and validated them against HNC datasets (c-index: 0.69) [11]. We have shown that the NPC IBM model performed well against the HNC dataset (c-index: 0.67), supporting the generalization of prognostic IBMs.
When IBMs were added in clinical models, the c-index of the NPC and HNC datasets significantly improved from 0.69 to 0.75 (p=0.019) and from 0.72 to 0.75 (p<0.001), respectively. Figure 2 showed that the Kaplan-Meier curve separation and hazard ratio between high-low risk patients are larger if the patients are stratified with the combined IBM-NPC and IBM-HNC models (Figure 2E and 2F) than with the separate clinical (Figure 2A and 2B) and IBM (Figure 2C and 2D) models. The exact survival probability difference was shown in Table S3. For example, two years survival difference between low and high risk groups improved from 10.8% with the clinical model to 14.2% with the combined model in the NPC group.
Standardization of image acquisition and reconstruction between different institutions are necessary for sharing quantitative image analyses. Different CT slice thickness and convolution kernel may influence the IBM model accuracy. Although the IBM model was validated externally for the HNC dataset, the clinical models and also the combined models were only fitted to the respective cohort without external validation, external validation of the combined models is the subject of future study. Moreover, a more
2
systematic and thorough analysis using different endpoints, such as disease free survival and locoregional control, will also need to be investigated in the future.
Conclusion
In conclusion, the addition of image biomarkers from the primary tumor and positive lymph nodes improved the performance of clinical prediction models significantly for both NPC and HNC datasets. This addition could facilitate the pre-treatment individualized prediction of survival for head and neck cancer patients.
Online supplementary materials
References
1. Torre LA, Bray F, Siegel RL, Ferlay J, Lortet-tieulent J, Jemal A. Global cancer statistics, 2012. CA Cancer J Clin 2015;65:87–108.
2. Howlader N, Noone AM, Krapcho M, et al. SEER cancer statistics review, 1975-2013, based on November 2015 SEER. http://seer.cancer.gov/csr/1975_2013/.
3. Pignon JP, le Maître A, Maillard E, Bourhis J, MACH-NC Collaborative Group. Meta-analysis of chemotherapy in head and neck cancer (MACH-NC): an update on 93 randomised trials and 17,346 patients. Radiother Oncol 2009;92:4-14.
4. Blanchard P, Baujat B, Holostenco V, et al. Meta-analysis of chemotherapy in head and neck cancer (MACH-NC): A comprehensive analysis by tumour site. Radiother Oncol 2011;100:33-40.
5. Aerts HJ. The potential of radiomic-based phenotyping in precision medicine: a review. JAMA Oncol 2016;2:1636-42.
6. Jia WH, Huang QH, Liao J, et al. Trends in incidence and mortality of nasopharyngeal carcinoma over a 20-25 year period (1978/1983-2002) in Sihui and Cangwu counties in southern China. BMC Cancer 2006;6:178.
7. Argiris A, Karamouzis MV, Raben D, Ferris RL. Head and neck cancer. Lancet 2008;371:1695–1709.
8. Liu MT, Chang TH, Lin JP, Huang CC, Wang AY. Prognostic factors affecting the outcome of nasopharyngeal carcinoma. Jpn J Clin Oncol 2003;33:501–508.
9. Ma J, Mai HQ, Hong MH, et al. Is the 1997 AJCC staging system for nasopharyngeal carcinoma prognostically useful for Chinese patient populations? Int J Radiat Oncol Biol Phys 2001;50:1181-89
10. Pfister DG, Spencer S, Brizel DM, et al. Head and neck cancers, version 1.2015. J Natl Compr Canc Netw 2015;13:847–55.
11. Aerts HJ, Velazquez ER, Leijenaar RT, et al. Decoding tumour phenotype by noninvasive imaging using a quantitative radiomics approach. Nat Commun 2014;5:4006 12. Grove O, Berglund AE, Schabath MB, et al. Quantitative computed tomographic
descriptors associate tumor shape complexity and intratumor heterogeneity with prognosis in lung adenocarcinoma. PLoS One 2015;10: e0118261.
