Bone suppression in chest radiographs by means of anatomically specific multiple massive-training ANNs combined with total variation minimization smoothing and consistency processing

Abstract

Most lung nodules that are missed by radiologists as well as computer-Aided detection (CADe) schemes overlap with ribs or clavicles in chest radiographs (CXRs). The purpose of this study was to separate bony structures such as ribs and clavicles from soft tissue in CXRs. To achieve this, we developed anatomically (location-) specific multiple massive-training artificial neural networks (MTANNs) which is a class of pixel-based machine learning combined with total variation (TV) minimization smoothing and a histogram-matching-based consistency processing technique. Multi-resolution MTANNs have previously been developed for rib suppression by use of input CXRs and the corresponding teaching images for training. Although they were able to suppress ribs, they did not suppress rib edges, ribs close to the lung wall, and the clavicles very well because the orientation, width, contrast, and density of bones are different from location to location and the capability of a single set of multi-resolution MTANNs is limited. To address this issue, the anatomically specific multiple MTANNs developed in this work were designed to separate bones from soft tissue in different anatomic segments of the lungs. Each of multiple anatomically specific MTANNs was trained with the corresponding anatomic segment in the teaching bone images. The output segmental images from the multiple MTANNs were merged to produce a whole bone image. Total variation minimization smoothing was applied to the bone image for reduction of noise while edges were preserved. This bone image was then subtracted from the original CXR to produce a soft-tissue image where the bones were separated out. In order to ensure the contrast and density in different segments were consistent, a histogram-matching technique was applied to the input segmental images. This new method was compared with the conventional MTANNs by using a database of 110 CXRs with pulmonary nodules. Our new anatomically (location-) specific MTANNs separated rib edges, ribs close to the lung wall, and the clavicles from soft tissue in CXRs to a substantially higher level than the conventional MTANNs did, while the visibility of lung nodules and vessels was maintained. Thus, our image-processing technique for bone-soft-tissue separation by means of our new anatomically specific multiple MTANNs would be potentially useful for radiologists as well as for CAD schemes in detection of lung nodules on CXRs.