automated tumors extraction and visualization in nasal and
Transcription
automated tumors extraction and visualization in nasal and
AUTOMATED TUMORS EXTRACTION AND VISUALIZATION IN NASAL AND PARANASAL SINUSES Mohammed A. Berbar 1,Omar A. El-Banhawy 2, Aliaa H. El-Seady3 1 Department of Computer Engineering and Sciences, Faculty of Electronic Engineering, El Menoufyia University, Egypt 2 Departments of ENT, Faculty of Medicine, El Menoufyia University, Egypt 3 Department of Mathematics, Faculty of Sciences, El Menoufyia University, Egypt ABSTRACT This research is concerned with biomedical image analysis and visualization and enabling the 3D (Three-Dimensional) representation of biomedical image data sets. These data sets are input as a sequence of 2D (Two-Dimensional) CT images of some patients suffering from tumor or extra bone in nasal and paranasal sinuses. The developed system can isolate and produce 3D model for the tumor or extra bone. The segmentation is simply done by limiting lower and upper density in 2D. The developed system could be used to detect the tumor or extra bone, isolate it, forming a 3D model, calculating the improvements will happen in air volume in nasal and paranasal after removing that tumor or the extra bone to show the benefits of the operation for the patient. This work used CT/MRI (Computerized Topography and Magnetic Resonance Imaging) of patients with juvenile angiofibroma and CT/MRI of patients with bilateral simple nasal polyp. They were diagnosed at ENT (Ear, Nose and Throat) out patient clinic of El-Menoufyia University Hospital, Shebin El-Kom, Egypt.. Experimental work proved that the tumor surrounding could be extracted completely. A large prospective randomized study has been done before considering this system. With these advances in hand, there are several important applications possible to be delivered soon that will have a significant impact on the practice of medicine and on biological research. Key Words Biomedical image analysis, Visualization, Segmentation, 3D representation. 1. INTRODUCTION Many advances in medicine have arisen over the past twenty years due to the large-scale integration of computers. It was quickly recognized that computer technology could be used for storing and correlating information, and as the technological explosion continued computers began being used increasingly with diagnostic medicine, surgery planning [1-5], and automatic segmentation of biomedical images [6]. Current high-performance computers and advanced image processing capabilities have facilitated major progress toward realization of forming the images in more intuitive than the 2D representation provided by paper and slides. The effect of turning a set of 2D images into a 3D representation as building a graph from a set of values. Producing 3D worlds from medical data sets is generally referred to as scientific visualization and many computer scientists from both images processing and computer graphics backgrounds have contributed to the work done in this area [7 - 13]. Medical imaging system such as computer topography (CT) and magnetic resonance (MR) imaging (MRI) are used to produce accurate 3D images of different organs and tissues [14]. Such as 3D images consist of a stack of 2D images in which the stack forms the third dimension as shown in Figure 1. Figure 1: An Example of 3D images consists of a stack of 2D images In this work, we concerned with using CT/MRI images of nasal and paranasal as an application of the developed system so the complexity of the paranasal sinuses anatomy is explained [15] and shown in Figure 2. Figure 2: The paranasal sinuses anatomy There are a total of four paired sinuses. They include the frontal, ethmoid, maxillary and sphenoid sinuses. These sinuses are essentially mucosa-lined airspaces within the bones of the face and skull. Their development begins in the womb, but results in only two clinically-relevant sinuses by birth--the maxillary and ethmoid sinuses. The development of the lateral nasal wall begins with a smooth, undifferentiated structure. The first outgrowth is the maxilloturbinal which will eventually become the inferior turbinate. Subsequently, another mound of mesenchyme forms which is the ethmoturbinal, destined to become the middle, superior, and supreme turbinates by subdividing into the second and third ethmoturbinals. This growth is followed by the development of the agger nasi cells, uncinate process, and ethmoid infundibulum. The sinuses then begin to develop. The resultant system of cavities, depressions, ostia, and processes is a complex system of structures which must be understood in detail before surgical management of sinus disease can be safe and effective. 