Improved Contour Tracing Method: With applications to fractured bone detection using X-RAY images==
ABSTRACT
editThe rapid advances in imaging technologies depict a future wherein computers would independently diagnose patients and independently recommend treatment methods. However, such high level technology can only be attained by developing medical expert systems that incorporate robust computer vision algorithms. It is widely accepted that careful analysis of object outlines enhances the object recognition process. To date, most of the available image segmentation algorithms hardly achieve perfect segmentation. Since edges typically describe object boundaries, edge detection techniques were among some of the earliest segmentation methods that computer vision researchers used.The major weakness of the edge-based segmentation approach inherently lies in the way edges are defined and computed. Edges merely represent regions of sharp lighting discontinues. This research proposes an enhanced edge detection algorithm that fuses techniques from the Sobel edge detector with those some gradient based edge tracing methods. Such an approach will enable us to trace fine edges, including those of occluded regions. A slightly similar approach has been implemented by Canny. The major weakness of Canny’s method lies on its emphasis on tracing out only the strong edges, and hence the reason why the Canny edge detector often misses out the outlines of some occluded object regions.
INTRODUCTION
editThe advent of the X-RAY technology has enabled medical practitioners to vividly see the bone structure and the human tissues underlying the human skin. More often than not, X-RAY technology becomes very handy when diagnosing internal bone fractures, lung infections and other bodily disorders. Human experts can intuitively classify objects by analyzing data about the objects’ crude outlines (Ballard, 1982; Russ, 1995). Such analytical skills can be exceptionally achieved by computers only if sufficient information about the objects’ outlines could be passed into the computers. One of the surest ways to achieve this is by designing enhanced boundary tracing algorithms. Since edges typically represent object outlines, edge detection techniques were, and are still a prime choice of several computer vision researchers in their attempt to automatically label different objects. However, edge based segmentation was largely unsuccessful for unsupervised systems since edges merely represent regions of lighting discontinuities (Sonka,1998 ) and not necessarily the true object boundaries. In some instances, image edges are generated by image noise(Yi, 2000). As has been noted in , no pervasive generic segmentation algorithm exists as most applications require a careful study of the alternatives and even an invention of new techniques. In one of their recent papers, (Addmore and Vladimir B. Bajic, 2007),despite using a multiple-cue based approach for segmenting gesturing hands, still could not evade the generation of jagged object boundaries and the needless loss of slow moving object regions, which to some extent, disrupts the image recognition process.
Medical images often taken under noisy environments. Quite often, deformable models are used to process medical images (Xu, 2000) since deformable models are robust against noise and the sampling artifacts that commonly affect medical images. However, at times the parameters and the geometry that convincingly describe a deformable model are too expensive to compute. On the other hand, the Sobel edge detector solely depends on the choice of the threshold value when determining which edge pixels to highlight or erode. In fact, the Sobel edge detector is mainly designed to highlight relatively strong edges, and hence it often produces curvatures of different thickness, which, on most occasions, are non-continuous in nature. Such a result makes it almost impossible to accurately describe an object using data about its outlines. In this research we propose an inexpensive algorithm that makes use of both the orientation and the gradient of connected edge-based pixels in order to trace out the boundaries of an object irrespective of the conditions under which it was filmed. Such an algorithm is useful for automating the analysis of image medical images. Idealy the algorithm will first be deployed for identifying fractured bones, but the same concepts will generally be extended to all forms of medical images.
RESEARCH PROBLEM
editComputers and imaging technologies have advanced to a stage whereby computers are increasingly complimenting human effort when diagnosing patients. However, computers have not yet been empowered to independently diagnose patients. It is envisaged that if computers can understand and interpret images, then there is a high likelihood that computers can diagnose all forms of cancers, fractured bones and other forms of diseases that cause the degeneration of the human tissues. This research is attempt to integrate computers’ high processing power with their extraordinary imaging ability in order to produce machines that can artificially diagnose some of the diseases without human aid.
HYPOTHESIS
editIt is assumed that advances in imaging technology will soon enable computers to perfectly interpret medical images. Once computers can read and understand medical images, and if the necessary decision systems are developed, then computers would expertly diagnosis patients. This theory is investigated in this research based on the automatic detection bone fractures.
OBJECTIVES
editOur main goal is to develop a system that automatically identifies the fractured regions of the human bones. The following sub-objectives help us to achieve this goals:
- Design a robust boundary tracing algorithm that correctly identifies object boundaries irrespective of the presence or not of image noise.
- Test for the robustness of the boundary tracing algorithm using X-RAYS of humans’ endoskeleton.
- Develop a mini-decision support system that diagnoses fractured bones.
- Integrate the decision support systems with the image processing algorithms.
- Test for the automatic detection of fractured bones.
LIMITATIONSAND ASSUMPTIONS
editDue to the limited equipment that is available for this research, ordinary human images will be used in place of X-RAYS of the human endoskeleton. We assume that such images adequately represent most of the factors that affect X-RAY images.
EXPECTED BENIFITS
editAutomatic medical diagnostic systems (AMDS) are very important for countries with insufficient numbers of medical practitioners. They also help to alleviate the workload of some of the medical personal who often work at odd hours. Even if the AMDS are not used as autonomous systems, they still would go a long way in aiding accuracy and timely diagnosis of patients. Doctors would have something to compare their results against, and hence enabling optimum patient diagnosis to take place.
REFERENCES
edit- Media:Addmore Machanja, Vladimir B. Bajic; Probabilistic Tracking of Moving Hands in Video Sequences with Applications to Dynamic Gesture Recognition on Mobile Handsets; Southern African Telecommunication Networks Applications Conference (SATNAC), Sugar Beach, Mauritius, ISBN: 978-0-620 39351-5; 2007]]]* Ballard D H, Brown C M; Computer Vision; Prentice-Halldaidb, 1982* Russ J C; The Image Processing Hand Book; CRC Press; 1995
- Sonka M, Hlavac V, Boyle R; Image Processing, Analysis and Machine Vision; Brooks and Cole Publishing; 1998
- Media:Xu C, Pham D L, Prince J L; Medical Image Segmentation Using Deformable Models, Handbook of Medical Imaging -- Volume 2: Medical Image Processing and Analysis, pp. 129-174, edited by J.M. Fitzpatrick and M. Sonka, SPIE Press, May 2000
- Media:Yi L, Tie-Qi C, Jie C, Anthony T, Jiaxin C; An Advanced Machine Vision System for VFD Inspection; 6th International Conference on Manufacturing; 6-8 Sept 2000