Image Processing and Analysis System for Computer Aided Classification of Female Androgenetic Alopecia
Date Issued
2008
Date
2008
Author(s)
Kung, Chien-Wei
Abstract
Traditional methods for grading stages of female alopecia, as described in published references, use scales such as the Ludwig scale or the Savin scale. Although these scales are widely accepted, they are often based on subjective evaluation of an investigator matching a subject with pictures on a pictorial classification scale. Consistency between investigators is low. With the assistance of computers, this study aimed to create a system that objectively analyzes alopecia severity in the hopes of increasing diagnostic accuracy. The purpose of this study was to grade alopecia through the analysis of an image containing a subject’s balding pattern. In order to achieve that, a photograph of the subject was obtained and the image processed. The most important step was segmenting the region of baldness. This study used the level set methods for segmenting region of baldness, thus eliminating subjective evaluation. Then, using this identified region, two features were extracted. One was the width of the bald region; the second was the area of decreased hair density, which was defined as region around the bald patches that still contain hair, albeit much thinner. Evaluation of female alopecia was carried out using the two features and support vector machines. This study included 44 patients suffering from alopecia and 1 normal subject. The standard procedure for acquiring the images was taking two consecutive photographs while the subject stayed immobile. After grooming the hair, the process was repeated again. The rationale behind two consecutive photographs was that differences in brightness may affect the final result of the analysis, so the two sets were used for comparison. This way, the effects can be elucidated. The reason for grooming and repeating the procedure was to see if and how different hairstyles affected the results of the system. Using the features of the width of the bald region and area of decreased hair density, patients rated as type 2 or under on the Lugwig scale were successfully divided into six categories. When taking into account changes in brightness and effect of grooming, the percentage of error is around 15%.
Subjects
Female pattern hair loss
Ludwig scale
Savin scale
level set method
Support Vector Machines
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-97-R95548037-1.pdf
Size
23.53 KB
Format
Adobe PDF
Checksum
(MD5):9f39430bc0d1b74ad2dac068901cd06b