2. MATERIALS AND METHODOLOGY The data sets are input to the developed biomedical image analysis and visualization system as sequences of slices of 2D images come from CT machine or from a scanner and digitizer. In the case of input slices of 2D images from the scanner, they are isolated into equal size slices using vertical and horizontal histogram. The slices have been co-registered [16-17] to each other. This registration ensures that the information from each of the images refers to the same physical structure and then feed into the 3D construction module. The CT/MRI images of the patients included in this study are taken directly from CT machine or from hard copy film, which digitized by Multi RAD 860 “film digitizer”. The used digitizer has the capability to capture the data in the hard copy in two standardized format either DICOM as that used in [18]or JPEG format [19]. For Example Figure 3 showed coronal CT scan of a case of type IIIb juvenile angiofibroma as captured by the digitizer in JPEG format. Figure 3: An Example of coronal CT scan in JPEG format. This data set consisted of 20 contiguous slices. The system extracts the CT slices by using vertical and horizontal histogram to find the separation rows and separation columns as shown in Figure 4 and Figure 5. Figure 5 : Horizontal histogram Figure 4 : Vertical histogram The results of segmentation of cronal CT scan that shown in Figure 3 are shown in Figure 6. Figure 6: The input image after segmentation into 20 slices. The proposed biomedical image analysis and visualization software is developed using IDL (Interactive Data Language) capable to: 1. View the sequence of 2D images as next and previous as shown in Figure 7 and construct them into 3D form to help the doctors in medical fields. 2. Determine the density of the diseased tissues in comparison to the rest of the surrounding areas, and remove the diseased tissues by the thresholding technique and the region growing technique in 3D space. The very small areas within 5-voxel-region were removed as false-positive areas 3. Visualizing and identify the important structures that may be at risk during surgery a.g. optic nerve or internal carotid artery. 4. Give a view of the field of surgery after diseased tissues subtraction and comparing this to the postoperative results. 5. Isolating a specific regions manual by setting the threshold lower and upper density by the user and display the results as shown in Figure 8, and building the 3D from them as shown in Figure 9. Figure 7: The main program interface. Figure 8: Isolating region from CT by threshold lower and upper density. Figure 9: Building 3D for the extracted tumor. 3. RESULTS The testing data for developed system contain three Groups of patients. The first group we will show Pre- and post- operative CT/MRI of patients with juvenile angiofibroma, and the postoperative CT/MRI were done after 2, 3 and 6 years of the operation. The second group we will show Preoperative CT/MRI of patients with juvenile angiofibroma was analyzed to determine the density of the diseased tissues in comparison to the rest of the surrounding areas and show visualization and identify the important structures that may be at risk during surgery a.g. Optic nerve or internal carotid artery. In the third and last group we will show Pre- and post- operative CT/ MRI of two patients with bilateral simple nasal polypi . We show in this document one patient from each group as shown in Figure 10, 11, 12. 3.1 Group 1 tumor (a) Post-contrast coronal MRI shows large heterogeneously enhanced soft tissue mass invading the left infratemporal fossa, orbital region and maxillary sinus. Tumor has been removed (b) Cronal view of CT scan of the same patient preoperative showing complete clearance of the tumor. (c) Two years post-operative post-contrast axial and cronal MRI of the same patient showing complete clearance of the tumor. (d) View of 3D for the tumor extracted using the developed software. Figure 10 Pre- and post-operative CT/MRI of a patient with juvenile angiofibroma. 3.2 Group 2 Tumor in orange color (a) Post-contrast cronal MRI shows large heterogeneously enhanced soft tissue Mass. Tumor has been removed (b) Visualization for live image of cronal view of CT scan of the same patient preoperative showing complete clearance of the tumor. Diseased tissues (c) Visualization for live contour by density to determine the diseased tissues in comparison to the rest of the surrounding areas . (c) View of 3D for the tumor (represented in blue) extracted using the developed software. Figure 11 Pre-operative CT/MRI of patient with juvenile angiofibroma. 3.3 Group 3 (a) Coronal view of CT scan for 12 years old boy showing R antrochoanal polyp originating from R maxillary sinus and protruding through R middle meatus to R nasal cavity and nasopharynx. (b) Cronal view of CT scan of the same patient preoperative showing complete removal of the lesion and enlarged maxillary sinus openining. (c) Coronal View of CT of the same patient 6 months postoperative showing complete removal of the lesion and enlarged maxillary sinus openining. (d) View of 3D for the tumor extracted using the developed software. Figure 12: Pre- and post- operative CT of patient with bilateral simple nasal polypi. 