J Med Syst (2011) 35:463–471 DOI 10.1007/s10916-009-9382-6
ORIGINAL PAPER
Statistical Approach for Brain Cancer Classification Using a Region Growing Threshold Bassam Al-Naami & Adnan Bashir & Hani Amasha & Jamal Al-Nabulsi & Abdul-Majeed Almalty
Received: 28 June 2009 / Accepted: 22 September 2009 / Published online: 16 October 2009 # Springer Science + Business Media, LLC 2009
Abstract In brain cancer, a biopsy as an invasive procedure is needed in order to differentiate between malignant and benign brain tumor. However, in some cases, it is difficult or harmful to perform such a procedure, to the brain. The aim of this study is to investigate a new method in maximizing the probability of brain cancer type detection without actual biopsy procedure. The proposed method combines both image and statistical analysis for tumor type detection. It employed image filtration and segmentation of the target region of interest with MRI to assure an accurate statistical interpretation of the results. Statistical analysis was based on utilizing the mean, range, box plot, and testing of hypothesis techniques to reach acceptable and accurate results in differentiating between those two types. This method was performed, examined and compared on actual patients with brain tumors. The results showed that the proposed method was quite successful in distinguishing between malignant and benign brain tumor with 95% confident that the results are correct based on statistical testing of hypothesis. B. Al-Naami (*) : H. Amasha : J. Al-Nabulsi Department of Biomedical Engineering, Hashemite University, P.O. Box 150459, Zarqa 13115, Jordan e-mail:
[email protected] H. Amasha e-mail:
[email protected] J. Al-Nabulsi e-mail:
[email protected] A. Bashir Department of Industrial Engineering, Hashemite University, P.O. Box 150459, Zarqa 13115, Jordan e-mail:
[email protected] A.-M. Almalty Physical Therapy Department, Hashemite University, P.O Box 330018, Zarqa 13133, Jordan e-mail:
[email protected]
Keywords Brain cancer (BCa) . Filtration . Segmentation . Box plot . T-test . Inter quartile range (IQR)
Introduction Primary brain tumors continue to be relatively insensitive to cancer treatments like radiation and chemotherapy and little is still known about environmental or genetic risk factors. The number of different types of brain tumors known to date are around 126, ranging from relatively benign (meningiomas) to highly malignant (glioblastomas). The term “cancer” is generally not applied to brain tumors since they are not equivalent to most other malignancies that cause damage and ultimately death through metastasis. Collectively, primary brain tumors are considered a rare disease, with an anticipated incidence of 18,820 new cases and 12,820 deaths in the United States in the year of 2006 [1, 3]. In Jordan, where the analyzed data have taken place, the cancer is the second most frequent cause of death after heart disease. In 2004, according to the Jordan National Cancer Registry (JNCR), about 3,591 new cancer cases have been registered among Jordanians with an incidence rate of 67.1 per 100,000 populations (63.9 for males and 70.5 for females). Among the most common cancers affecting Jordanian population, the brain tumor was ranked as the second common type in children and the seventh in adults and the prevalence in males 5.5% more than females (3.5%) [2]. Different diagnostic procedures have been followed in attempt to differentiate between the benign and malignant tumor such as CT or MRI images, angiogram, skull X-ray, spinal tap, myelogram, and biopsy which is the most accurate procedure that determines the brain tumor type. A biopsy involves removing a piece of the tumor for viewing
464
