Automatic rib detection in chest radiographs

Computer aided diagnosis in chest radiography

“…Only papers with a clear description of the algorithm have been included. Note that several studies also detect (parts of) the contours of the lung fields [55,201,278] but do not evaluate these methods. The classical approach, which is followed by most researchers, is the following.…”

“…Wechsler [256,257] used edge detection and thresholding to find rib border candidates and proposed a slightly more elaborate rib model, consisting of a [55] used the derivatives of vertical profiles in the image to find pixels on rib boundaries and fitted parabolas to these. A general problem emerged from all these studies: not all detected edge pixels and extracted rib border candidates were actually located on a true rib border.…”

Semiautomatic Detection of Scoliotic Rib Borders From Posteroanterior Chest Radiographs

“…Also, techniques using global spatial filtering and rib reconnection through rule-based reasoning [2,3,4,5,7] would often miss the edges at rib crossings near the rib cage border, due to high overlapping and locally reduced contrast. In addition, methods based on vertical profile analysis [4,5,8,9] would also encounter major Semi-Automatic Detection of Scoliotic Rib Borders From Postero-anterior Chest Radiographs Frédéric Plourde, Farida Cheriet, Associate Member, IEEE, Jean Dansereau S limitations, as they greatly rely on assumptions that the ribs are strictly horizontally-oriented (in their dorsal portions) and that the intercostal space is approximately constant from one rib to another, which is clearly not the case with scoliotic ribs, as observed in [10]. Yet another approach [6] makes use of active contours (snakes) in order to better delineate actual rib borders, but still relies on the Hough transform to initialize the algorithm.…”

“…Existing rib detection methods are not suitable for detecting scoliotic ribs for the following reasons. First, because of their large variability in shape and curvature between different patients, scoliotic ribs would not be properly detected using parametric curve fitting techniques [2,3,4,5] or parametric curve-searching algorithms such as the modified Hough transform [3,6]. Also, techniques using global spatial filtering and rib reconnection through rule-based reasoning [2,3,4,5,7] would often miss the edges at rib crossings near the rib cage border, due to high overlapping and locally reduced contrast.…”

“…First, because of their large variability in shape and curvature between different patients, scoliotic ribs would not be properly detected using parametric curve fitting techniques [2,3,4,5] or parametric curve-searching algorithms such as the modified Hough transform [3,6]. Also, techniques using global spatial filtering and rib reconnection through rule-based reasoning [2,3,4,5,7] would often miss the edges at rib crossings near the rib cage border, due to high overlapping and locally reduced contrast. In addition, methods based on vertical profile analysis [4,5,8,9] would also encounter major Semi-Automatic Detection of Scoliotic Rib Borders From Postero-anterior Chest Radiographs Frédéric Plourde, Farida Cheriet, Associate Member, IEEE, Jean Dansereau S limitations, as they greatly rely on assumptions that the ribs are strictly horizontally-oriented (in their dorsal portions) and that the intercostal space is approximately constant from one rib to another, which is clearly not the case with scoliotic ribs, as observed in [10].…”