13. Huang YQ, Liang CH, He L, et al. Development and validation of a radiomics nomogram for preoperative prediction of lymph node metastasis in colorectal cancer. J Clin Oncol 2016;34:2157-64.
14. Cui Y, Song J, Pollom E, et al. Quantitative analysis of (18)F-fluorodeoxyglucose positron emission tomography identifies novel prognostic imaging biomarkers in locally advanced pancreatic cancer patients treated with stereotactic body radiation therapy. Int J Radiat Oncol Biol Phys 2016;96:102-9.
15. Haralick R, Shanmugan K, Dinstein I. Textural features for image classification. IEEE Trans Syst Man Cybern 1973;3:610–21.
2
16. Tang X. Texture information in run-length matrices. IEEE trans image process1998;7:1602–9.
17. Thibault G, Fertil B, Navarro C, et al. Texture indexes and gray level size zone matrix application to cell nuclei classification. Pattern Recognit Inf Process 2009;140–45. 18. Gillies RJ, Kinahan PE, Hricak H. Radiomics: images are more than pictures, they are
data. Radiology 2016;278:563-77.
19. Oken MM, Creech RH, Tormey DC, et al. Toxicity and response criteria of the Eastern Cooperative Oncology Group. Am J Clin Oncol 1982;5:649-55.
20. van Dijk LV, Brouwer CL, van der Schaaf A, et al. CT image biomarkers to improve patient-specific prediction of radiation-induced xerostomia and sticky saliva. Radiother Oncol 2017;122:185-91.
21. Ang KK, Harris J, Wheeler R, et al. Human papillomavirus and survival of patients with oropharyngeal cancer. N Engl J Med 2010;363:24-35.
22. Kano M, Kondo S, Wakisaka N, et al. The influence of human papillomavirus on nasopharyngeal carcinoma in Japan. Auris Nasus Larynx 2016; S0385-8146:30223-1. 23. Benjamini Y, Hochberg Y. Controlling the false discovery rate: a practical and powerful
approach to multiple testing. J R Stat Soc B 1995; 57, 289–300
24. van der Schaaf A, Xu CJ, van Luijk P, Van't Veld AA, Langendijk JA, Schilstra C. Multivariate modeling of complications with data driven variable selection: guarding against overfitting and effects of data set size. Radiother Oncol 2012;105, 115–21 25. Moons KGM, Altman DG, Reitsma JB, et al. Transparent Reporting of a multivariable
prediction model for Individual Prognosis Or Diagnosis (TRIPOD): Explanation and Elaboration. Ann Intern Med. 2015;162: W1-W73
26. Emerick KS, Leavitt ER, Michaelson JS, Diephuis B, Clark JR, Deschler DG. Initial clinical findings of a mathematical model to predict survival of head and neck cancer. Otolaryngol Head Neck Surg 2013;149:572-78.
27. Rodriguez CP, Adelstein DJ. Survival trends in head and neck cancer: opportunities for improving outcomes. Oncologist 2010;15:921-23.
28. Patel VJ, Chen NW, Resto VA. Racial and ethnic disparities in nasopharyngeal cancer survival in the United States: a SEER study. Otolaryngol Head Neck Surg 2017;156:122-31.
29. Sethi S, Ali-fehmi R, Franceschi S, et al. Characteristics and survival of head and neck cancer by HPV status : a cancer registry-based study. Int J Cancer 2012;131:1179–86. 30. Dayyani F, Etzel CJ, Liu M, Ho C, Lippman SM, Tsao AS. Meta-analysis of the impact of human papillomavirus (HPV) on cancer risk and overall survival in head and neck squamous cell carcinomas (HNSCC). Head Neck Oncol 2010;2-15.