under a microscope by a pathologist to differentiate the benign tumor cells from malignant ones. However, sometime this kind of procedure is not applicable for some patients with brain cancer and could be life threatening. Especially, those patients who are having tumors in locations that hard to be reached (e.g. brain stem) without causing crucial damage to the healthy brain tissues. Thus, the present study is to investigate a new method in diagnosing the type of brain tumor without having biopsy be done the patient using statistical analysis and image analysis. Detection of brain cancer types (BCaT) was presented in different techniques in the literature. The image techniques used for brain image segmentation can be categorized into three categories: (1) threshold-based segmentation, (2) statistical methods for brain segmentation and (3) region growing methods. The threshold-based segmentation category was proposed to use iterative thresholding, histogram analysis and morphological operations. Some of these techniques are followed by some refinement of the segmented zones [4– 10]. Taheri et al. [4] proposed a three-dimensional segmentation approach to achieve a proper estimation of tumor volume. They introduced a threshold-based algorithm for 3D tumor segmentation using level set depending on the complexity of the tumor shape. Prastawa et al. [5] provided a brain tumor segmentation framework based on outlier detection. He used T2 MRI to make brain tumor segmentation by detection the abnormal regions in the brain, then determine intensity properties in the tissues and test if the edema exists with the tumor. The geometric and spatial constraints then applied to the detected tumor and edema regions. Segmentation procedure has been applied to three real datasets, representing different tumor shapes, locations, sizes, image intensities, and enhancement. Continuation for this methodology combines physical and statistical modeling to generate synthetic multi-modal 3D brain MRI with tumor and edema, along with the underlying anatomical ground truth, with emphasizing on simulation of the major effects known for tumor MRI [6]. These effects include contrast enhancement, local distortion of healthy tissue, infiltrating edema adjacent to tumors, destruction and deformation of fiber tracts, and multi-modal MRI contrast of healthy tissue and pathology. Statistical methods represent another important category in the segmentation process and most of the approaches proposed in this category were using some statistical classifications combined with different image processing techniques in order to segment the MRI images. Also, the use of statistical methods in texture analysis has been proposed [11–13]. Justin et al. have used a statistical analysis of fractal-based brain tumor detection algorithms [11]. They studied brain tumors as geometric objects that
J Med Syst (2011) 35:463–471
have a noninteger fractal dimension (FD). The used technique shows that the FD is useful in the detection of the possible presence and location of brain tumors when a reference nontumor image is available and then followed by statistically validating the results of FD analysis (p mmb Þ;
Benign tumor
Malignant tumor
134.6316 14.84816 136 54 109 163 125 139.5
205.035 31.0622 196 149 160 255 186.5 235.5
19
23
Some weaknesses of the algorithm are the small area of the overlap between the extreme values of the two different tumor types. The case can be seen for the cases 5, 8 of malignant tumor in Table 1. To overcome this drawback, it is suggested to increase the number of samples used for the study.
Conclusion
Where: H0: is the null hypothesis Ha: is the alternative hypothesis μmm: is the max. pixel value of malignant images μmb: is the max. pixel value of benign images The results for using the two sample (t) test are shown below; these parameters are used to calculate the (t) statistics, n1 ¼ 23; n2 ¼ 19; df ¼ 40; x1 ¼ 205:04; x2 ¼ 134:63; S12 ¼ 964:86; S22 ¼ 220:48; S 2 ¼ 629:88; t ðaÞ ¼ 1:684 and t ðmeasuredÞ ¼ 9:05 Since the t(measured) is located in the rejection region, we reject the null hypotheses H0 and strongly conclude that the maximum pixel value for malignant is greater than the maximum pixel value for benign images.
