A New Methodology for Diagnosis of Appendicitis Using Sonographic Images in Image Mining

Balu Ramu
Ph.D Research Scholar Department of Computer Applications Bharathiar University, Coimbatore - 46 +91 98420 12361

Devi Thirupathi
Associate Professor and Head i/c Department of Computer Applications Bharathiar University, Coimbatore - 46 +91 97900 04351

rvkbalu@yahoo.co.in ABSTRACT
Appendicitis is a most common abdominal condition that necessitates emergency surgery. Sonographic images become very handy in diagnosing the condition of the appendicitis in a patient. Sonographic image mining technique is a dynamic, noninvasive, rapid and inexpensive technique. This research work proposes a new methodology based on image mining for diagnosing appendicitis by considering sonographic images as input. For case study, a reputed hospital in India has been considered and 12 patients of that hospital are observed over a period of four months. The sonographic images of these patients are used as input for the study and analysis. The results obtained out of the proposed method based on sonographic images are compared with the real results, results obtained through euclidean distance methods and geodesic distance. This paper describes a brief literature on tests for appendicitis, proposed algorithm, the experimental method and results of sonographic image mining.

tdevi5@gmail.com
prone to the condition. But it may happen to anyone without warning, which is why NASA is considering preventative appendectomies for astronauts before they embark on future space missions [8]. A new methodology for diagnosis of appendicitis has been proposed and it uses sonographic images and image mining technique. Image mining is explained in next section. Section III discusses various existing methods for appendicitis, details about the euclidean and geodesic distances; Section IV describes proposed algorithm; Section V shows the results and performance evaluation. Finally, Section VI concludes the paper.

2. IMAGE MINING
Image mining denotes the synergy of data mining and image processing technology to aid in the analysis and understanding in an image-rich domain. It is an interdisciplinary endeavor that draws upon the expertise in computer vision, image processing, image retrieval, data mining, machine learning, database and artificial intelligence [2]. Broadly speaking image mining deals with the extraction of implicit knowledge, image data relationship, or other patterns not explicitly stored in the images and between image and other alphanumeric data [3]. Image mining is a future trend, although it is a new research area, its development shows great potential, as far as now its application has expanded to various domains and gained good results. Mining on medical images is to acquire valuable knowledge and modes, which can later be used for discovering abnormal situations not consistent with the previous common modes. This can act as a reference and help doctors diagnose diseases. Image mining is a technique which handles the mining of information, image data association, or additional patterns not unambiguously stored in the images [4, 5].

Categories and Subject Descriptors

[Image Processing]: Enhancement [Data Mining]: Image Applications Image mining

General Terms
Measurement, Performance, Design, Experimentation

Keywords
Image Mining, Euclidean Distance, Geodesic Distance, Ultrasound, Appendicitis

1. INTRODUCTION
Appendicitis is a painful inflammation and infection of the appendix. The primary cause is obstruction of the appendiceal lumen. An important cause of abdominal pain in children, appendicitis often requires emergency surgery [7]. Appendicitis is a dangerous condition and it usually occurs when the appendix becomes infected and blocked by a build-up of thick mucus, faeces or some foreign object such as parasitical worms. Global statistics show that one in 15 people will develop appendicitis in their lifetime, and people between the ages of 10 and 30 are most
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee. CCSEIT-12, October 26-28, 2012, Coimbatore [Tamil nadu, India] Copyright 2012 ACM 978-1-4503-1310-0/12/10$10.00.

3. EXISTING METHODS
The common test done for appendicitis includes blood test, urine test and sometimes plain abdominal x-ray, ultrasound and CT scan. Most patients suspected of having appendicitis would be asked to do a blood test. 50% of the times, the blood test may be normal. So it is not fool proof in diagnosing appendicitis. Urine test in appendicitis is usually normal. It may however show blood if the appendix is rubbing on the bladder, causing irritation. A urine test or urinalysis is compulsory in women, to rule out pregnancy in appendicitis, as well to help ensure that the abdominal pain felt and thought to be acute appendicitis is not in fact, due to ectopic pregnancy.

