Evaluation of Evaporative Dry Eye Disease Using Thermal Images of Ocular Surface Regions with DWT and Gabor Transform

Abstract

Tear film instability is a major cause of dry eye (DE) disease. The lack of stability of the tear film may be associated with optical aberrations, visual disturbances, and ocular surface damage. It is clinically important to detect tear instability in DE as the treatment may involve specific measures such as chronic eyelid warming therapy. To achieve this, a practical and rapid method to analyze the relevant features from different regions of the ocular surface in DE will be useful. Thus, in this chapter, efficiency of using the upper half and lower half regions of the ocular surface (cornea + conjunctiva) in the detection of evaporative dry eye is assessed using infrared thermography images. Here, we define the ocular surface as the exposed area of the cornea and the bulbar conjunctiva during natural blinking conditions. Infrared thermography images are acquired from each eye of normal and DE participants. Discrete wavelet transform (DWT) and Gabor transform are used to extract the salient features from the 1st, 5th, and 10th frames of the infrared thermography images after the first blink is subjected to segmentation to obtain the upper half and lower half ocular regions. Each segmented region is decomposed up to three levels using DWT and Gabor transform is performed on the DWT coefficients. Principal component analysis (PCA) is performed on these extracted features to reduce the number of features, and PCA coefficients are ranked using t-value and fed to support vector machine (SVM) classifier. Using the 1st, 5th, and 10th frames of the upper half of ocular region after the first blink, we achieved classification accuracies of (i) 82.3, 89.2, 88.2% for the left eye and (ii) 93.4, 81.5, 84.4% for the right eye, respectively. Similarly, using 1st, 5th, and 10th frames of lower half of ocular regions we achieved accuracies of (i) 95.0, 95.0, 89.2% and (ii) 91.2, 97.0, 92.2% for the left and right eyes, respectively. This study shows that the lower half of the eye is superior to the upper half for the purpose of DE detection using our technique. The proposed algorithm is efficient, simple, and may be employed in polyclinics or hospitals for faster DE assessment.