In this paper, we propose a medical image retrieval scheme which could be of immense help to the medical practitioners. The basic idea is to input a 2D Brain Magnetic Resonance (MR) query slice and output visually similar slices pertaining to the MR image database. Query slice pertaining to a new case can be related to slices belonging to previous case histories in the stored database. It would prove to be a useful diagnostic aid for the medical expert and could also serve as a teaching aid for students and researchers in medical field. Similar pathological cases can also aid in treatment apart from diagnosis. The challenge then is to device an algorithm which is fast, accurate and could be run on low power devices that are installed in clinics and hospitals. To this end, we propose a scheme consisting of a) MR image slice representation by only a few visually significant feature points and b) Similarity distance metric used for retrieval taking into account the spatial information of the local structures.