Home > Products > An Object-Oriented Method for Segmentation of Bone in Digital Hand X-Ray Image

An Object-Oriented Method for Segmentation of Bone in Digital Hand X-Ray Image

Company : CompuMed Inc

UMDC Code : 101172

The purpose of this paper is to evaluate an image processing system that segments the middle phalangeal bones of hand x-ray images. This system was imbedded in the PC-based software the Automated OsteoGram (CompuMed, Inc., Los Angeles, CA) which utilizes radiographic absorptiometry (RA) for calculating BMD. An important input for phalangeal BMD calculations is the contour of target bones; hence, the reliability of the calculations depends on accurately detecting the contours of all regions of interests including bones. The segmentation system uses an object-oriented approach comprising of several sub-stages designed to extract target objects. These objects are organized hierarchically beginning with the most general objects to be extracted, and proceeding in a succession of sub-stages to the most specific and final target object. Each sub-stage level has a collection of methods and attributes specific to that particular sub-stage target object. Some of these methods can be re-used in other sub-stages thereby minimizing the developmental effort. Because the system is object-oriented in its approach it has three distinct advantages. First, it uses feedback loops at each of the sub-stages for QA verification, which minimizes the propagation of error. Second, it allows for operator intervention at key stages to assist in segmenting misidentified target objects. Third, new and improved methods and attributes developed for a sub-stage target object can easily be integrated into the system and tested without restructuring the main design.The system was tested on a clinical study that included 248 hand x-rays images with a total of 4 final target objects per image. The target objects were the middle phalangeal bones of the 2nd, 3rd, and 4th digits, in addition to a proprietary reference aluminum wedge. The hand x-rays were digitized using a medical-imaging quality desktop scanner. The digitized image from each x-ray exposure spanned a rectangular area that included the 4 target objects of interest. The general orientation of the 4 objects was vertical, with the fingers pointing upward. All 992 objects were successfully extracted from the 248 images. Of the 992 objects, 92% of them were automatically extracted while 8% required operator intervention by means of mouse-clicking on the misidentified object.In conclusion, a new segmentation system was developed and implemented using an object-oriented approach. The system was successfully tested on digital hand x-rays to segment three phalangeal bones for a BMD software.

Product URL : http://compumed.ning.com/page/an-objectoriented-me

Last Updated : 03/16/2013

Please provide the following information and a representative of CompuMed Inc will contact you to assist with any questions you might have.