Chisako Muramatsu PhD, Presenter: Nothing to Disclose

Tokiko Endo MD, Abstract Co-Author: Nothing to Disclose

Mikinao Ooiwa, Abstract Co-Author: Nothing to Disclose

Misaki Shiraiwa MD, Abstract Co-Author: Nothing to Disclose

Kunio Doi, Abstract Co-Author: Nothing to Disclose

Hiroshi Fujita PhD, Abstract Co-Author: Nothing to Disclose

As digital mammography and monitor reading are becoming common practice, process for storing and retrieving images is much easier and faster. The vast data including diagnostic and pathologic images, clinical data, and treatment data can be effectively used as a reference and teaching materials. We have investigated an automatic retrieval of similar mammographic mass images and their effect on observers’ differential diagnosis.

Observers’ confidence ratings on a mass being malignant were recorded before and after the retrieval of reference images. The study included 48 benign and 50 malignant masses that were visible on two views. The majority of the benign cases were cysts and fibroadenomas, whereas the majority of the malignant cases were invasive ductal carcinomas. For each view, 5 most similar images were retrieved and a similarity map showing the distribution of the database was provided. Observers had to review both views before marking their confidence ratings. Eleven physicians who are routinely reading mammograms and eleven radiological technologists who were trained to read mammograms participated.

Overall, the number of cases that were affected beneficially by the retrieval of reference images was larger than that were affected detrimentally for both groups of observers. There was no trend in pathologies of the beneficially affected cases, although it is difficult to draw a strong conclusion based on the small number of cases. When majorities of the reference images were from the same pathologic group, i.e., benign or malignant, as the inquiry on both views, they were likely to be beneficial. When discrepancy occurred between the views, the observers, especially the technologists, had tendency to follow the suspicious ones and increased their ratings.

It is important to understand how reference images can affect observers’ confidence in order to improve an image retrieval system. Radiologists may effectively use such retrieval systems by understanding the system’s behavior and the characteristics of retrieved images.

Muramatsu, C, Endo, T, Ooiwa, M, Shiraiwa, M, Doi, K, Fujita, H, Retrieval of Reference Cases: For Which Cases Can Similar Cases be Useful?.  Radiological Society of North America 2013 Scientific Assembly and Annual Meeting, December 1 - December 6, 2013 ,Chicago IL. http://archive.rsna.org/2013/13024746.html