Novel Classification of Current Methods, Available ... - CiteSeerX

7 downloads 273387 Views 358KB Size Report
Those AI-Based segmentation methods are classified in four classes which include ..... [25] http://www.medgadget.com/archives/2006/08/model_based_seg.html.
Novel Classification of Current Methods, Available Softwares and Datasets in Medical Image Segmentation Maryam Rastgarpour1 and Jamshid Shanbehzadeh2 Department of Computer Engineering, Saveh Branch, Islamic Azad University, Saveh, Iran 2 Department of Computer Engineering, Tarbiat Moallem University, Tehran, Iran

1

Abstract- Disease type, image features, modality, dimension of imaging and etc. make segmentation challenging in medical applications. This results in plenty of literatures which take beginner researchers in trouble. Classification of literatures can be a good outline to overview simply, properly and quickly. Moreover, Artificial Intelligence techniques improve efficiency of segmentation severely. This paper proposes a novel literature classification based on applied segmentation method. The segmentation methods are generally based on non-AI techniques and AI techniques. Those AI-Based segmentation methods are classified in four classes which include of Digital image processing techniques, combination of digital image processing techniques and other AI techniques, knowledge-based and rule-based system and registration-based algorithms respectively. Some available softwares and tools are declared and classified based on application, modality, class of segmentation method, dimension of supported images and corresponding body parts. Available datasets are also mentioned for validity finally. Keywords: Image Segmentation, Artificial Intelligence, Medical Application, Dataset, Tools, Classification

1 INRODUCTION Quantitative analysis of medical images, namely the measurement of volumes, needs to describe anatomy structures. Medical Image Analysis (MIA) can obtain this information. In MIA, segmentation is required to simplify later steps like feature extraction, image measurement and ROI representation. So it is crucial and important, because incorrect segmentation leads to incorrect analysis [1- 7]. Unfortunately segmentation is very difficult and is often done manually by human operator. It leads to segmentation be too time-consuming. So many scans segmentation is not possible at all. Additionally the number of images which should be analyzed is growing strongly due to technological advances. Hence, manual segmentation becomes less efficient possibility in clinical operations as well as human interpretation may not be produced suitable. Thus intelligent tools are so essential to segment semi or

full automatically. In medical applications, segmentation identifies the boundaries of ROIs including the bony structures like brain parts and tumors, breast calcification, prostate, iris, abdomen, pulmonary fissure and etc [8]. An example of prostate segmentations is shown in figure 1.

.

.

Figure 1 a sample of medical image segmentations. Original image of prostate on the left and segmentation of prostate tumor on the right [15]

Medical segmentation methods depend on many factors like disease type, image features, modality and dimension of imaging. So these dependencies result in Medical Image Segmentation (MIS) remains challenging as well as there is a significant growth of literatures annually [2, 9]. It takes beginner researchers in difficult to get a comprehensive and proper overview. There are some literature articles in journals and conferences proceeding related to medicine, biomedical engineering and computer engineering like [2, 46, 10-18]. Classification of the literatures can be a good outline to overview them simply, properly and quickly. Thus there are a few literature classifications [2, 4-6, 10-18] based on following aspects:  Author’s view  Human intervention  Disease features  Model and prior knowledge application  Evaluation and validation of segmentation method  Local or general data  Modality and dimension of imaging. Some ones are summarized in the following. Hemant et al. [19] presented a good survey paper which classified the segmentation methods in four sets for brain tumor of MRI. They included of Non AI techniques, Neural techniques, Fuzzy techniques and Hybrid of Fuzzy and Neural

