Thursday, June 27, 2013

CARS 2013 - Content based retrieval from DICOM images


In a very interesting presentation, dr. Caramella stated that making full use of PACS content goes beyond the images and involves teaching files, content based retrieval, dose information extraction both regarding radiation and contrast media use, and proactive quality assurance.

Dr. Kozuka showed image based retrieval in a clinical database of lung CT images. They defined image features in the lung CT scans with a new approach to dynamically assign weights to image features for each query by also using written findings descriptions to extract information from the database. They showed higher succes rates with their weighted method (71%) when compared to the conventional method (61%).

The next presentation was about user interface design to enhance human interpretation of content based image retrieval by dr. Kubar. Using 8 literature based UI requirements and recommendations. Their novel approach is a graph showing a representation of the data by providing anatomical region nodes and tumour region nodes and the connection between the node that define their relationship. The nodes are linked to the images allow the user to utilize the nodes to jump to the correct image. A full paper on the presenation can be found in the Int. journal of CARS.

BastiĆ£o from Portugal presented on analyzing of efficiency and service quality of digital imaging laboratories. They use DICOM (meta)data to perform knowledge extraction to get quality indicators and evaluate performance. They used a DICOM data mining tool called dicoogle available at www.dicoogle.com which permits extraction of DICOM meta data allowing a variety of analyses based on the DICOM header information.

No comments:

Post a Comment