Automated analysis of medical images for diagnostic and data mining purposes are being widely developed. An essential step in high throughput processing is to know what anatomy is included in a given image so that the right analysis can be triggered automatically.
CVIB PhD student Xiaoyong Wang, working with faculty and staff scientists, has developed an image classification technique using a feature vector based on image subsampling. Machine learning was performed using a large clinical trial database of labeled images. The approach is designed for robustness and generalizability to any imaging modality (CT, MRI, PET, etc) and anatomy type. This is a key competitive advantage over the small number of other papers in this area.
Xiaoyong presented this working at the SPIE Medical Imaging Conference in San Diego, California. His experimental results show high classifier accuracy and will be published in the conference proceedings. The algorithm and software have been registered with the UCLA Office of Intellectual Property for potential commercial licensing.