Andreas Burner, Abstract Co-Author: Nothing to Disclose

Rene Donner, Abstract Co-Author: Nothing to Disclose

Marius Erik Mayerhoefer, Abstract Co-Author: Nothing to Disclose

Georg Langs, Presenter: Nothing to Disclose

In hospital routine, thousands of medical images are processed on a daily basis. These images are typically inspected once during diagnosis and then remain unused. It is desirable to access the rich information in those images and the corresponding reports efficiently during diagnosis of new cases. We propose and evaluate an efficient method for content-based image retrieval (CBIR) that supports physicians during their diagnosis by finding most similar anomalies in the PACS system. We evaluate the system with data of various lung diseases.

Results indicate that region-based retrieval is able to find corresponding anomalies based on imaging data. Region-based retrieval outperforms full volume retrieval. Future use of the method is the support of diagnosis by making efficient access to previously diagnosed cases possible.

Results show that the average recognition rate for correctly retrieving the three top rated images is 67%, which outperforms full volume retrieval , 24%, significantly.

The proposed method automatically obtains a texture vocabulary from training data by extracting the dominant structure (tissue structure, trabecular structure, etc.) using an extension of Local Binary Patterns (LBP). The texture properties are captured via descriptor histograms of texture bags. During retrieval, a physician marks a region of interest in a query case. The algorithm retrieves similar texture regions in the entire data set. We evaluate the performance of our method on a set of 21 healthy and emphysema-diseased lung cases in a 7-fold cross-validation by the recognition of pathologies.

Burner, A, Donner, R, Mayerhoefer, M, Langs, G, An Efficient Region-based Image Retrieval of Anomalies in Medical Imaging Data.  Radiological Society of North America 2012 Scientific Assembly and Annual Meeting, November 25 - November 30, 2012 ,Chicago IL. http://archive.rsna.org/2012/12037184.html