Individual nodule tracking in micro-CT images of a longitudinal lung cancer mouse model
Graphical abstract
Introduction
Lung cancer is the leading cause of cancer-related deaths worldwide (Herbst et al., 2008). Its low – 15.9% – 5-year survival rate (National Cancer Institute, 2013), relates to the fact that lung cancer usually becomes symptomatic at advanced stages, when it is likely to have metastasized to distant organs or it is too widespread in the lung as to be surgically removed or efficiently treated by chemotherapy or radiotherapy. Early detection and an accurate differential diagnosis of suspicious lung lesions are critical to reduce mortality. In clinical settings, computed tomography (CT) has been proven as a useful, cost effective tool for non-invasive detection of early stage lesions (Henschke et al., 2006, National Lung Screening Trial Research Team, 2011) and to evaluate tumor progression during clinical treatment (Meyer et al., 2009).
The biology of the disease and the effect of novel therapeutic drugs can be efficiently studied using mouse models. In one such model, lung lesions are induced by intraperitoneal injection of urethane in A/J mice, producing small and preneoplastic lesions, whose incidence increased with progression from lung hyperplasias to adenomas and to carcinomas (Cazorla et al., 1998). Tumor growth in mouse models is observed much faster than in humans, which allows for relatively short preclinical studies. Tumor size and growth are common predictors of tumor malignancy and of the response to treatment in human (Therasse et al., 2000). Therefore, accurate nodule growth quantification is extremely important also in preclinical research to assess therapeutic response (Chen et al., 2012, Nathan et al., 1962, Weiss, 1971). X-ray computer tomography is a technique especially well suited to image the lungs. Tumor growth measured on micro-CT images has been found to correlate well with histology (Hori et al., 2008), tumor biopsy (Cody et al., 2005), or bioluminescence-based measurements (Fushiki et al., 2009). Being a non-invasive imaging technique, micro-CT can be used to monitor tumor growth in longitudinal studies, significantly reducing the number of animals required to obtain valid results. Disadvantages of this method are the amount of radiation delivered to the animals in each scan –which may interfere with the process being studied- and the relative low image quality caused by motion-related artifacts. To increase the image quality at a reasonable radiation dose, we and others have combined respiratory gating with animal intubation and artificial ventilation. This way we can obtain high-resolution micro-CT images of an almost motion-free lung during controlled breath-holds. Using this protocol, we have quantitatively characterized longitudinally several models of lung disease (Artaechevarria et al., 2011, Artaechevarria et al., 2010).
Manually segmenting and matching high number of tumors is extremely time consuming. Therefore, longitudinal nodule progression has been assessed by surrogate tumor and vessel burden (Haines et al., 2009, Rodt et al., 2012), semi-automatically tracking diameter changes in low tumor burden cases (Hori et al., 2008), over short-term period (Namati et al., 2010) and interactively using commercial tool (Cody et al., 2005). In this study, we have developed an automated, quantitative framework to volumetrically segment and track individual nodules. The techniques employed are inspired by methods recently developed in the clinical settings.
Several computer-aided detection (CAD) systems (Van Ginneken et al., 2010) have been recently developed, some of them with automatic pulmonary nodule matching modules (Hong et al., 2008, Koo et al., 2012). These systems facilitate the manual process of tracking suspicious nodules. The use of these systems in human studies concludes that nodule matching greatly improves the efficiency of CT interpretation (Koo et al., 2012), but it is still an unresolved problem in patients with large number of nodules (Lee et al., 2007, Tao et al., 2009).
In clinical settings, low-cost chest X-ray has been the most common radiological procedure for lung nodule detection and assessment. The practice of measuring nodule linearly on a plane continued onto other modalities such as CT. In the interest of maximizing reproducibility, lung nodule size was typically assessed using simple linear measurements such as RECIST (Response Evaluation Criteria in Solid Tumors) (Schwartz et al., 2000, Therasse et al., 2000), based on the measurement of the maximum diameter of a nodule from a single slice. These criteria are practical and easy to calculate but assume that tumors grow isometrically, and the corresponding quantification has proven poorly reproducible (Erasmus et al., 2003, Picozzi et al., 2006, Revel et al., 2004). More accurate metrics such as tumor volume became feasible with the advent of helical CT, which allows acquiring entire anatomical volumes – such as thorax – in a single breath hold. However, manually segmenting nodules in 3D image stacks, often surrounded by other anatomical structures, is highly time-consuming and leads to high inter-observer variability (Schilham et al., 2006, Wormanns et al., 2000). There are also automatic and semi-automatic nodule segmentation software packages, but their accuracy is highly dependent on the acquisition protocol, reconstruction parameters, and nodule characteristics (Gavrielides et al., 2009), leading to poorly reproducible results. Taking all these factors into account, various groups have proposed automatic nodule volume estimation using either Gaussian fitting techniques (Okada et al., 2005) or explicit nodule segmentation techniques (Dehmeshki et al., 2008, Ko et al., 2003, Kostis et al., 2003, Kuhnigk et al., 2006, Reeves et al., 2006).
