A review on segmentation of positron emission tomography images

https://doi.org/10.1016/j.compbiomed.2014.04.014Get rights and content

Abstract

Positron Emission Tomography (PET), a non-invasive functional imaging method at the molecular level, images the distribution of biologically targeted radiotracers with high sensitivity. PET imaging provides detailed quantitative information about many diseases and is often used to evaluate inflammation, infection, and cancer by detecting emitted photons from a radiotracer localized to abnormal cells. In order to differentiate abnormal tissue from surrounding areas in PET images, image segmentation methods play a vital role; therefore, accurate image segmentation is often necessary for proper disease detection, diagnosis, treatment planning, and follow-ups. In this review paper, we present state-of-the-art PET image segmentation methods, as well as the recent advances in image segmentation techniques. In order to make this manuscript self-contained, we also briefly explain the fundamentals of PET imaging, the challenges of diagnostic PET image analysis, and the effects of these challenges on the segmentation results.

Introduction

Structural imaging techniques such as computed tomography (CT) and magnetic resonance imaging (MRI) are widely utilized in clinical practice to examine anatomical abnormalities caused by disease. The three dimensional (3D) images produced by these techniques usually give detailed structural information about one׳s anatomy that can be used for diagnostic and therapeutic purposes [1]. However, structural imaging is not well suited for pathology detection applications where cellular activity is more significant than anatomical features [2]. The need for functional characterization leads researchers to develop PET scanners, which provide molecular information on the biology of many diseases. When combined with CT or MRI, utilizing both functional (PET) and structural information leads to a higher sensitivity and specificity than is achievable using either modality alone. Although the sensitivity of PET scans is usually much higher than conventional structural images, anatomical information from another modality (CT or MRI) is still needed to properly interpret and localize the radiotracer uptake and the PET images are somewhat limited due to low resolution. Hence, there is a frequent need for assessing functional images together with structural images in order to localize functional abnormalities and distinguish them from normal uptake of PET radiotracers, which tend to normally accumulate in the brain, heart, liver, kidneys, etc. [3], [4], [5]. PET-CT imaging and more recently MRI-PET have been used to combine complementary diagnostic information from different imaging modalities into a single imaging device, removing the need for registration [6]. Using these scanning techniques, disease can be labeled and identified such that an earlier diagnosis with more accurate staging for patients may potentially be delivered [7].

Some of the statistics for the use of PET imaging in the U.S. is summarized in Fig. 1(a). Over 1,700,000 clinical PET and PET-CT studies were reported nation-wide for 2011 only. Compared to single PET imaging, the use of PET-CT is relatively higher and continuing to increase. PET imaging is mostly used for (i) diagnosis, (ii) staging, (iii) treatment planning, and (iv) therapy follow-up, in different fields of medicine such as (1) oncology, (2) cardiology, and (3) neurology (Fig. 1(b)). PET is widely used in staging and follow-up therapy in oncology applications (Fig. 1(c)). For instance, radiation therapy, as a common cancer treatment in oncology, aims to target the boundary and volume of abnormal tissue and irradiates the targeted area with a high dosage of radiation, intending to eliminate all cancerous cells. In practice, the determination of this boundary (i.e., delineation) should be kept as small as possible to minimize damage to healthy tissue, but the boundary must ensure the inclusion of the entire extent of the diseased tissue [2]. PET is also used in cardiac applications such as quantifying blood flow to the heart muscle and quantifying the effects of a myocardial infarction [8]. More recently, PET has been used for imaging inflammation and infection in the lungs [9] with 18F-FDG because this glucose analog localizes to activated and proliferated inflammatory cells. The new norm in clinical practice is acquiring PET-CT images instead of a single PET scan to take advantage of the functional and structural information jointly.

In pre-clinical and clinical applications, physicians and researchers use PET imaging to determine functional characterization of the tissues. Owing to this, clinical trials are now placing a greater reliance on imaging to provide objective measures in before, during, and after treatment processes. The functional morphology (the area, volume, geometry, texture, etc.) as well as activity measures – such as standardized uptake value (SUV) of the tissues–are of particular interest in these processes. Accurately determining quantitative measures enables physicians to assess changes in lesion biology during and after treatment; hence, it allows physicians to better evaluate tumor perfusion, permeability, blood volume, and response to therapy. Among these measures, functional volume (i.e., the volume of high uptake regions) has been proven useful for the definition of target volumes [11]. Therefore, an accurate image segmentation method, other than the conventional region of interest (ROI) analysis, is often needed for diagnostic or prognostic assessment. This functional characterization has a higher potential for proper assessment due to recent advances in PET imaging. Indeed, this higher potential has renewed interest in developing much more accurate (even globally optimal) segmentation methods to turn hybrid imaging systems into diagnostic tools [11]. Specifically, after the adoption of multi-modal imaging systems (i.e., PET-CT, MRI-PET), optimal approaches for precise segmentation and quantification of metabolic activities were crucial.

