Diagnostic Tool for Diabetic Retinopathy using Thermal Eye Images

Nithya Rajagopalan, Nirmala K, Vahini M, Nisha S and Rakshitha S


Diabetic Retinopathy (DR) occurs as an effect of Diabetes, affecting the retina causing loss of vision. Worldwide 27% of the diabetic individuals are estimated to have DR, leading to 0.4 million blindness. As the DR population increases every year, early diagnosis is the need of the hour to avoid further progression of the disease thereby preventing blindness. In this work, DR diagnosis is performed using thermal images of the eye. According to earlier research on the eye, the increase in ocular surface temperature caused by dilatation is less in individuals with diabetic retinopathy than in control participants, making it a possible parameter for DR diagnosis. This imaging modality is non-invasive and less time consuming making it an easier diagnostic tool for large scale screening. For this work real time images are acquired from both the eyes of 20 DR participants and 16 control participants, with a total of 72 thermal eye images. The significant features from the pre-processed thermal eye images are extracted using Scale invariant transform and Gabor Transform methods. The extracted features are processed using optimal Machine learning algorithms to classify DR images from normal images. The diagnostic accuracy of the proposed tools is 96%. The results indicate that the IR thermal imaging system could be an efficient method for the non- invasive and non-contact detection of thermal anomalies in the eyes of patients with diabetes.

Modification Method for Facial Skin Temperature Image Considering Facial Artery Structure

Masahito Takano, Kent Nagumo and Akio Nozawa


We proposed a new method to modify facial skin temperature image considering facial artery structure. The proposed modification method facilitates more attention to blood flow information on the thermal face image. The method was adapted to an originally collected dataset to evaluate its performance. The dataset consisted of thermal face images under good or poor subjective health conditions. The result of the evaluation of the discriminant performance indicates that valuable information for health conditions estimation could be extracted from thermal face images by using the proposed modification method.