Content-based image retrieval has been proposed as a viable alternative to text-based image retrieval. Mainstream content-based image retrieval methods, however, do not satisfy the complex demands created by biomedical images. The success of content-based biomedical image retrieval hinges on the consideration of domain-specific image characteristics. We strive for image retrieval with a high-level of content understanding, a high degree of query completion and integral user-computer interaction. Departing from the view that user interaction is mandatory for query completion and that query completion is mandatory for content-understanding, we explore the prospects of incrementally and interactively learning concepts during image retrieval for precise formalization of the user's perception of image information. As concepts closely relate to a user's preferences and subjectivity, and in addition can be considered as generalizations of a query instance, they are expected to allow more accurate and reliable image retrieval. In this paper, we discuss a method for concept-based image retrieval by population-based incremental learning of multi-feature image segmentations using query-by-example and relevance feedback. Image retrieval is demonstrated using digitized vertebral X-ray images from the NHANES II database.