Segmentation of anatomical structures in chest radiographs using supervised methods: a comparative study on a public database
Introduction
A large amount of literature in the medical image analysis research community is devoted to the topic of segmentation. Many methods have been developed and tested on a wide range of applications. Despite these efforts, or perhaps because of the large number of algorithms that have been proposed, it remains very difficult for a system designer to decide which approach is best suited for a particular segmentation task. Fortunately, there is a growing awareness in the medical image research community that evaluation and performance characterization of segmentation methods is a critical issue (Jannin et al., 2002, Bowyer et al., 2001). Such evaluations are greatly facilitated by the availability of public image databases with manual annotations on which researchers can test and compare different algorithms. For this study, we have annotated a public database, and have made the manual segmentations available (Image Sciences Institute Research Databases).
We compare three methods for segmenting five important anatomical structures in the single most acquired medical image: the standard posterior–anterior (PA) chest radiograph. To this end, these structures – the lung fields, the heart, and the clavicles – have been segmented manually by two observers independently in 247 radiographs from the publicly available JSRT (Japanese Society of Thoracic Radiology) database (Shiraishi et al., 2000). The fact that each object has been manually segmented twice allows one to use one manual segmentation as gold standard and compare the performance of automatic methods with that of an independent human observer. The web site of the annotated JSRT database (Image Sciences Institute Research Databases) allows other researchers to upload the results of other segmentation algorithms applied to the database and we invite the medical image analysis research community to do so.
Accurate segmentation of anatomical structures in chest radiographs is essential for many analysis tasks considered in computer-aided diagnosis. These include various size measurements, the determination of the presence of pulmonary nodules or signs of interstitial lung disease. Knowledge about the location of the clavicles can be used to reduce false positive findings or to detect lesions hiding ‘behind a clavicle’ more reliably.
The methods considered here are active shape models (ASM) (Cootes et al., 1995, Cootes and Taylor, 2001), active appearance models (AAM) (Cootes et al., 2001) and pixel classification (PC). ASM is a popular segmentation method, with many internal parameters. We consider how to tune these parameters. AAM has recently found widespread application in medical image segmentation. In this work we use an implementation available in the public domain (Stegmann et al., 2003) and compare the standard AAM scheme with an extension in which the surroundings of objects are modeled as well. PC is a classical segmentation method, but the basic concept is so general that it can be implemented in many different ways. We propose an implementation in which both position and local image derivatives are used as input features and show how a multi-resolution implementation and an approximate k-nearest neighbor classifier lead to a relatively fast scheme that yields accurate segmentations. Finally, we also consider three hybrid approaches. The first one fuses the results of the best performing ASM, AAM and PC scheme by majority voting. The other hybrid schemes uses a “tissue map” produced from the probability output of the PC scheme as input for the ASM and AAM method, respectively.
Each of the methods examined here is supervised. This means that example images with the desired output need to be supplied for training. This makes the methods versatile; by supplying different training images and annotations, each method can be applied to many different segmentation tasks, including the ones investigated here. This is in contrast to rule-based schemes that are specifically designed to handle one segmentation task.
The article is organized as follows. Section 2 briefly reviews previous work on segmentation of lung fields, heart and clavicles in chest radiographs. Section 3 describes the data. The segmentation methods are presented in Section 4. Section 5 presents the results, followed by a discussion in Section 6. Section 7 concludes.
Section snippets
Previous work
Segmentation of lung fields in PA chest radiographs has received considerable attention in the literature. Rule-based schemes have been proposed by Li et al. (2001), Armato et al. (1998), Xu and Doi, 1995, Xu and Doi, 1996, Duryea and Boone (1995), Pietka (1994), and Brown et al. (1998). Lung segmentation by pixel classification using neural networks has been investigated by McNitt-Gray et al. (1995), and Tsujii et al. (1998). Vittitoe et al. (1998) developed a pixel classifier for the
Image data
The chest radiographs are taken from the JSRT database (Shiraishi et al., 2000). This is a publicly available database with 247 PA chest radiographs collected from 13 institutions in Japan and one in the United States. The images were scanned from films to a size of 2048 × 2048 pixels, a spatial resolution of 0.175 mm/pixel and 12 bit gray levels. 154 images contain exactly one pulmonary lung nodule each; the other 93 images contain no lung nodules.
