A Learning by Sample Approach for the Detection of Features in Medical Images.

Abstract

The detection of specific features in medical images is often a key support for diagnosis. Taking advantage of large bases of images where features of interest have been localized by clinicians, a modular system has been developed to spot similar features on new images. Images are scanned through a window, the size of which being previously fitted to the feature of interest. The recognition process involves a coding phase followed by a classification phase. These phases rely on unsupervised and supervised learning respectively for their implementation. Applications in Dermatology and Ophthalmology are presented.