For the literature search, we used Pubmed™, IEEEXplore™, Google Scholar™, and ScienceDirect™ and listed all the relevant articles from 1983 to March 2013. Our search also included the methods specifically developed for MRI and CT for comparison (Fig. 2). The number of publications for PET image segmentation is further separated by publication type (conference, journal, and total) in Fig. 2(a). As a reflection of the improvements in multi-modality imaging technology (PET-CT and MRI-PET), there was a dramatic increase in the number of publications in 2008 and 2011. For a comparison, Fig. 2(b) shows how the number of publications in PET image segmentation methods compare to the number of CT and MRI based segmentation methods in the literature. Notably, the number of PET image segmentation publications has always been lower than both CT and MRI and was significantly lower before 2007. Fig. 2(c) gives the breakdown on the number of publications for segmentation techniques for PET images from 1984 to 2013. We also noted that only 2% of the articles were review papers and almost half of the total articles are journal papers (42% journal publications and 54% conference publications). For the last 6 years, Fig. 2(d) shows a snapshot of publication types from 2007 to 2013, during which the dramatic increase of PET image segmentation publications was observed. It appears that the growing interest in PET and hybrid imaging will further accelerate the methods for segmentation and quantification of lesions.

In this work, we systematically review state-of-the-art image segmentation methods for PET scans of body images, as well as the recent advances in PET image segmentation techniques. In order to have a complete review on the topic, the necessary knowledge of the physical principles of PET imaging are also given, along with the source of the challenges for segmentation inherent to PET images in Section 2. The state-of-the-art segmentation methods for PET images, their comparison, and recently developed advanced PET image segmentation methods are extensively analyzed in later sections, and the methods are divided into the following groups for clarity: manual segmentation and ground truth reconstruction (Section 3), thresholding-based (Section 4), stochastic and learning-based (Section 5), region-based (Section 6), boundary-based (Section 7), and multi-modality methods (Section 8). These categories are shown in Fig. 3. Due to the large number of segmentation methods, we have categorized the state-of-the-art methods into intuitive groups for easier comprehension and better contrasting of the methods. Finally, discussions are made in Section 9, followed by conclusions in Section 10.

Section snippets

Radiotracers

The basic concept of PET is to label a radio-pharmaceutical compound with a biologically active ligand to form a radiotracer and inject it intravenously into a patient. The PET scanner then measures the distribution and concentration of the radiotracer accumulation throughout the patient׳s body as a function of time [12]. To do this, PET utilizes positron emitting radioisotopes as molecular probes so the biochemical process can be measured through imaging in vivo [13]. There have been many

Ground truth construction

An overview of the categories that PET image segmentation methods are classified into is given in Fig. 3. Before introducing the various PET image segmentation methods as summarized in Fig. 3, it is useful and necessary to know the standard ways of evaluating the accuracy of segmentation for proper comparison. In order to evaluate an image segmentation algorithm, the true boundary of the object of interest should be identified. Unfortunately, there is no ground truth available if

Thresholding-based methods

Thresholding is a simple, intuitive, and popular image segmentation technique that converts a gray-level image into a binary image by defining all voxels greater than some value to be foreground and all other voxels are considered as background [76]. The thresholding-based PET image segmentation methods utilize the probability of intensities, usually by using the histogram of the image. An intuitive view on this process is that the objects of interest in the PET image, usually referred to as

Stochastic and learning-based methods

Stochastic methods exploit differences between uptake regions and surrounding tissues statistically. Learning-based methods, similarly, use pattern recognition techniques to statistically estimate dependencies in the data. Since there are strong similarities between learning-based methods and stochastic methods, in this section we introduce the core concepts of the both groups together (Fig. 7).

Region-based segmentation methods

Another distinct type of PET segmentation technique is region-based segmentation methods where the homogeneity of the image is the main consideration for determining object boundaries. While it is true that the region-based segmentation methods also utilize the intensities of the image, they are much more concerned about the local distribution (homogeneity) of the intensities on the image. The region-based methods are mainly divided into two subgroups when considering PET images: Region Growing

Boundary-based methods

Instead of using the statistics of the entire image or the homogeneity of the image for segmentation, boundary-based segmentation methods were designed to locate and identify the boundaries of the objects in PET images. However, locating the boundaries of the objects in PET images is challenging due to the low resolution and noise of the PET images. The boundary-based methods can be categorized into two subgroups, Level Set/Active Contours and Gradient Based methods, as shown in Fig. 11.

