A novel algorithm has been developed which makes it possible to reliably recognise specific objects in images, even if the objects have a wide variability in their appearance. Researchers at the University of Zurich say the method is particularly well suited for object recognition in histological slides or radiological images.
Conventional image analysis software based on threshold-methods is well established in the medical and life science field. Unfortunately, such software only inadequately detects objects that show a vast variability in their appearance. The results are therefore not sophisticated enough for the daily work of scientists and medical doctors, or for automated diagnosis.
A new image recognition algorithm has been developed which is based on a network of classifier modules. The algorithm is able to successfully detect objects with a vast variability in their appearance, such as objects without well-defined boundaries or diffuse objects. The usability of the method has been demonstrated by a working prototype which has been successfully applied to detect plaques, connective tissue and blood cells in histological samples and axons in electron microscope images. Examples can be accessed under http://www.pathol.uzh.ch/ir (user: irtest password: neuronlike).
The image recognition technique is expected to be particularly useful for medical applications and in life sciences.