Please use this identifier to cite or link to this item:
|Title:||Bioimage informatics approach to automated meibomian gland analysis in infrared images of meibography||Authors:||Celik, T.
Dry eye syndrome
Meibomian gland segmentation
|Issue Date:||Oct-2013||Citation:||Celik, T.,Lee, H.K.,Petznick, A.,Tong, L. (2013-10). Bioimage informatics approach to automated meibomian gland analysis in infrared images of meibography. Journal of Optometry 6 (4) : 194-204. ScholarBank@NUS Repository. https://doi.org/10.1016/j.optom.2013.09.001||Abstract:||Background Infrared (IR) meibography is an imaging technique to capture the Meibomian glands in the eyelids. These ocular surface structures are responsible for producing the lipid layer of the tear film which helps to reduce tear evaporation. In a normal healthy eye, the glands have similar morphological features in terms of spatial width, in-plane elongation, length. On the other hand, eyes with Meibomian gland dysfunction show visible structural irregularities that help in the diagnosis and prognosis of the disease. However, currently there is no universally accepted algorithm for detection of these image features which will be clinically useful. We aim to develop a method of automated gland segmentation which allows images to be classified. Methods A set of 131 meibography images were acquired from patients from the Singapore National Eye Center. We used a method of automated gland segmentation using Gabor wavelets. Features of the imaged glands including orientation, width, length and curvature were extracted and the IR images enhanced. The images were classified as 'healthy', 'intermediate' or 'unhealthy', through the use of a support vector machine classifier (SVM). Half the images were used for training the SVM and the other half for validation. Independently of this procedure, the meibographs were classified by an expert clinician into the same 3 grades. Results The algorithm correctly detected 94% and 98% of mid-line pixels of gland and inter-gland regions, respectively, on healthy images. On intermediate images, correct detection rates of 92% and 97% of mid-line pixels of gland and inter-gland regions were achieved respectively. The true positive rate of detecting healthy images was 86%, and for intermediate images, 74%. The corresponding false positive rates were 15% and 31% respectively. Using the SVM, the proposed method has 88% accuracy in classifying images into the 3 classes. The classification of images into healthy and unhealthy classes achieved a 100% accuracy, but 7/38 intermediate images were incorrectly classified. Conclusions This technique of image analysis in meibography can help clinicians to interpret the degree of gland destruction in patients with dry eye and meibomian gland dysfunction. © 2013 Spanish General Council of Optometry.||Source Title:||Journal of Optometry||URI:||http://scholarbank.nus.edu.sg/handle/10635/109947||ISSN:||18884296||DOI:||10.1016/j.optom.2013.09.001|
|Appears in Collections:||Staff Publications|
Show full item record
Files in This Item:
There are no files associated with this item.
checked on Aug 18, 2019
checked on Aug 16, 2019
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.