techniques. They considered Maximum Likelihood (ML) approach and Expectation Maximization (EM) algorithm in set of non-AI techniques. Multilayer perceptron (MLP) algorithm and Kohonen algorithm were classified in set of Neural techniques as well as Fuzzy C-Means and Fuzzy watershed were categorized in set of Fuzzy techniques. Withey classified literatures in three generation [4, 12]. The first generation includes application of image processing methods. It is the simplest form of image analysis and the lowest processing level. Second generation applies the models, optimization methods and uncertainty models and generally avoids discoveries. Finally, the methods of third one need to higher level of knowledge like prior information, some rules defined by experts and models of ROI form. Yan [20] divided current MR brain image segmentation algorithms into three categories: classification-based, region-based, and contour-based, and discussed the advantages and disadvantages of these approaches. Dellepiane [18] classified literatures in tree form which partitioned algorithms based on directorial parameters to their goal. Main group was based on density, topology, and geometry. Pan [17] proposed four groups for segmentation methods. They include interactive thresholding, edge detection, regions split and merge and finally hybrid methods. Thresholding techniques comprise all the pixels with a threshold and identify those pixels within a range of specific areas. Selected threshold is vital in these methods. Boundary based methods use edge detection techniques like gradient filter to locate region boundary. But these methods are noise sensitive. Region based techniques claim that region pixels have similar features. A general growing procedure starts some seed points and comprises each pixel by neighbors. If merging criterion satisfied, pixel is classified to that class. Merge criterion selection is vital to success segmentation process. Clarke et al. [6] reviewed MRI segmentation method until 1995. This article looked over them in aspect of single or multiple spectral and supervised or unsupervised. The rest of this paper is organized as follows. Section 2 proposes a novel literature classification based on its

segmentation method in five classes. Section 3 declare some available tools and softwares for MIS and classifies based on application, modality, class of segmentation method, corresponding body parts and so on. Some useful datasets are mentioned for validity in section 4. Finally this paper concludes in section 5.

2 NOVEL CLASSIFICATION OF CURRENT METHODS Comprehensive and proper overview is provided by a good literature classification. Inspired of [19, 21] one can classify available literatures based on applied segmentation method in aspect of non-AI techniques and AI techniques generally. Figure 2 summarized medical segmentation methods. Those segmentation methods based on AI techniques can be included in four classes: Class 1: Based on Digital image processing techniquesIn literatures of this class [15,17,22], segmentation needs low-level processing and use thresholding, edge detection, and region growing for segmentation. Class 2: Based on Combination of digital image processing techniques with other AI techniques- These segmentation methods[23,24] use pattern recognition and machine learning algorithms (i.e., c- means clustering, artificial neural networks, active contours, level set, hidden markov models and so on) in combination of Image processing techniques and tries to be semi-automatic. Class 3: Knowledge-based and rule-based systemsThese systems [5,25,26] applies the experts’ knowledge as rules, models and atlases with combination of AI methods to segment full automatically. Class 4: Registration-based algorithms- These algorithms [6,11,18,27-29] use multispectral and multimodal images in expert system of MIS. In this class, images need to be registered properly before segmentation so it’s called Co-registration Segmentation. This segmentation method uses more several features of multiple modalities than the other classes to increase accuracy.

Medical Image Segmentation Methods

Non-AI techniques based

AI techniques based

Maximum Likelihood approach

Digital Image Processing techniques

Expected Maximization algorithm

Hybrid methods

Energy Minimization approach

Knowledge-based and rule-based methods Registration based methods

Figure 2 Summarization of available segmentation methods based on application of AI methods and non-AI methods.

Each class appears one after the other. Complex and accuracy would be increased and Human intervention would be decreased in progress of classes. There are many efforts to dominate the problems and difficulties in the methods of each class but the result remains already data dependent. Most traditional segmentation techniques use images in one modality like MR or CT. But Class 4 applies multiple images of an organ to have several features by using variant modalities such as CT, MR, PET, ultrasound, or collection of images over time. These features result in closer segmentation. These methods are called multispectral or multi-modal. There are many intelligent methods applied in this class such as k-means, fuzzy c-means and adaptive template moderated spatially varying statistical classification etc.

