Spatio-temporal Saliency for Microscopic Medical Data

Manual tracking and analysis of medical images have become tedious and impractical because of the advancement in medical modalities and the amount of medical streams being produced daily. The analysis of the motion of the immune system can be used for diagnostic purposes. A spatio-temporal saliency map is proposed to detect and segment the moving immune system from microscopic video data. The temporal saliency is computed based on tracklets computed using extended Lucas Kanade technique and their clustering is performed based on coherence. The spatial saliency is computed by feature extraction. Experimental results demonstrate the significance of the proposed spatial and temporal saliency maps fusion. In addition, a database for moving the immune system (DMI) has also been developed.