717

In 10% of patients with appendicitis, plain abdominal x-ray may demonstrate hard formed faeces in the lumen of the appendix. The normal appendix is not frequently visible on ultrasound scan. If seen, it is most likely that the appendix is inflamed. Ultrasound scan may demonstrate free fluid around a swollen appendix. An outer thickness of greater than 6mm on scan is also highly suggestive of inflammation of the appendix. CT scan or computed tomography scan is a specialised form of x-ray. The patient is passed through a big doughnut-like machine and x-rays 400 times the normal is used to look at the body in slices of about 6millimeter thickness [6].

surface along the geodesic paths produces better results [9, 10] and minimum distortion parameterization or remeshing of 3D models are based on the knowledge of geodesic distances [11-13]. The applications include isometry-invariant shape classification [14, 15], skinning [16], medical imaging and geophysics [17]. Fig. 3. 1. Shows difference between Euclidean Distance and Geodesic Distance on a circle.

3.1 Euclidean Distance

Euclidean distance is the distance between two points in Euclidean space. Take two points P and Q in two dimensional Euclidean spaces. This describes P with the coordinates (p1, p2) and Q with the coordinates (q1, q2). Now construct a line segment with the endpoints of P and Q. This line segment will form the hypotenuse of a right angled triangle. The distance between two points p and q is defined as the square root of the sum of the squares of the differences between the corresponding coordinates of the points; for example, in two-dimensional Euclidean geometry, the Euclidean distance between two points a = (ax, ay) and b = (bx, by) is defined as:

Fig.3.1. Euclidean and geodesic distances on a circle [17]

d a, b

p1

q1

p2

q2

This algorithm computes the minimum Euclidean distance between a column vector x and a collection of column vectors in the codebook matrix cb. The algorithm computes the minimum distance to x and finds the column vector in cb that is closest to x. It outputs this column vector, y, its index, idx, cb, and distance, the distance between x and y.

p1

q1

d a, b p2

q2

|p

q|

pn

qn

Geodesic has shortest paths, and perturbation of a geodesic curve will increase its length. The minimal length path between two points on the surface is the minimal geodesics connecting those points and minimizes the euclidean distance on a surface. A geodesic distance between two vertices of a triangle mesh surface can be computed using a shortest path algorithm on the mesh graph with an edge is its length. The Dijkstra algorithm [18], can compute path lengths very quickly, but produce paths quite different from true geodesics and the paths they create pass only through the mesh vertices. The research on geodesic computation on a surface focuses on solving the eikonal equation [19]. Kimmel and Sethian [20] proposed an optimal time algorithm for computing the geodesic distances and extracting shortest paths on triangulated manifolds using the Fast Marching Method [21].

In one dimension, the distance between two points, x1 and x2, on a line is simply the absolute value of the difference between the two points [3] .

In two dimensions, the distance between P = (p1,p2) and q = (q1,q2) is:

X2

X1

|X2 p2

X1 |

The geodesic paths between two vertices in O(n log n) time per path by numerically solving the eikonal equation (n is the number of vertices in the mesh). Mitchell et al presented an algorithm for determining the shortest path between a source and a destination on an arbitrary polyhedral surface, and seeking to approximate distance maps on a parametric surface and the eikonal equation on a discrete grid obtained by sampling the parametric domain. An efficient O(n) numerical algorithm for first-order approximation of geodesic distances on parametric surfaces, where n is the number of points on the surface [22].

