Medical science is a typical discipline that greatly relies on cumulative transmission of knowledge and experience, particularly for medical personnel to be familiar with various categories of complicated diseases. For example, medical doctors usually keep memorizing specific patterns shown in pathological photos in an atlas in order to help in diagnosis. Hence, a tool will be helpful if it provides functionality of retrieving relevant photos and information by providing some descriptive words. In this study, the demand is tackled as semantic retrieval of images using annotations from an atlas of ophthalmology. Two-way association rule mining and group association relevance are proposed. Three examples with retrieved photos are given for demonstration.