Joint segmentation methods

Image fusion involves combining two or more images of differing modalities to create a composite that contains complementary information from the inputs. Before PET-CT and MRI-PET hybrid scanners were developed, image registration techniques were being used to align images. It is evident that fused images are more suitable for visual perception, particularly for radiologists as analyzing fused images reduces uncertainty and minimizes redundancy in the output while maximizing relevant

Discussion

This review gives an overview of the current image segmentation techniques for PET images. The similarities, main ideas, assumptions, and quantitative comparisons between the many PET segmentation methods have been outlined to give researchers and clinicians an idea of which method is applicable for most situations. When choosing which method is appropriate for a specific quantification application, it should be noted that there are some important details outside the scope of this manuscript.

Conclusions

PET imaging provides quantitative functional information on diseases, and image segmentation is of great importance for extracting this information. In this paper, we presented the state-of-the-art image segmentation methods that are commonly used for PET imaging, as well as the recent advances in techniques applicable to PET, PET-CT, and MRI-PET images. We investigated different segmentation methods in detail; results were listed and compared throughout this review. Given the vast number and

Conflict of interest statement

None declared.

Acknowledgments

This research is supported by the Center for Infectious Disease Imaging (CIDI), the Intramural Program of the National Institutes of Allergy and Infectious Diseases (NIAID), and the National Institutes of Bio-imaging and Bioengineering (NIBIB) at the National Institutes of Health (NIH). We thank Dr. Sanjay Jain for kindly providing the rabbit TB images.

References (185)

  • U. Bagci et al.

    Joint segmentation of functional and anatomical imagesapplications in quantification of lesions from PET, PET-CT, MRI-PET, and MRI-PET-CT images

    Med. Image Anal.

    (2013)
  • J. Fox et al.

    Does registration of PET and planning CT images decrease interobserver and intraobserver variation in delineating tumor volumes for non-small-cell lung cancer?

    Int. J. Radiat. Oncol. Biol. Phys.

    (2005)
  • C. Caldwell et al.

    Observer variation in contouring gross tumor volume in patients with poorly defined non-small-cell lung tumors on CTthe impact of 18 FDG-hybrid PET fusion

    Int. J. Radiat. Oncol. Biol. Phys.

    (2001)
  • R. Steenbakkers et al.

    Reduction of observer variation using matched CT-PET for lung cancer delineationa three-dimensional analysis

    Int. J. Radiat. Oncol. Biol. Phys.

    (2006)
  • C. Fiorino et al.

    Intra-and inter-observer variability in contouring prostate and seminal vesiclesimplications for conformal treatment planning

    Radiother. Oncol.

    (1998)
  • P. Giraud et al.

    Conformal radiotherapy for lung cancerdifferent delineation of the gross tumor volume (GTV) by radiologists and radiation oncologists

    Radiother. Oncol.

    (2002)
  • S. Breen et al.

    Intraobserver and interobserver variability in GTV delineation on FDG-PET-CT images of head and neck cancers

    Int. J. Radiat. Oncol. Biol. Phys.

    (2007)
  • H. Vorwerk et al.

    The delineation of target volumes for radiotherapy of lung cancer patients

    Radiother. Oncol.

    (2009)
  • N. Webb et al.

    Reliability coefficients and generalizability theory

    Handbook of Statistics

    (2006)
  • K.H. Zou et al.

    Statistical validation of image segmentation quality based on a spatial overlap index sup 1 supscientific reports

    Acad. Radiol.

    (2004)
  • D. Schinagl et al.

    Comparison of five segmentation tools for 18f-fluoro-deoxy-glucose–positron emission tomography-based target volume definition in head and neck cancer

    Int. J. Radiat. Oncol. Biol. Phys.

    (2007)
  • C. Caldwell et al.

    Can PET provide the 3d extent of tumor motion for individualized internal target volumes? A phantom study of the limitations of CT and the promise of PET

    Int. J. Radiat. Oncol. Biol. Phys.

    (2003)
  • A. Paulino et al.

    Comparison of CT-and FDG-PET-defined gross tumor volume in intensity-modulated radiotherapy for head-and-neck cancer

    Int. J. Radiat. Oncol. Biol. Phys.

    (2005)
  • E. Deniaud-Alexandre et al.