3.2 Geodesic Distance

p1

q1

The geodesic distances on a three dimensional surface, is important in many fields of computer graphics and geometric modelling. Normally, geodesic paths on a surface are critical in surface segmentation and editing methods since cutting the

Fig. 3.2. Difference between Euclidean Distance and Geodesic Distance [23]

718

4. PROPOSED ALGORITHM
To review our method authors described below, the algorithm is as follows.

5. EXPERIMENTS AND RESULTS

An experiment has been conducted on sonographic scan image based on the proposed algorithm as shown in Fig 4.1. The sonographic image is first captured as an input image, and then decorrelation stretch has been used to remove the noise component from that image. Decorrelation stretch has been used to find the edge feature in the ultrasound scan image. Using this technique the noise component is removed from the image. 5.1 Sample Appendicitis

Sonographic Images

Image Enhancemen t

Image Segmentatio n

Image Extraction

Image Classification

Measure of Distance

Test Results

Fig. 4.1 Phases in proposed algorithm for diagnosis of appendicitis

Fig. 5.1 Normal Image vs Appendicitis Image

Step1: Image Capturing Image Capturing is a process to acquire the digital image into Matlab. Ultrasound image from the database is obtained through imread command in matlab. The exact path of the image should be given as the argument to imread command. In matlab, captured image is displayed using imshow command by passing a variable as the argument and the image can also be displayed in the image viewer using imtool command. Fig. 5.2 Image Enhancement Step2: Image Enhancement The image noise is most apparent in image regions with low signal level such as shadow regions or under exposed images. In conventional enhancement techniques such as low pass filter, median filter, gabor filter, gaussian filter, prewitt edge-finding filter, normalization method are employable for this work. Step3: Image Segmentation Segmentation subdivides an image into its constituent regions or objects. To find out the transformation between two images precisely, they should be preprocessed to improve their quality. These images are too noisy or blurred, they should be filtered and sharpened and for removing the unwanted portions of the image, region-based segmentation algorithm has been applied. Step4: Image Extraction Image extraction from the region of interest (ROI) characteristics such as size, shape, density and smoothness etc. Step5: Image Classification Linear Discriminant Analysis (LDA) is traditional method for classification. It is construct decision boundaries by optimizing certain criteria to classify cases into one of mutually exclusive classes. Step5: Similarity Measures In this paper, the appendicitis is detected using the distance measure in order to confirm the patient is diagnosed with appendices. The distance is predominant to bring out the diagnosis. Step6: Suggestion of Diagnosis Fig. 5.4 Image Extraction 719 Fig. 5.3 Image Segmentation

A sonologist has been consultated and the results obtained from the sonologist on the same 12 samples reveal that 12 patients are affected by appendicitis. A comparison of the results obtained from the proposed system with the results obtained from the sonologist has been done and the results are as shown in table 6.2. Table 5.2 : Comparison of Proposed Method and Real Result .2

Size 6 mm
Fig. 5.5Appendicitis Extraction using Proposed Method Appendicitis

Results from Proposed method 12 0

Results from Sonologist 12 0

< 6 mm

5.2 Results and Discussions

The study period was from Jan 2012 Apr 2012 Ultrasound 2012. imaging was done in all patients. Acute appendicitis of sonographic image findings in 12 patients was done. In the experimental study the total number of 12 instances studied where in the range of 16 42 years. Male and female sex ratio is in the range of 1: 5. The column chart clearly shows the sex . distribution for these 12 images.

5.3 Performance Evaluation .3

All these features were extracted for this particular domain. Authors used the features extracted from the training image, on the testing images. Hence in essence, authors have developed a system, which is trained once and then applies the same technique to other images. The goodness of these features can be judged by certain evaluation criteria. The second set of images was used for comparison and validation. The validation images have the same size and hence can be compared with the extracted images. For a two-class problem, there can be 4 class possible outcomes of a prediction [17]. The outcomes are True Positives (TP), True Negatives (TN), False Positives (FP) and False Negatives (FN). Based on commonly used performance ves measures in two statistical measures were computed to assess system performance namely recall and precision. Table Table5.3 shows the values of recall and precision of each diagnosed images using the proposed system.