3 NOVEL CLASSIFICATION OF THE CURRENT TOOLS There are a lot of tools and softwares for MIS in [2, 1113, 30-39]. Some of available novel tools is listed in table 2. For more information, they are classified based on the modality, application, dimension, class of segmentation method and corresponding body parts as well. Table2 Classification of novel Tools in MIS Name and Ref. [25] Royal Philips Electronics

Segment [30]

Image Dime nsion

Modality

Body parts

Seg. method

Cla ss

3D

CT MRI

cardiacvascular, artery, all organs

Model-based

3

2D 3D

IATR[31] 3D

MRI CT PET SPECT MRI CT

MDSTK [38]

2D 3D

All Image slice

ITKSNAP [39]

3D

MRI

myocardial perfusion , cardiovascular image (also Fast level set wide range of radiology and robust edgecardiology applications) tracking step wrist with an active contour algorithm , level set Edge All organs detection, thresholding, RG, FCM, Watershed Brain, Liver, level set, Kidney and so active on contours methods

scapula, humeral, clavicle, fingers, spine and pelvis are mentioned and can be available for more researches. Different sets of medical images with pathology to test on high-resolution images with dimensions 4 and 5 in different modalities are available in [41] and even combination of several modalities such as PET-CT of different organs and also teeth is possible. 63 obtained data sets from CT scanners and MRI of the human and other creatures such as monkeys and rabbits, etc. are in [42]. The images of human are related to the head, brain, female breast, knee, teeth, claws and toe. In [43] some real data in the form of brain MRI has been simulated for validation pathologies which lead to tumors and lesions in the brain. In this dataset, synthetic brain MR images can be built up to show tumors by putting pathology to a healthy brain MRI image with known ground truth. Set of MR brain images with guidance segmentation boundaries provided manually has been compiled to research in [44]. [45, 46] Simulated brain database with the known truth has been presented that has the ability to customize. In [47] useful information to create the database of MR and CT images using the available tools in [35,48] is mentioned.

5 CONCLUSION This paper presented the novel classification of the most recent important methods based on human intervention decrement a nd application of AI methods or non-AI methods. It also proposed the novel classification of available medical softwares declared and classified based on modality, application, applied segmentation method, dimension of supported images and corresponding part of body. Some validity dataset was mentioned at last.

6 REFERENCES 3

4

1

2

4 VALIDITY DATASET Additions of [4], there are many dataset which can be explained more in the next. In [40] several data sets from CT-Scan images related to human bone structures, such as

[1] Bankman, I.: Handbook of Medical Image Processing and Analysis, Elsevier, 2009. [2] Zaidi, H.: Medical image segmentation Quo Vadis – comparison. J. Elsevier Computer Methods and Programs in Biomedicine, Vol. 84, Issues 2-3, pp.63—65, 2006. [3] Melonakos, J.: GEODESIC TRACTOGRAPHY SEGMENTATION FOR DIRECTIONAL MEDICAL IMAGE ANALYSIS. Phd thesis, georgia institute of technology, 2009. [4] Withey, D.J. ,Koles, Z.J.: Three Generations of Medical Image Segmentation-Methods , Available Software. International J. of Bioelectromagnetism, Vol. 9 No. 2 , 2007. [5] Pham, D.L., Xu, Ch., Princo, J.L.: a survey of current methods in medical image segmentation. Annual Review of Biomedical Engineering, vol. 2, p.p. 315—338, 2000. [6] Clarke, L.P., Velthuizen, R.P., Camacho, M.A., Heine, J.J., Vaidyanathan, M., Hall, LO, Thatcher, R.W., Silbiger, M.L.: MRI segmentation: Methods and applications. Magn. Reson. Imag., vol. 13, pp. 343–368, 1995. [7] Atkins, M.S. Mackiewich, B.T.: Fully Automatic Segmentation of the Brain in MRI. Medical Imaging, IEEE Transactions on, vol. 17, Issue 1, pp. 98-107, 1998. [8] M.Rastgarpour, J.Shanbehzadeh, and H. Farahanirad ―The Status Quo of Artificial Intelligence Methods in Automatic Medical Image‖ IEEE Proc. Of ICICA International Conference on Information and Computer Applications (ICICA 2011), Dubai, p.p. 413-417, 2011. [9] M.Rastgarpour and J.Shanbehzadeh, ―The Problems, Applications and