31. Zang J, Li C, Zhao LN, et al. Prognostic model of death and distant metastasis for nasopharyngeal carcinoma patients receiving 3DCRT/IMRT in nonendemic area of China. Medicine (Baltimore) 2016;95:e3794.
32. Chen LL, Nolan ME, Silverstein MJ, et al. The impact of primary tumor size, lymph node status and other prognostic factors on the risk of cancer death. Cancer 2009;115:5071–83.
33. Qin L, Wu F, Lu H, Wei B, Li G, Wang R. Tumor volume predicts survival rate of advanced nasopharyngeal carcinoma treated with concurrent chemoradiotherapy. Otolaryngol Head Neck Surg 2016;155:598-605.
34. Chicklore S, Goh V, Siddique M, Roy A, Marsden PK, Cook GJR. Quantifying tumor heterogeneity in 18F-FDG PET/CT imaging by texture analysis. Eur J Nucl Med Mol Imaging 2013;40:133–40.
35. Tixier F, Le Rest CC, Hatt M, et al. Intratumor heterogeneity characterized by textural features on baseline 18F-FDG PET images predicts response to concomitant radiochemotherapy in esophageal cancer. J Nucl Med 2011;52:369–78.
36. Gerlinger M, Rowan AJ, Horswell S, et al. Intratumor heterogeneity and branched evolution revealed by multiregion sequencing. N Engl J Med 2012;366:883-92. 37. Zhang XC, Xu C, Zhang B, et al. Tumor evolution and intratumor heterogeneity of an
oropharyngeal squamous cell carcinoma revealed by whole-genome sequencing. Neoplasia 2013 ;15:1371-78.
38. Balluff B, Hanselmann M, Heeren RM. Mass Spectrometry Imaging for the Investigation of Intratumor Heterogeneity. Adv Cancer Res 2017;134:201-30. 39. Alizadeh AA, Aranda V, Bardelli A, et al. Toward understanding and exploiting tumor
heterogeneity. Nat Med 2015;21:846-53.
40. Lipinski KA, Barber LJ, Davies MN, Ashenden M, Sottoriva A, Gerlinger M. Cancer evolution and the limits of predictability in precision cancer medicine. Trends Cancer 2016;2:49-63.
41. O'Connor JP, Rose CJ, Waterton JC, Carano RA, Parker GJ, Jackson A. Imaging intratumor heterogeneity: role in therapy response, resistance, and clinical outcome. Clin Cancer Res 2015;21:249-57.
CHAPTER 3
The prognostic value of CT-based
image-biomarkers for head and neck cancer patients
treated with definitive (chemo-)radiation
Published in: Oral Oncology 2019 August;95:178-186.
Tian-Tian Zhaia,*, Johannes A. Langendijka, Lisanne V. van Dijka, Gyorgy B. Halmosc,
Max J.H. Witjesd, Sjoukje F. Oostinge, Walter Noordzijf, Nanna M. Sijtsemaa, Roel J.H.M.
Steenbakkersa a Department of Radiation Oncology, University of Groningen, University Medical
Center Groningen, Groningen, The Netherlands
b Department of Radiation Oncology, Cancer Hospital of Shantou University Medical
College, Shantou, China
c Department of Otorhinolaryngology, Head and Neck Surgery, University of Groningen,
University Medical Center Groningen, Groningen, The Netherlands
d Department of Maxillofacial Surgery, University of Groningen, University Medical
Center Groningen, Groningen, The Netherlands
e Department of Medical Oncology, University of Groningen, University Medical Center
Groningen, Groningen, The Netherlands
f Department of Nuclear Medicine and Molecular Imaging, University of Groningen,
University Medical Center Groningen, Groningen, The Netherlands *Corresponding author
Abstract Objectives
The aim of this study was to investigate whether quantitative CT image-biomarkers (IBMs) can improve the prediction models with only classical prognostic factors for local-control (LC), regional-local-control (RC), distant metastasis-free survival (DMFS) and disease-free survival (DFS) for head and neck cancer (HNC) patients.