Object delineation
Each object has been delineated by clicking
Active shape model segmentation
The following is a brief description of the ASM segmentation algorithm. The purpose is mainly to point out the free parameters in the scheme; the specific values for these parameters are listed in Table 1. Cootes et al., 1994, Cootes et al., 1995 first introduced the term active shape model. However, Cootes et al. (1995) do not include the gray level appearance model and Cootes et al., 1994, Cootes et al., 1995 do not include the multi-resolution ASM scheme. Both of these components are
Point distribution model
The analysis of the shape vectors x gives insight in the typical variations in shape of lungs, heart and clavicles that occur in chest radiographs, and their correlation. This is an interesting result in its own right, and therefore the first few modes of variation are displayed in Fig. 2. In Fig. 3 the spread of each model point after Procrustes alignment is displayed. This is another way of visualizing which parts of the objects exhibit most shape variation.
Folds
The 247 cases in the JSRT database
Discussion
Some of the presented results obtained by computer algorithms are very close to human performance. Therefore, we start this discussion by considering the limitations of manual segmentations which were used to determine the gold standard, and discuss the representativity of the data. Then the results for lung segmentation, heart segmentation, clavicle segmentation and the automatic determination of the CTR are discussed. After pointing out the fundamental differences between pixel
Conclusions
A large experimental study has been presented in which several versions of three fully automated supervised segmentation algorithms have been compared.
The methods were active shape models (ASM), active appearance models (AAM) and pixel classification (PC). The task was to segment lung fields, heart, and clavicles from standard chest radiographs. Results were evaluated quantitatively, and compared with the performance of an independent human observer. The images, manual annotations and results
Acknowledgements
The authors gratefully acknowledge R. Nievelstein for supervising the manual segmentation process and G. Mochel and A. Scheenstra for each clicking around 75,000 points to segment the images.
References (56)
- et al.
Automated lung segmentation in digitized postero-anterior chest radiographs
Academic Radiology
(1998) - et al.
Roentgenographic determination of total lung capacity. A new method evaluated in health, emphysema and congestive heart failure
American Journal of Medicine
(1960) - et al.
Anatomic landmarks to estimate the length of the diaphragm from chest radiographs: effects of emphysema and lung volume reduction surgery
Chest
(2001) - et al.
Comparing methods of measurement: why plotting difference against standard method is misleading
Lancet
(1995) - et al.
Knowledge-based method for segmentation and analysis of lung boundaries in chest X-ray images
Computerized Medical Imaging and Graphics
(1998) - et al.
A fully automatic and robust brain MRI tissue classification method
Medical Image Analysis
(2003) - et al.
Active shape models – their training and application
Computer Vision and Image Understanding
(1995) - et al.
Using the chest radiograph to determine intravascular volume status: the role of vascular pedicle width
Chest
(2002) - et al.
Computer diagnosis of heart disease
Radiological Clinics of North America
(1971) - et al.
Relaxation labelling algorithms – a review
Image and Vision Computing
(1985)
Improved method for automatic identification of lung regions on chest radiographs
Academic Radiology
Face alignment using texture-constrained active shape models
Image and Vision Computing
An optimal algorithm for approximate nearest neighbor searching in fixed dimensions
Journal of the ACM
Review of MR image segmentation techniques using pattern recognition
Medical Physics
Principal warps: thin-plate splines and the decomposition of deformations
IEEE Transactions on Pattern Analysis and Machine Intelligence
The Laplacian pyramid as a compact image code
IEEE Transactions on Communications
The use of active shape models for locating structures in medical images
Image and Vision Computing
Active appearance models
IEEE Transactions on Pattern Analysis and Machine Intelligence
A minimum description length approach to statistical shape modeling
IEEE Transactions on Medical Imaging
Adapting active shape models for 3D segmentation of tubular structures in medical images
Pattern classification
A fully automatic algorithm for the segmentation of lung fields in digital chest radiographic images
Medical Physics
Practical methods of optimization
Valmet: a new validation tool for assessing and improving 3D object segmentation
Procrustes methods in the statistical analysis of shapes
Journal of the Royal Statistical Society B
Cited by (511)
GAN inversion-based semi-supervised learning for medical image segmentation
2024, Biomedical Signal Processing and ControlHybrid CNN-transformer network for interactive learning of challenging musculoskeletal images
2024, Computer Methods and Programs in BiomedicineExpertNet: Defeat noisy labels by deep expert consultation paradigm for pneumoconiosis staging on chest radiographs
2023, Expert Systems with ApplicationsA deep unsupervised saliency model for lung segmentation in chest X-ray images
2023, Biomedical Signal Processing and Control