It has been proven that a simple, harmless and accurate statistical technique can efficiently distinguish between the malignant and benign tumor. The huge advantage of this approach is that there will be no need to make any further tests or examinations on the patient after making the MRI. Especially where some of these tests or examinations are difficult to be performed or may be dangerous such as brain biopsy. The decision on the tumor type can be decided by investigating the range, the mean and the maximum of the pixel values of the image. It is also noted that filtration is an essential process in this analysis due to the fact that most of the images contain noise. The results of the experiments show that this method proposed to detect the brain cancer type may also be applied to the other areas of the medical image analysis. Box Plot 300 275 250 225 200 175 150 125 100 75 50 25 0
150 125 100 75 50 25 0 benign
Pixels
Box Plot
malignant
Type of image Fig. 10 Box plot for the average pixel value
benign
malignant
Fig. 11 Box plot for the maximum pixel value
Pixels
Parameters
J Med Syst (2011) 35:463–471
References 1. National Cancer Institute, USA. http://www.cancer.gov. 2. King Hussein Cancer Centre, Jordan. http://www.khcc.jo. 3. Jemal, A., Murray, T., Ward, E., et al., Cancer statistics 2005, CA. Cancer J. Clin. 55:10–30, 2005. 4. Taheri, S., Sim Heng O., Chong, V., Threshold-based 3D tumor segmentation using level set (TSL). IEEE Workshop on Applications of Computer Vision, pp. 45–45, 2007. 5. Prastawa, M., Bullitt, E., Ho, S., and Gerig, G., A brain tumor segmentation framework based on outlier detection. Med. Image Anal. 8 (3)275–283, 2004. 6. Prastawa, M., Bullitt, E., and Gerig, G., Simulation of brain tumors in MR images for evaluation of segmentation efficacy. Med. Image Anal. 13 (2)297–311, 2009. 7. Fletcher-Heath, L., Dmitry, O., Goldgof, B., and Reed Murtagh, F., Automatic segmentation of non-enhancing brain tumors in magnetic resonance images. Artif. Intell. Med. 21 (1–3)43–63, 2001. 8. Clark, M. C., Hall, L. O., Goldgof, D. B., Velthuizen, R., Murtagh, F. R., and Silbiger, M., Automatic tumor-segmentation using knowledge-based techniques. IEEE TMI. 117:187–201, 1998. 9. Kaus, M., Warfield, S., Nabavi, A., Black, P. M., Jolesz, F. A., and Kikinis, R., Automated segmentation of MR images of brain tumors. Radiology. 218:586–591, 2001. 10. Leemput, K., Maes, F., Vandermeulen, D., and Suetens, P., Automated model-based tissue classification of MR images of the brain. IEEE TMI. 18:897–908, 1999. 11. Justin, M., and Iftekharuddin, K., Statistical analysis of fractalbased brain tumor detection algorithms. Magn. Reson. Imaging. 23 (5)671–678, 2005. 12. Ashraf, M., Evangelia, I., Dinggang, S., and Christos, D., Deformable registration of brain tumor images via a statistical model of tumor-induced deformation. Med. Image Anal. 10 (5) 752–763, 2006.
471 13. Xiao, X., and Qingmin, L., Statistical structure analysis in MRI brain tumor segmentation. Image and Graphics, ICIG 2007. Fourth International Conference, pp. 421–426, 2007. 14. Mancas, M., Gosselin, B., and Macq, B., Segmentation using a region growing thresholding. Proceedings of the Electronic Imaging Conference of the International Society for Optical Imaging SPIE. 5672:388–398, 2005. 15. Sezgin, M., and Sankur, B., Survey over image thresholding techniques and quantitative performance evaluation. J. Electron. Imaging. 13 (1)146–165, 2004. 16. Chowdhury, M., and Little, W., Image thresholding techniques. IEEE Pacific Rim Conference on Communications, Computers, and Signal Processing. 17–19:585–589, 1995. 17. Pan, Zhigeng, P., and Jianfeng, L., A Bayes-based region-growing algorithm for medical image segmentation. Comput. Sci. Eng. 9 (4)32–38, 2007. 18. Chuang, C. -L., and Chen, C. -M., A novel region-based approach for extracting brain tumor in CT images with precision. World Congress on Medical Physics and Biomedical Engineering. IFMBE Proceedings. 14 (4)2488–2492, 2006. 19. Chunyan, J., Xinhua, Z., Wanjun, H., and Meinel, C., Segmentation and quantification of brain tumor, virtual environments, human–computer interfaces and measurement systems. IEEE Symp. 12–14:61–66, 2004. 20. Velthuizen, R., Clarke, L., Phuphianich, S., Hall, L., Bensaid, A., Arrington, J., Greenberg, H., and Siblinger, M., Unsupervised measurement of brain tumor volume on MR images. J. Magn. Reson. Imaging. 5 (5)594–605, 1995. 21. Vinitski, S., Gonzales, C., Mohamed, F., Iwanaga, T., Knobler, R., Khalili, K., and Mack, J., Improved intracranial lesion characterization by tissue segmentation based on a 3D feature map. Magn. Reson. Med. 37 (3)457–469, 1997. 22. Gonzalez, R. C., Woods, R. E., and Eddins, S. L., Digital image processing, 2nd edition. Prentice Hall, Upper Saddle River, 2004.
