| 2028 Computerized Boundary Detection of Oral Mucosal Lesions | ||
|
A. CHODOROWSKI, Göteborg University, Sweden, and U. MATTSSON, Central Hospital, Karlstad, Sweden Objectives: Digital image recording of the oral cavity is now a standard technique in dental clinics and produces a huge amount of images. To efficiently analyse this material we need procedures to automatically extract the relevant information. The purpose of the present study was to develop a computerized method for automatic segmentation of digital color images of oral mucosal lesions, with the specific aim to detect the boundaries of these lesions. Methods: The clinical appearance of the mucosal lesions was recorded using a 3 CCD digital color camera. The test material consisted of 51 images of oral leukoplakia and 74 cases of oral lichenoid reactions (atrophic, plaqueformed and reticular). The heterogenous appearance of many mucosal lesions makes it difficult to find a fully automated segmentation method. The aim was therefore to elaborate a semi-automatic solution, where the knowledge and experience of the clinician is supported by the computer. The first step consisted of detecting lesion-specific features which could be used for objective function (cost-function) design and for the subsequent segmentation process. The final lesion outlining was performed using a real time user-steered live wire technique with dynamic programming. Results: The developed semi-automatic segmentation technique reduced the registration time and the number of user interventions was low. It reduced the variability in outlined contours, when compared to the manual delineation. Image features obtained from semi-automatic and manual delineation were comparable. Conclusions: The technique eliminates the tedious manual lesion outlining and improves the segmentation quality. The developed method can be used in clinical practice for delineation and storing objects of interests in digital images in odontological databases. The segmentation will simplify extraction of relevant features for further processing in computer aided diagnostic systems to support detection of potentially precancerous lesions. | ||
| Seq #210 - Digital Imaging & Models/Motion Assessment/Colorimetry/Non-imaging Diagnositcs 3:45 PM-5:00 PM, Friday, 27 June 2003 Svenska Massan Exhibition Hall B | ||
|
Back to the Diagnostic Systems Program
| ||