Tumor size and doubling rate measured from CT images are features commonly used to determine follow-up periods and further disease management. In the context of preclinical longitudinal studies, to this day, malignancy is defined by tumor phenotype, which is determined after histological characterization of the resected tissue upon sacrifice of the affected mice. Tumor size quantification is currently based on the measurement of the diameter of the outermost nodules visible on the lung surface at the moment of lung dissection. Alternatively nodule average size can be assessed by the estimation of the tumor area in 2D histological sections. Any of these two ways of measuring tumor size remains a predictor for the response to treatment. A technique that performs correct tracking of nodules in living animals and quantitative assessment of the differences in tumor growth upon different treatment regimens, becomes an extremely valuable tool to study the efficacy of new drugs.
In this paper, we present a novel, fully automatic, open-source framework that integrates all the steps needed for the longitudinal characterization of individual nodules in a urethane-based mouse model of lung cancer. This is a particularly challenging task, due to the high-incidence of the tumors, and the length of the study, which accounts for tumor numbers and densities that are far beyond those found in human cancers. Therefore, we have designed a framework to volumetrically segment the nodules from interactively planted seeds, and track their evolution in time after proper registration of the lungs. Our results have been validated and calibrated using a small animal scale phantom. The novelty of the proposed framework has been applied to the characterization of individual nodule’s growth and doubling rate in a long-term urethane-induced model of lung cancer, correlating these parameters with tumor phenotype. This information could be essential for researchers investigating phenotypic growth dynamics of tumors and may be instrumental in the future to study the response to pharmacological treatments.
The manuscript is organized as follows: In Section 2, we describe the animal work, histology sample preparation and micro-CT imaging protocol. In Section 3, we describe the algorithm developed to track and segment the nodules. In Section 4, we describe the validation and volume correction process. In Section 5, we present the experimental results. In Section 6, we discuss the challenges and future extensions.
Section snippets
Animal preparation
The experimental protocol is shown in Fig. 1. At 10 weeks of age, fourteen male A/J mice (Harlan UK Ltd., Oxon, UK) with mean weight 20.3 ± 2.4 g, received a single intraperitoneal injection of urethane (100 mg/100 g, Urethane U2500 Sigma, St Louis, Missouri). At 8, 22, and 37 weeks after urethane administration, all mice underwent respiratory-gated micro-CT scans. The institutional ethics committee approved the proposed animal experiment.
Micro-CT imaging
Before the scan, mice were anaesthetized, endotracheally
Micro-CT image analysis
Fig. 2 summarizes the main steps of the nodule progression estimation pipeline. The inputs for the pipeline are the micro-CT lung images and the coordinates of all interactively detected nodules. Nodule matching requires first the registration of the longitudinal lung images. Nodule matching combined with nodule segmentation provides quantitative assessment of tumor progression (i.e., growth rate, doubling rate) for each individual nodule. The accuracy of the nodule segmentation is validated
Validation
As tumors do not expand isometrically, one-dimensional measurements are not appropriate to quantify nodule growth (Namati et al., 2010). The volumetric quantification of tumors minimizes variability and decreases measurement error (Yankelevitz et al., 2000), but its accuracy is highly dependent on the shape, dose, reconstruction algorithm, nodule size, and partial volume averaging effects (Gavrielides et al., 2010, Ko et al., 2003, Yankelevitz et al., 2000). This requires proper validation of
Experimental results
Experimental results are presented here for each stage of the tracking and volumetric measurement pipeline. An example of a juxtapleural nodule tracked in time shown in 3-D reconstruction, along with the whole lung volume, is illustrated in Fig. 7.
Discussion and conclusions
Automatic and quantitative longitudinal characterization of lung nodule progression in small animals is a complex task. In this paper, we have presented a turnkey solution that combines longitudinal imaging with nodule matching and volumetric nodule segmentation resulting in a powerful tool for preclinical research. Tracking nodule growth individually permits researchers to discover individual nodule’s growth rate and correlate this to the phenotype.
We have achieved high matching accuracy in
Acknowledgment
This project was partially funded by the “UTE Project CIMA”; the Spanish Ministry of Health under Grants PI070751, RTICC RD12/0036/0040 and RD12/0036/0062 subprogram for unique strategic projects of the Spanish Science and Innovation Ministry under Grants MICINN PSE SINBAD and PSS 0100000-2008-2; the program for non-fundamentally directed research grants of the Spanish Science and Innovation Ministry under Grants MCYT TEC2005-04732, MICINN DPI2009-14115-C03-03 and MINECO DPI2012-38090-C03-02.
In
Reference (50)
- et al.
Comparing and combining algorithms for computer-aided detection of pulmonary nodules in computed tomography scans: the ANODE09 study
Med. Image Anal.
(2010) - et al.
A quantitative volumetric micro-computed tomography method to analyze lung tumors in genetically engineered mouse models
Neoplasia
(2009) - et al.
Automatic lung nodule matching on sequential CT images
Comput. Biol. Med.
(2008) - et al.
Quantitative imaging to assess tumor response to therapy: common themes of measurement, truth data, and error sources
Translational Oncol.
(2009) - et al.