    Impact of computed tomography and 18f-deoxyglucose coincidence detection emission tomography image fusion for optimization of conformal radiotherapy in non-small-cell lung cancer

    Int. J. Radiat. Oncol. Biol. Phys.

    (2005)
  • R. Hong et al.

    Correlation of PET standard uptake value and ct window-level thresholds for target delineation in CT-based radiation treatment planning

    Int. J. Radiat. Oncol. Biol. Phys.

    (2007)
  • A. Van Baardwijk et al.

    Pet-ct-based auto-contouring in non-small-cell lung cancer correlates with pathology and reduces interobserver variability in the delineation of the primary tumor and involved nodal volumes

    Int. J. Radiat. Oncol. Biol. Phys.

    (2007)
  • W. Yu et al.

    Gtv spatial conformity between different delineation methods by 18 FDG PET/CT and pathology in esophageal cancer

    Radiother. Oncol.

    (2009)
  • M. Wanet et al.

    Gradient-based delineation of the primary GTV on FDG-PET in non-small cell lung cancera comparison with threshold-based approaches, CT and surgical specimens

    Radiother. Oncol.

    (2011)
  • Q. Black et al.

    Defining a radiotherapy target with positron emission tomography

    Int. J. Radiat. Oncol. Biol. Phys.

    (2004)
  • J. Davis et al.

    Assessment of 18f PET signals for automatic target volume definition in radiotherapy treatment planning

    Radiother. Oncol.

    (2006)
  • T. Seute et al.

    Detection of brain metastases from small cell lung cancer

    Cancer

    (2008)
  • S. Basu et al.

    Fundamentals of PET and PET/CT imaging

    Ann. NY Acad. Sci.

    (2011)
  • D. Lardinois et al.

    Staging of non-small-cell lung cancer with integrated positron-emission tomography and computed tomography

    New Engl. J. Med.

    (2003)
  • L. Kostakoglu et al.

    Clinical role of FDG PET in evaluation of cancer patients

    Radiographics

    (2003)
  • M. Judenhofer

    Simultaneous PET-MRIa new approach for functional and morphological imaging

    Nat. Med.

    (2008)
  • D. Evanko

    Two pictures are better than one

    Nat. Methods

    (2008)
  • P. Kaufmann et al.

    Myocardial blood flow measurement by PETtechnical aspects and clinical applications

    J. Nucl. Med.

    (2005)
  • L.S. Zhao, B., L.H. Schwartz, Imaging surrogates of tumor response to therapy: anatomic and functional biomarkers, J....
  • I.M.I. Division, Pet Market Summary Report, Technical Report,...
  • V. Gregoire et al.

    Pet based treatment planning in radiotherapya new standard?

    J. Nucl. Med.

    (2007)
  • J. Votaw

    The aapm/rsna physics tutorial for residents. Physics of PET

    Radiographics

    (1995)
  • S. Basu et al.

    Quantitative techniques in PET-CT imaging

    Curr. Med. Imaging Rev.

    (2011)
  • M. Evelina et al.

    Positron emission tomography (PET) radiotracers in oncology–utility of 18f-fluoro-deoxy-glucose (FDG)-PET in the management of patients with non-small-cell lung cancer (NSCLC)

    J. Exp. Clin. Cancer Res.

    (2008)
  • G. Kramer-Marek et al.

    Pet of her2-positive pulmonary metastases with 18f-zher2342 affibody in a murine model of breast cancer: comparison with 18f-FDG

    J. Nucl. Med.

    (2012)
  • R. Wahl et al.

    From recist to percistevolving considerations for PET response criteria in solid tumors

    J. Nucl. Med.

    (2009)
  • V. Lowe et al.

    Semiquantitative and visual analysis of FDG-PET images in pulmonary abnormalities

    J. Nucl. Med.

    (1994)
  • M. Lodge et al.

    Noise considerations for PET quantification using maximum and peak standardized uptake value

    J. Nucl. Med.

    (2012)
  • C. Kim et al.

    Standardized uptake values of FDGbody surface area correction is preferable to body weight correction

    J. Nucl. Med.

    (1994)
  • M. Kelly, SUV: advancing comparability and accuracy, White Paper...
  • K. Zasadny et al.

    Standardized uptake values of normal tissues at PET with 2-[fluorine-18]-fluoro-2-deoxy-d-glucosevariations with body weight and a method for correction

    Radiology

    (1993)
  • Cited by (318)

    • Artificial intelligence in skeletal metastasis imaging

      2024, Computational and Structural Biotechnology Journal
    • What You See Ain't Necessarily What You Got

      2024, International Journal of Radiation Oncology Biology Physics
    View all citing articles on Scopus
    View full text