Kenji Suzuki PhD, Presenter: Nothing to Disclose

Hiroyuki Abe MD, Abstract Co-Author: Nothing to Disclose

Feng Li MD, PhD, Abstract Co-Author: Nothing to Disclose

Heber MacMahon MD, Abstract Co-Author: Nothing to Disclose

Kunio Doi PhD, Abstract Co-Author: Nothing to Disclose

The detection of nodules in chest radiographs can be difficult for radiologists as well as computer-aided diagnostic (CAD) schemes when nodules are overlapped with ribs and clavicles. Our purpose was to develop a new image-processing technique for separating ribs and soft tissue in chest radiographs using a multi-resolution massive training artificial neural network (MTANN).

A multi-resolution MTANN consisted of three MTANNs for three different resolution images. The MTANN is a highly nonlinear filter that can be trained with input chest images and the corresponding “teacher” images. A bone image obtained by dual-energy subtraction was used as the teacher image for separation of ribs. The multi-resolution MTANN was trained by using the input and teacher images with three resolutions. After training, the multi-resolution MTANN was able to provide an image similar to a dual-energy bone image. Then, the “bone image” was subtracted from the chest image to produce a "soft-tissue" image where ribs were substantially suppressed. The major advantages of our scheme compared to dual-energy chest radiography are that no specific devices for generating dual-energy images, and no additional radiation dose are required. The database used in this study consisted of 137 chest radiographs with solitary nodules that overlap with ribs or clavicles.

When the trained multi-resolution MTANN was applied to non-training chest images acquired with conventional radiography systems, ribs and clavicles were separated from the chest images substantially. In the soft-tissue images, ribs and clavicles were suppressed remarkably, while the visibility of nodules and lung vessels was maintained. The effect of the rib suppression was evaluated by measuring the contrast of ribs. Results demonstrated that the ribs in chest images almost disappeared, and contrast was reduced to 8% in processed images, while the contrast of nodules was comparable in unprocessed and processed images.

A new image-processing technique for separating ribs and soft tissue in chest radiographs has potential to assist radiologists and to be useful for CAD schemes in the detection of nodules on chest radiographs.

K.D.,H.M.: K.D., H.M. are shareholders in R2 Technology, Inc., Sunnyvale, CA, and K.D. is a shareholder in Deus Technologies, Inc., Rockville, MD.

Suzuki, K, Abe, H, Li, F, MacMahon, H, Doi, K, Separation of Ribs and Soft Tissue in Single Chest Radiographs by Means of Massive Training Artificial Neural Networks.  Radiological Society of North America 2004 Scientific Assembly and Annual Meeting, November 28 - December 3, 2004 ,Chicago IL. http://archive.rsna.org/2004/4411692.html