Detection of Tubes in Radiographs Using Canny Edge Detection and Progressive Hough Transforms
Abstract
An automated method for detecting tubes and catheters in chest radiographs could
improve patient safety and healthcare efficiency by helping radiologists to more quickly and
accurately identify mal-positioned tubes. We propose a method for automatically detecting
tubes that first uses a Canny edge detector for the initial identification of edges, followed by
a windowed variant of the Hough transform, a common line detection algorithm, which is
used to identify potential tube pixels.
Our method employs repeated applications of a parallel-line-specific Hough
transform to the same image with progressively lower thresholds for minimum line length.
Information about the parallel lines identified in the initial Hough transforms is retained and
used to help later, lower threshold runs to more selectively identify potential tube sections. The resultant technique gives an average recall of greater than 80% when measured
by its ability to detect feeding tubes only. The precision rate is low, partially due to its ability
to identify other types of tubes in the image. This could potentially be exploited for tube subclassification
by including other types of tubes in the target set, or by developing additional
algorithms that distinguish between the various types of tubes in the radiograph.
Table of Contents
Introduction -- Review of literature -- Methodology -- Evaluation -- Conclusions and future work
Degree
M.S.