AREA DETERMINATION OF DIABETIC FOOT ULCER IMAGES USING A CASCADED TWO-STAGE SVM BASED CLASSIFICATION

Aim:

             The mainstay of the project is to determine the wound boundary on a foot ulcer image.

 Introduction:

             Doctors base their wound assessment primarily on visual examination and manual measurements performed either directly on the wounds or on high resolution wound images. However, such an ad hoc assessment approach does not establish a comprehensive clinical benchmark. A reliable wound assessment can only be achieved by regularly performing accurate measurements of the wound area, analyzing its colors and the relative sizes of different wound tissues, including proliferation, infected area, slough or necrosis. Due to the lack of consistency, even with the assistance of tools such as PUSH (Pressure Ulcer Scale for Healing), an objective assessment of wound healing rate cannot be ensured. Hence, PC/laptop-based evaluation of foot ulcers using computer vision and image processing techniques represents an improved approach to accurate chronic wound assessment. Automatic detection of foot ulcer size and tissue composition is especially useful for both clinicians and diabetic patients to monitor the wound healing status and for more effective wound care.

Proposed scheme:

           Recognition scheme based on machine learning was applied to wound tissue classification and indirect wound area determination (by grouping all the regions classified as one type of wound tissue). Generally, these approaches consist of following three key steps: Step 1: image segmentation; Step 2: feature extraction within each segment; and Step 3: classifier training on a large number of features from either the wound or non-wound segments.