<p>Meibomian gland dysfunction (MGD) is a common cause of dry eye disease. It occurs when oil producing glands do not provide enough oil for eye greasing, due to fewer glands, blockage or inflammation. In this paper, the main idea is to segment the meibomian glands present in eyelids, which have complex structures and location variability. They help to prevent evaporation of tears and thus can further aid in detecting MGD by providing the details of the region occupied by the glands. A new deep learning model based on modified U-Net with proposed attention mechanism has been developed for automatic segmentation of meibomian glands from the infrared images. Adding the proposed attention layers to the modified U-Net architecture have improved the segmentation performance by effectively highlighting the glands regions. The proposed attention mechanism is based on the additive activation attention gate, which combines the activation feature maps from the skip connection and the gating signal. It is then followed by addition of original feature map from skip connection. This attention mechanism helps in extracting glands information from the infrared images. The performance of the developed model has been evaluated using Dice Similarity Coefficient, Intersection over Union, Sensitivity, and Specificity. MGD-1K dataset has been used for the experiments to validate the efficacy of our developed model. In all experiments, the proposed model achieved high segmentation accuracy and outperformed existing methods.</p>