Automatic method for classifying 3D images

Method for extracting information from three-dimensional images, by identifying general properties regardless of orientation. The developed algorithm allows to extract and classify information from any image and can be applied in diagnostic imaging by extracting feature from PET images, for example.

How does it work?

Extracting robust features from images is often affected by orientation and normalization problems that often compromise the generality and performance of the method used. These problems become particularly relevant for low-resolution, low-contrast 3D images. The proposed method, developed and optimized for brain PET imaging, is a general tool for automatic characterization, classification and sorting of any type of image. Such a method is applicable to any homogeneous dataset representing a population of objects described by an n-dimensional scalar matrix. ELBA, the efficient algorithm behind the method, is able to extract the relevant properties from the image by calculating features based on isosurfaces and offers a valid alternative to intensity-based methods.

Applications

• Evaluation of amyloid PET in clinical trials;
• Recognition of supernovae (astronomical images);
• Object recognition;
• Characterization and classification of images (also in time series);
• Data mining;
• Night vision improvement;
• Mole classification and monitoring.

Advantages

• It provides stable features, extractable from any n-D matrix;
• Innovative computer vision approach;
• It provides information complementary to that provided by standard methods;
• It avoids image normalization;
• Extremely flexible