Growing Interest in Automatic Segmentation of Medical Images from the year 2000 till now‖ IEEE Proc. Of ICICA International Conference on Information and Computer Applications(ICICA 2011), Dubai, p.p. 409412 , 2011. [10] Yang, D., Zheng, J., Nofal, A., Deasy, J., El Naqa, I.M.: Techniques and software tool for 3D multimodality medical image segmentation. J. RADIATION ONCOLOGY INFORMATICS, VOL. 1, NO. 1, 2009. [11] Olabarriag, S.D, Smeulders, A.W.M: Interaction in the segmentation of medical images: A survey. Med.Image. Anal., Elsevier, pp. 127—142, 2001. [12] Withey, D.J., Kole, Z.J.: Medical Image Segmentation: Methods and Software. IEEE Proc. Of NFSI&ICFBI, China , pp. 140- 14, 2007. [13] Pitiot, A., Delingette, H., Thompson, P.M.: Automated Image segmentation: issues and applications. Ch. 6 of ―Medical Imaging Systems Technology: Methods in general anatomy (Medical Imaging Systems Technology)‖ (v. 3) by Cornelius T. Leondes pp. 195—241, 2005. [14] Sharma, N., Aggarwal, L.M.: Automated medical image segmentation t e c h n i q u e s . J. of medical physics, pp. 3—14 2010. (http://www.ncbi.nlm.nih.gov/pmc/articles/PMC2825001/) [15] Neufeld, E., Samaras, T., Chavannes, N., Kuster, N.: ROBUST, HIGHLY DETAILED MEDICAL IMAGE SEGMENTATION, 2006.(http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.116.408 0&rep=rep1&type=pdf) [16] Souplet, J.Ch.: Medical Image Navigation and Research Tool by INRIA (MedINRIA). 2007. [17] Pan, Zh., Lu, J.: A Bayes-Based Region-Growing Algorithm for Medical Image Segmentation. IEEE trans. Vol.9, issue 4, pp. 32— 38, 2007. [18] Dellepiane S.: The active role of 2–D and 3–D images: Semiautomatic segmentation. In: Roux C Coatrieux JL, (eds.), Contemporary Perspectives in Three-Dimensional Biomedical Imaging, IOS Press, pp.165–189, 1997. [19] D. JUDE HEMANTH, C. KEZI SELVA VIJILA, J. ANITHA ―A SURVEY ON ARTIFICIAL INTELLIGENCE BASED BRAIN PATHOLOGY IDENTIFICATION TECHNIQUES IN MAGNETIC RESONANCE IMAGES (Survey Paper)‖, International Journal of Reviews in Computing, p.p. 30-45, 2009. [20] Hong Yan ―Current methods in the automatic tissue segmentation of 3D magnetic resonance brain images, Current medical imaging reviews‖, 2006. [21] M.Rastgarpour and J.Shanbehzadeh, ―Application of AI Techniques in Medical Image Segmentation and Novel Categorization of Available Methods and Tools‖ IAENG Proc. Of 6th International Multiconference Of Engineers And Computer Scientists (IMECS), Vol I, Hong Kong, p.p. 519-523, March 2011. [22] Batenburg, K.J., Sijbers, J.: Optimal Threshold Selection for Tomogram Segmentation by Projection Distance Minimization. IEEE trans. Medical Imaging, Vol. 28, Issue 5, pp.676 – 686, 2009. [23] Hatt, M., Cheze le Rest, C., Turzo, A., Roux, C., Visvikis, D.: A Fuzzy Locally Adaptive Bayesian Segmentation Approach for Volume Determination in PET. IEEE trans. On Medical Imaging, Vol. 28, Issue 6 , Pp. 881 – 893, 2009. [24] Segonne, F., Fischl, B.: Integration of Topological Constraints in