Materials and Methods
The cohort included 240 and 204 HNC patients in the training and validation analysis, respectively. Clinical variables were scored prospectively and IBMs of the primary tumor and lymph nodes were extracted from planning CT-images. Clinical, IBM and combined models were created from multivariable Cox proportional-hazard analyses based on clinical features, IBMs, and both for LC, RC, DMFS and DFS.
Results
Clinical variables identified in the multivariable analysis included tumor-site, WHO performance-score, tumor-stage and age. Bounding-box-volume describing the tumor volume and irregular shape, IBM correlation representing radiological heterogeneity, and LN_major-axis-length showing the distance between lymph nodes were included in the IBM models. The performance of IBM LC, RC, DMFS and DFS models (c-index(validated):0.62, 0.80, 0.68 and 0.65) were comparable to that of the clinical models (0.62, 0.76, 0.70 and 0.66). The combined DFS model (0.70) including clinical features and IBMs performed significantly better than the clinical model. Patients stratified with the combined models revealed larger differences between risk groups in the validation cohort than with clinical models for LC, RC and DFS. For DMFS, the differences were similar to the clinical model.
Conclusion
For prediction of HNC treatment outcomes, image-biomarkers performed as good as or slightly better than clinical variables.
3
Introduction
Head and neck squamous cell carcinoma (HNSCC) is primarily managed by surgery and/ or radiotherapy (RT) with or without systemic treatment. At present, the 5-year overall survival rate is around 60% [1]. However, 30%-50% of patients with locally advanced HNSCC still experience treatment failures, predominantly occurring at the site of the primary tumor, followed by regional failures and distant metastases [2]. Risk assessment of local control (LC), regional control (RC), distant metastasis-free survival (DMFS) and disease-free survival (DFS) of HNSCC patients becomes increasingly important to optimise treatment [3-7]. For HNSCC patients, molecular-based factors such as human papilloma-virus (HPV) [3,4] and patient-specific factors such as age and World Health Organization performance-status (WHO PS) have been identified as prognostic clinical factors for LC, RC, DMFS and DFS [5-7]. However, these clinical factors are not sufficient for identifying patients that will benefit most from specific treatment strategies. For this purpose, more detailed information is required, including factors that reflect the characteristics of the whole tumor, such as tumor volume, shape and heterogeneity [8-16].
A wide variety of medical images is generated for diagnostic and staging purposes, such as TNM staging. In current clinical practice, these images are used for guiding treatment decision-making [3]. Image-biomarkers (IBMs) may also be extracted from these medical images, transforming image data into quantitative information that describes intensity, shape and textural characteristics of the whole tumor. IBMs can provide more spatial and textural information about tumor features than TNM staging [8-10]. For patients treated with primary non-surgical modalities, like (chemo)radiotherapy, where only limited pathological information is available, the use of image-biomarkers might improve the prediction of treatment outcomes and improve medical decision-making [9].
IBMs have demonstrated their value to predict treatment outcome and complications for patients with head and neck, lung, breast, pancreatic, and colorectal cancers [9,11-16]. In a previous study, we showed that the quantitative computed-tomography (CT) IBMs were good substitutes for the qualitative clinical variable N-stage, and that they improved the performance of multivariable prediction models for overall survival, compared to models consisting of clinical variables alone [17]. The next challenge is to find IBMs that allow for a better prediction of local-regional failure and distant
metastasis. Based on our previous study, we hypothesized that IBMs would provide similar or better predictive information than clinical variables for LC, RC, DMFS and DFS. The aim of this study was to investigate whether multivariable prediction models for LC, RC, DMFS and DFS, consisting of both clinical variables and IBMs perform better than prediction models with only classical prognostic factors.