4. Discussion and Conclusions This research provides a software package for medical applications. The developed software package implemented by using IDL (Interactive Data Language). The developed software package takes an image data set from various sources and restructures it into a form that is more intuitive and formative for diagnostic use by expert. It has been tested and works well with image data sets of the sinus over time. The developed system could be used to detect the tumor or extra bone, isolate it, forming a 3D model, calculating the improvements will happen in air volume in nasal and paranasal after removing that tumor or the extra bone to show the benefits of the operation for the patient. The 3D visualization created by the software package can be run in a 3D browser, which in turn, can be viewed over the Internet. The browser enables the user to examine the visualization from any angle, and allows inspection of the data sets in a dynamic environment. The research also discuss the performance of the visualization. The future for this type of medical imagery is very promising. Powerful PC’s will make visualization tools with complex rendering algorithms practical. The availability of 3D visualizations over the Internet will prove invaluable to the medical personnel associated with data and conferencing tools within a virtual reality may soon be realized. 5. References [1] Kennedy D. W, “ functional endoscopic sinus surgery”, theary and diagnostic evaluation, archotolarygol head neck surge (1985) 111-576. [2] Shin Huha, Terence A. Ketterb, Kwang Hoon Sohna, Chulhee Leea, “ Automated cerebrum segmentation from three-dimensional sagittal brain MR images ” , Computers in Biology and Medicine 32 (2002) 311–328. [3] Anshul Sehgala, U.B. Desaib , “ 3D object recognition using Bayesian geometric hashing and pose clustering ”, Pattern Recognition 36 (2003) 765 – 780. [4] David R. Holmes III, Brian J. Davisb, Charles J. Brucec, Richard A. Robba, “ 3D visualization, analysis, and treatment of the prostate using trans-urethral ultrasound ”,Computerized Medical Imaging and Graphics 27 (2003) 339–349. [5] C.M. Wong, W.H. Chan, T.W. Lam, K.Y. Yip, “ Surface mapping of three-dimensional objects by a planar light scanning technique ” , Journal of Materials Processing Technology 139 (2003) 96–102. [6] Yan Kang, Klaus Engelke , Willi A. Kalender, “ Interactive 3D editing tools for image segmentation ” , Medical Image Analysis 8 (2004) 35–46. [7] Fabio Remondino, “ 3-D reconstruction of static human body shape from image sequence ”, Computer Vision and Image Understanding 93 (2004) 65–85. [8] Said Benameura, Max Mignottea, Stefan Parentd, Hubert Labelled, Wafa Skallie, Jacques de Guise, “ 3D/2D registration and segmentation of scoliotic vertebrae using statistical models ”, Computerized Medical Imaging and Graphics 27 (2003) 321–337. [9] K.Achour , M.Benkhelif, “ A new approach to 3D reconstruction without camera calibration ”, Pattern Recognition 34 (2001) 2467-2476. [10] D.P. Hanson, R.A. Robb, S. Aharon, K.E. Augustine, B.M. Cameron, J.J. Camp, R.A. Karwoski, A.G. Larson, M.C. Stacy, E.L. Workman, New software toolkits for comprehensive visualization and analysis of three-dimensional multimodal biomedical images, Journal of Digital Imaging 10 (2) (Suppl.) (1997) 1-2. [11] R.A. Robb, Three-Dimensional Biomedical Imaging - Principles and Practice, VCH Publishers, New York, NY, (1994). [12] David G. Gobbia,b, Terry M. Petersa, “ Generalized 3D nonlinear transformations for medical imaging: an object-oriented implementation in VTK ”, Computerized Medical Imaging and Graphics 27 (2003) 255–265. [13] X.M. Pardo , P. Radeva , D. Cabello, “ D iscriminant snakes for 3D reconstruction of anatomical organs ”, Medical Image Analysis 7 (2003) 293–310. [14] Jiayin Zhoua, Tuan-Kay Lima, Vincent Chongb, Jing Huang , “Segmentation and visualization of nasopharyngeal carcinoma using MRI ” ,Computers in Biology and Medicine 33 (2003) 407– 424. [15] Omar A. El-Banhawy, Abd El-Hafiz Shehab El-Dine, Talal Amer, “ Endoscopic-assisted midfacial degloving approach for type III juvenile angiofibroma ”, International Journal of Pediatric Otorhinolaryngology 68 (2004) 21-28. [16] Baowei Feia, Corey Kempera, David L. Wilson, “ A comparative study of warping and rigid body registration for the prostate and pelvic MR volumes ”, Computerized Medical Imaging and Graphics 27 (2003) 267–28. [17] Barbara Zitova, Jan Flusser, “ Image registration methods: a survey ” , Image and Vision Computing 21 (2003) 977–1000. [18] Tianhu Lei, Jayaram K. Udupa, Dewey Odhner, La´szlo´ G. Nyu´l, Punam K. Saha,” 3DVIEWNIX-AVS: a software package for the separate visualization of arteries and veins in CE-MRA images”, Computerized Medical Imaging and Graphics 27 (2003) 351–362. [19] Andrew P. Bradleya, and Fred W.M. Stentifordb,” Visual attention for region of interest coding in JPEG 2000 “,J. Vis. Commun. Image R. 14 (2003) 232–250.