A computer-aided diagnosis system for detection of lung nodules in chest radiographs with an evaluation on a public database
Med. Image Anal.
(2006) - et al.
Computerized scheme for determination of the likelihood measure of malignancy for pulmonary nodules on low-dose CT images
Med. Phys.
(2003) - et al.
Automated computerized scheme for distinction between benign and malignant solitary pulmonary nodules on chest images
Med. Phys.
(2002) - et al.
Evaluation of micro-CT for emphysema assessment in mice: comparison with non-radiological techniques
Eur. Radiol.
(2011) - et al.
Longitudinal study of a mouse model of chronic pulmonary inflammation using breath hold gated micro-CT
Eur. Radiol.
(2010) - et al.
Ki-ras gene mutations and absence of p53 gene mutations in spontaneous and urethane-induced early lung lesions in CBA/J mice
Mol. Carcinog.
(1998)
A murine lung cancer co-clinical trial identifies genetic modifiers of therapeutic response
Nature
Murine lung tumor measurement using respiratory-gated micro-computed tomography
Invest. Radiol.
Segmentation of pulmonary nodules in thoracic CT scans: a region growing approach
IEEE Trans. Med. Imaging
Measures of the amount of ecologic association between species
Ecology
Interobserver and intraobserver variability in measurement of non-small-cell carcinoma lung lesions: implications for assessment of tumor response
J. Clin. Oncol.
Investigating the effect of longitudinal micro-CT imaging on tumour growth in mice
Phys. Med. Biol.
Quantification of mouse pulmonary cancer models by microcomputed tomography imaging
Cancer Sci.
A resource for the assessment of lung nodule size estimation methods: database of thoracic CT scans of an anthropomorphic phantom
Opt. Express.
Noncalcified lung nodules: volumetric assessment with thoracic CT
Radiology
Survival of patients with stage I lung cancer detected on CT screening
New Engl. J. Med.
Lung cancer
New Engl. J. Med.
Periodic analysis of urethane-induced pulmonary tumors in living A/J mice by respiration-gated X-ray microcomputed tomography
Cancer Sci.
Automatic lung segmentation for accurate quantitation of volumetric X-Ray CT images
IEEE Trans. Med. Imaging
Cooperation of the agonistic DR5 antibody apomab with chemotherapy to inhibit orthotopic lung tumor growth and improve survival
Clin. Cancer Res.
Elastix: a toolbox for intensity-based medical image registration
IEEE Trans. Med. Imaging
Cited by (21)
Automated segmentation of lungs and lung tumors in mouse micro-CT scans
2022, iScienceCitation Excerpt :Published preclinical lung and lung tumor segmentation methods utilize a wide range of techniques, including rules-based segmentation, a mix of rules-based and machine learning methods, machine learning methods with manual intervention and fully automated deep learning approaches. For preclinical lung segmentation, multiple semi-automated methods have been proposed.6,7,8 However, they rely on a combination of manual input and rules-based segmentation, which can be time-consuming and potentially biased.
Principles of Micro X-ray Computed Tomography
2021, Molecular Imaging: Principles and PracticeA Novel Method for Quantifying Total Thoracic Tumor Burden in Mice
2018, Neoplasia (United States)Citation Excerpt :In addition, these methods measure only the largest tumors and ignore small tumors that, in aggregate, may contribute significantly to the total tumor burden. Other methods that have been used to quantify tumor burden include tumor nodule segmentation [18]; segmentation of the aerated lung volume with respiratory gating [19]; manual segmentation of the chest space [16]; tracking of individual nodules [20]; modeling of tumors as ellipsoids [21]; and volumetric measurement of the combined tumor and vasculature from a threshold-based region growing algorithm with manual and semi-automated segmentation [22]. However, these methods are labor-intensive, require specific skill sets in radiology, and do not take all thoracic metastases into consideration.
Early Lung Adenocarcinoma in Mice: Micro-Computed Tomography Manifestations and Correlation with Pathology
2017, Translational OncologyCitation Excerpt :The results indicated that micro-CT could clearly demonstrate the morphological characteristics of the tumor and tumor-bronchus relationship. Previous studies [17–23] used micro-CT to investigate the vessel supply, angiogenesis and therapeutic response of mouse lung cancer, as well as airway changes in a mouse asthma model. However, it could only demonstrate the presence of lung tumors and could not clearly show the internal structure of tumors.
Quantification of tumor burden in a genetically engineered mouse model of lung cancer by micro-CT and automated analysis
2015, Translational OncologyCitation Excerpt :This typically makes an analysis method time-consuming, vulnerable to intra-reader and inter-reader variability and potential reader bias, and generally requires identification of the tumor tissue by the reader. Semi-automated methods [16–19] have been reported for segmenting and tracking individual lung nodules, but there remains a need for a time-efficient method for quantifying lung tumor burden in a robust manner. Fully automating the image analysis would offer significant benefits in drug development studies where objective and reproducible quantification of the tumor burden is crucial, and high throughput is necessary for gaining statistical power.
- 1
Present address: Sunnybrook Health Sciences Centre, Toronto, Ontario, Canada.