Medical Image Segmentation. Bimed. Im. Analysis: Methods and applications, 2007. [25] http://www.medgadget.com/archives/2006/08/model_based_seg.html [26] Isgum, I., Staring, M., Rutten, A., Prokop, M., Viergever, M.A., van Ginneken, B.: Multi-Atlas-Based Segmentation with Local Decision Fusion—Application to Cardiac and Aortic Segmentation in CT Scans. Medical Imaging, IEEE Trans. Vol. 28, issue 7, pp. 1000— 1010, 2008. [27] Ardekani, B.A.,Braun, M.,Hutton, B.F.,Kanno, I., Iida, H.:A fully automatic multimodality image registration algorithm. Journal of Computer Assisted Tomography. Vol. 19, issue4, pp. 615–623, 1995. [28] Andersen, A.H., Zhang, Z.,Avison, M.J., Gash, D.M.: Automated segmentation of multispectral brain MR images. J Neurosci Methods, vol.122, issue1, pp.13–23, 2002. [29] Farag, A.A., El-Baz, A.S., Gimel’farb G.: Precise segmentation of multimodal images. IEEE Trans Image Process.,vol.15, issue 4, pp.952– 968, 2006. [30] Heiberg, E., Sjögren, J., Ugander, M., Carlsson, M., Engblom, H., Arheden, H.: Design and Validation of Segment - Freely Available Software for Cardiovascular Image Analysis. BMC Medical Imaging, 10:1, 2010 (http://segment.heiberg.se) [31] http://www.cma.mgh.harvard.edu/iatr/ [32] Yoo, T.S., Metaxas, D.N.: Open science - combining open data and open source software: Medical image analysis with the Insight Toolkit. J. Medical Image Analysis, Elsevier B.V., vol. 9, issue 6, pp. 503—506, 2005. [33] http://medfloss.org/node/62 [34] Duryea, J., Magalnick, M., Alli, S, Yao, L., Wilson, M., Goldbach-Mansky, R.: Semiautomated three-dimensional segmentation software to quantify carpal bone volume changes on wrist CT scans for arthritis assessment. J. Med Phys. ,35(6), pp. 2321-30, 2008. [35] http://www.itk.org/ [36] http://white.stanford.edu/~brian/mri/segmentUnfold.htm [37] http://www.cma.mgh.harvard.edu/seg/tools.html [38] http://sourceforge.net/projects/mdstk/ [39] http://www.itksnap.org [40] Duryea, J., Magalnick, M., Alli, S, Yao, L., Wilson, M., GoldbachMansky, R.: Semiautomated three-dimensional segmentation software to quantify carpal bone volume changes on wrist CT scans for arthritis assessment. J. Med Phys. ,35(6), pp. 2321-30, 2008. [41] http://www.celebisoftware.com/Dataset.aspx [42] http://www9.informatik.uni-erlangen.de/External/vollib/ [43] Prastawa, M., Bullitt, E., Gerig, G.: Simulation of Brain Tumors in MR Images for Evaluation of Segmentation Efficacy. J. Medical Image Analysis (MedIA). Vol. 13, No. 2, pp. 297—311, 2009. [44] http://www.cma.mgh.harvard.edu/ibsr/data.html [45] http://mouldy.bic.mni.mcgill.ca/brainweb/ [46] http://www.ucnia.org/softwaredata/5-tumordata/10simtumordb.htmlhttps://www.rad.upenn.edu/sbia/software/index.html#w mls [47] http://www.sci.utah.edu/cibc/wiki/index.php/Segmentation_of_CT_and_ MRI_datasets_using_3D_Slicer#Segmentation [48] http://www.na-mic.org/

Suggest Documents