Materials and Methods Patient selection and treatment
This was a retrospective analysis in a prospective cohort study, which was composed of 707 consecutive non-surgically treated HNSCC patients. The tumors originated in the oral cavity, oropharynx, nasopharynx, hypopharynx or larynx and were primarily treated with definitive radiotherapy at the University Medical Center Groningen between July 2007 and December 2015. We excluded 202 patients without contrast-enhanced planning CT-scans, 45 patients with metal or motion artifacts in the region of the primary tumor (PT) or positive lymph nodes (LN), and 16 patients with previous neck dissection. Overall 444 patients with standard contrast-enhanced planning CT-scans (Somatom Sensation Open, Siemens, Forchheim, Germany; voxel size: 1.0×1.0×2.0 mm; scan voltage: 120kV; and convolution kernel: B30) were included. Overall 240 patients treated before June 2012 were enrolled in the training cohort and 204 patients treated after June 2012 in the validation cohort. All patients were treated with definitive three-dimensional conformal radiotherapy (3D-CRT), intensity-modulated radiotherapy (IMRT) or volumetric modulated arc therapy (VMAT) to a total dose of 70 Gy with fractions of 2 Gy in 6-7 weeks, with or without chemotherapy or cetuximab. Detailed radiation protocols have been published previously [17,18].
Clinical parameters
Clinical parameters including age, gender, TNM-stage, clinical stage, treatment modality and WHO PS were collected from our prospective data registration program. TNM and clinical stage were defined according to the 7th edition of the American Joint Committee on Cancer Staging Manual [3]. Tumor site was included in the analysis, and tumors originating in the oropharynx were further stratified by HPV status as HPV-positive, HPV-negative and HPV-unknown (Table 1). HPV-status was assessed by p16
3
Fig. 1. The imag
e-biomark
er
s (IBMs) e
xtr
action and analy
sis w
orkflo
immunohistochemistry followed by DNA polymerase chain reaction in cases of p16-positivity in OPC patients. Tumor volume was included in the analysis as a geometric IBM, and not as a clinical parameter.
CT image-biomarkers
An overview of the IBM extraction process and analysis is shown in Fig. 1. The IBMs were extracted using in-house developed Matlab based software (version R2014a; Mathworks, Natick, USA). For a more detailed description of the IBMs, we refer to previous work [11] and Supplementary A. All IBMs were reported complying with th
REMARK guidelines [18] and IBM formulas are in line with the “Image biomarker standardisation initiative” [10].
CT intensity and geometric IBMs
The primary tumor (PT) and pathological lymph node (LN) were delineated on the planning CT-scans by experienced head and neck radiation oncologists. Thirty-six intensity and 40 geometric IBMs were extracted from both the PT and LN. All IBMs from LN were marked as LN_IBMs. The intensity IBMs were obtained from the histogram of the voxel intensities of the delineated structures, e.g. mean represents the average voxel intensity and the skewness quantifies the degree of asymmetry around the mean value. The geometric IBMs, such as volume, bounding-box-volume and major-axis-length, were extracted from the three-dimensional (3D) contoured structures. The LN_IBMs from patients without lymph node metastasis were defined as 0.
CT textural IBMs
Forty-four textural CT IBMs describing the radiological heterogeneity of the PT tissue were derived from three different matrices: the gray level co-occurrence matrix (GLCM) [20], the gray level run-length matrix (GLRLM) [21] and the gray level size-zone matrix (GLSZM) [22]. Those matrices provide a statistical view of image texture based on the relationship between neighbouring pixels. GLCM IBMs describe the number of voxel transitions of certain gray levels, e.g. the IBM correlation is larger in case of larger areas of similar gray levels. GLRLM IBMs assess the number of directional gray level repetition, e.g. low run length non-uniformity means consecutive voxels with the same gray level are distributed homogeneously. The GLCM and GLRLM IBMs were computed from each 3D directional matrix and averaged over 13 directions. GLSZM quantifies the volumetric