10

10 5 0
No. of Patients

Male
2

Female

Graph 5.1 Patients Ratio

Precision: Defined as the diagnosed images, which is relevant.

The images are classified in two different sizes based on the thickness of appendicitis with greater than 6 mm and less than 6 mm. The proposed system is tested with 1 images and the 12 results obtained are: out of 12 images, 12 instances show thickness measured as greater than 6 mm.

Recall: Defined as the diagnosed images versus all database

images.

Table 5.3 Recall and precision using the proposed system Table 5.1 : Comparison of Proposed Method, Real Result and .1 Geodesic Distance
S NO AGE SEX Real Result Euclidean Geodesic

Method Euclidean Real Result Geodesic

Precision 1 1 0.5

Recall 1 1 0.5

1 2 3 4 5 6 7 8 9 10 11 12

24 18 41 36 44 53 48 34 32 21 28 18

F F M M F F F F F F F F

+ + + + + + + + + + + +

+ + + + + + + + + + + +

1.2 1 0.8 0.6 0.4 0.2 0

Precision Recall

Euclidean

Real Result

Geodesic

Fig. 5.6 Recall and precision using the proposed system 720

6. CONCLUSION AND FUTURE WORK

This research paper proposed a new methodology for diagnosis of appendicitis using sonographic images and image mining techniques. The results obtained out of the proposed method based on sonographic images are compared with the real results, results obtained through euclidean distance methods and geodesic distance. The precision and recall obtained are 1. The developed system is expected to provide valuable support to the physicians in decision making while diagnosing the patients for appendicitis. This work can be further extended to include the module on automatic identification of the start and end points of measurement using image processing and automation techniques. [1] Funkhouser, Kazhdan, Shilane, Min, Kiefer, Tal, Rusinkiewicz, and Dobkin. Modelling by example, ACM Trans. Graph., 23(3):652663, 2004 Zhou, Synder, Guo, and Shum, Iso-charts: stretchdriven mesh parameterization using spectral analysis, In SGP 04: Proceedings of the 2004 Eurographics/ ACM SIGGRAPH symposium on Geometry processing, pages 4554, New York, NY, USA Peyre and Cohen, Geodesic remeshing using front propagation, Int. J. Comput. Vision, 69(1):145156, 2006 Sifri, Sheffer, and Gotsman,Geodesic-based surface remeshing, In Proc. 12th International Meshing Roundtable, pages 189199, 2003. Elad and Kimmel, On bending invariant signatures for surfaces, IEEE Transactions on Pattern Analysis and Machine Intelligence, 25(10):12851295, 2003. Hilaga, Shinagawa, Kohmura, and Kunii., Topology matching for fully automatic similarity estimation of 3d shapes, In SIGGRAPH 01: Proceedings of the 28th annual conference on Computer graphics and interactive techniques, pages 203212, New York, NY, USA. Sloan, Charles, Rose, and Cohen, Shape by example, In I3D 01: Proceedings of the 2001 symposium on Interactive 3D graphics, pages 135 143, New York, NY,USA. Sethian and Popovici, 3-d traveltime computation using the fast marching method, Geophysics, 64(2):516523, 2006. Skiena, The algorithm design manual, Springer, 2000. Bronstein, Bronstein, Devir, and Kimmel, Parallel algorithms for approximation of distance maps on parametric surfaces, Technical Report, 2007. Kimmel and Sethian, Computing geodesic paths on manifolds, In Proceedings of National Academy of Sciences, pages 84318435, 1998. Sethian, A fast marching level set method for monotonically advancing fronts, In Proc. Natl. Acad. Sci., volume 93, pages 15911595, February 1996. Mitchell, Mount, and Papadimitriou, The discrete geodesic problem, SIAM J. Comput.,16(4):647668, 1987.

[2]

[3]

[4]

[5]

[6]

[7]

[8]

[9] [10]

[11]

[12]

[13]

721