A web-accessible mammographic image database dedicated to combined training and evaluation of radiologists and machines

We designed and implemented a web-accessible database entitled MIRaCLe DB (Mammography Image reading for Radiologists' and Computers' Learning Database) that composes a dynamic repository for machines and radiologists training and evaluation. Up to now, 204 mammograms have been collected from 196 patients; they have been classified according to the type of the lesions, the type of the biopsy finding, the type of the mammography finding and the available CADs performance. A user can access the database in two modes: (a) for classification software evaluation and (b) for radiologists' education evaluation. In the mode for classification software evaluation, the user has the ability to query the database and retrieve cases with certain characteristics and certain difficulty. Also, there is the possibility to download the existing cases in order to facilitate the evaluation of a new classifier. In the other mode, the user (radiologist) can be trained in real time through a sequence of presentations and furthermore can be evaluated through different evaluating scenarios. At the duration of evaluation, the user can examine the mammography images through a web-based digital magnifier and process the corresponding image in real time. MIRaCLe DB is the first database that combines the machine and human needs for training and evaluation in mammographic image reading.