52 IEEE TRANSACTIONS ON MEDICAL IMAGING, VOL. 28, NO. 1, JANUARY 2009
The BoneXpert Method for Automated
Determination of Skeletal Maturity
Hans Henrik Thodberg*, Sven Kreiborg, Anders Juul, and Karen Damgaard Pedersen
Abstract—Bone age rating is associated with a considerable
variability from the human interpretation, and this is the motiva-
tion for presenting a new method for automated determination of
bone age (skeletal maturity). The method, called BoneXpert, re-
constructs, from radiographs of the hand, the borders of 15 bones
automatically and then computes “intrinsic” bone ages for each
of 13 bones (radius, ulna, and 11 short bones). Finally, it trans-
forms the intrinsic bone ages into Greulich Pyle (GP) or Tanner
Whitehouse (TW) bone age. The bone reconstruction method
automatically rejects images with abnormal bone morphology or
very poor image quality. From the methodological point of view,
BoneXpert contains the following innovations: 1) a generative
model (active appearance model) for the bone reconstruction; 2)
the prediction of bone age from shape, intensity, and texture scores
derived from principal component analysis; 3) the consensus bone
age concept that defines bone age of each bone as the best estimate
of the bone age of the other bones in the hand; 4) a common bone
age model for males and females; and 5) the unified modelling of
TW and GP bone age. BoneXpert is developed on 1559 images.
It is validated on the Greulich Pyle atlas in the age range 2–17
years yielding an SD of 0.42 years [0.37; 0.47] 95% conf, and on
84 clinical TW-rated images yielding an SD of 0.80 years [0.68;
0.93] 95% conf. The precision of the GP bone age determination
(its ability to yield the same result on a repeated radiograph) is
inferred under suitable assumptions from six longitudinal series
of radiographs. The result is an SD on a single determination of
0.17 years [0.13; 0.21] 95% conf.
Index Terms—Active appearance models, bone age, computer
aided diagnosis (CAD), shape analysis, skeletal maturity, quanti-
tative radiology.
I. INTRODUCTION AND OVERVIEW
T
HE assessment of skeletal maturity, or bone age, of chil-
dren is an important tool in pediatrics, but the procedure
has not improved over the last 30 years. It is still based on a sub-
jective evaluation of a hand radiograph and its reliability is com-
pletely dependent on the availability of careful and well-trained
raters. In a modern radiology department it is becoming increas-
ingly difficult to maintain a high standard of bone age rating,
because the rating is often done by different persons, and there
is a tendency that less time is spent on this task.
Manuscript received January 03, 2008; revised May 05, 2008. First published
May 23, 2008; current version published December 24, 2008. This work was
supported in part by Novo Nordisk A/S. Asterisk indicates corresponding au-
thor.
*H. H. Thodberg is with Visiana Aps, Søllerødvej 57C, DK-2840 Holte, Den-
mark (e-mail: thodberg@gmail.com).
S. Kreiborg is with the University of Copenhagen, Copenhagen, Denmark
(e-mail: sven@odont.ku.dk).
A. Juul and K. D. Pedersen are with Rigshospitalet, Copenhagen, Denmark
(e-mail: anders.juul@rh.region.dk; karen.damgaard@rh.region.dk).
Color versions of one or more of the figures in this paper are available online
at http://ieeexplore.ieee.org.
Digital Object Identifier 10.1109/TMI.2008.926067
As a consequence, bone age rating is associated with a con-
siderable rater variability, the size of which is often unknown.
The pediatrician, who is using the bone age to diagnose the child
or to monitor the treatment, is therefore often uncertain about
the reliability of the rating. A reliable rating is important as a
basis for a more patient-specific treatment, and bone age is a
fundamental characteristic of the child that influences the best
practice in many areas.
The aim of this paper is to change the status of bone age as-
sessment by introducing a new, computerized, and 100% auto-
mated approach called BoneXpert.
BoneXpert makes use of technologies from medical image
analysis, statistics, and machine learning, which have not been
used previously for this task, but at the same time the method
is to a large extent based on the insight into human biology ex-
posed in the classic book in this field Assessment of Skeletal Ma-
turity and Prediction of Adult Height (TW3 Method) [1], which
also defines the specific Tanner–Whitehouse (TW) method for
bone age rating.
It is therefore appropriate to start with a quote from this book
[1, p. 22]:
From the beginning it seemed reasonable to suppose
that bone age assessments were something that a computer
should be able to do better than a human operator.
The realization of this vision has been remarkably slow.
Tanner was among the first to present a computerized system,
CASAS [2], which was received with some interest in the
pediatric endocrinology community [3]. Other systems came
along from Hill [4], Sato [5], and Pietka [6], but none of these
became common in clinical practice.
A common problem of these systems is their limited ability
to reconstruct the bone borders, i.e., to automatically locate
anatomically meaningful points at the relevant locations on
each bone. As a result, these systems are not fully automated;
they are able to process at most 90% of the cases, so they must
be supervised by an expert. In the first versions of CASAS the
bone reconstruction was actually done manually by the user,
who placed the film under a video camera and adjusted the lo-
cation of the film and the magnification for each bone to match
a template and the interpretation was then done automatically.
This was an elegant way to initiate the computerization of the
task.
The architecture of BoneXpert [7] divides the processing into
three layers, shown in Fig. 1. Layer A reconstructs the bone
borders, layer B computes an intrinsic bone age value for each
bone, and layer C transforms the intrinsic bone age values to
either TW bone age or Greulich Pyle (GP) bone age [8] using a
relatively simple postprocessing. An overview of the layers and
related prior work follows. The Appendix introduces skeletal
maturity and the TW and GP rating systems.
Layer A: The bone reconstruction algorithm is based on ac-
tive appearance models (AAMs), a powerful paradigm in image
0278-0062/$25.00 © 2008 IEEE