Research into the segmentation of hand radiographs Research into the segmentation of hand radiographs

Lots of research has been performed into the segmentation of Hand Radiographs. In [1] and [2] we investigated the use of four different methods:

  1. Active Appearance Models;
  2. Canny Edge Detection;
  3. Otsu Thresholding; and
  4. A contouring Algorithm, which had previously not been proposed.
figure 1 amd figure 2 bone

We found none of these methods were robust to the sources of radiation found in hand radiographs (heel effect, under and over exposure). Hence we proposed the Ensemble Outline Detector, which applies the segmentation algorithm to a set of rescaled images and select the outline with the minimum median distance to a set of manual outline, using Dynamic Time Warping.  We found that the using the contour algorithm with the ensemble outline detector performs best, giving a correct outline on approximately 85% ofradiographs.

1 dimensional series hand outline taken from fig 2