You are on page 1of 4

IS J AA

Skeletal Bone Age Assessment from elbow Radiographs


1

International Journal of Systems , Algorithms & Applications

P.Thangam, 2K.Thanushkodi, 3T.V.Mahendiran, 4P.S.Sujith Assistant Professor, Sri Ramakrishna Engineering College, Coimbatore, Tamilnadu, India. 2 Director, Akshaya College of Engineering and Technology, Coimbatore, Tamilnadu, India. 3 Assistant Professor, Coimbatore Institute of Engineering and Technology Coimbatore, Tamilnadu, India. 4 PG Scholar, Sri Ramakrishna Engineering College, Coimbatore, Tamilnadu, India. e-mail: saithangam@gmail.com1
1

Abstract - The purpose of this work is to design and implement a computerized system for skeletal Bone Age Assessment (BAA) during puberty period, based on the Sauvegrain method. The system ensures accurate and robust estimation of bone age for the age group 10-13 years for girls and 12-15 years for boys. A Gaussian filter is employed for noise removal, Canny edge detector for edge detection, and thereafter Boundary fill algorithm for segmentation. From the segmented apophysis ROI of the olecranon, 8 discriminating features are extracted, which are fed in to an ID3 decision tree classifier to output the age class to which the radiograph is categorized, which is mapped onto the final bone age. The system was trained with 80 radiographs, 40 radiographs of girls in age group 10-13 years and 40 radiographs of boys in age group 12-15 years. After training, the system was tested on 100 radiographs (50 for girls and 50 for boys). The performance of the system was evaluated with the help of radiologist expert diagnoses. The results show that 42 out of 50 were correctly classified for girls (84%) and 40 out of 50 were correctly classified for boys (80%), with a minimal error rate of 0.66% for girls and 0.68% for boys. The system is very reliable, yielding excellent results with minimum human intervention. The performance of the system is excellent during puberty when compared to the conventional BAA systems. Keywords: Bone Age Assessment (BAA), Feature extraction, radiograph, ID3 Classification.

I. INTRODUCTION Bone age assessment is an important clinical tool in the area of pediatrics, especially in relation to endocrinological problems and growth disorders[1]. Based on a radiological examination of development of certain parts of the skeleton, bone age is assessed and compared with the chronological age. A discrepancy between these two values indicates abnormalities in skeletal development. The procedure is often used in the management and diagnosis of endocrine disorders and also serves as an indication of the therapeutic effect of treatment. It indicates whether the growth of a patient is accelerating or decreasing, based on which the patient can be treated with growth hormones. BAA is universally used due to its simplicity, minimal radiation exposure, and the availability of multiple ossification centers for evaluation of maturity. Many different methods are used to assess skeletal maturity. Each method makes use of different parts of the skeleton. Acheson proposed the Oxford method [2] based on the pelvis, Pyle and Hoerr [3], on the knee, Hoerr et al., on the foot [4], Tanner et al.[5], and Greulich and Pyle[6], based on the hand and wrist, and the Sauvegrain method [7], based on the elbow. Of all the above, the main

clinical methods for skeletal bone age estimation are the Greulich & Pyle (GP) method and the Tanner & Whitehouse (TW) method. GP is an atlas matching method while TW is a score assigning method [8]. GP method is faster and easier to use than the TW method. In GP method, a left-hand wrist radiograph is compared with a series of radiographs grouped in the atlas according to age and sex. The atlas pattern which superficially appears to resemble the clinical image is selected. TW method uses a detailed analysis of each individual bone, assigning it to one of eight classes reflecting its developmental stage. This leads to the description of each bone in terms of scores. The sum of all scores assesses the bone age. This method yields the most reliable results. Cundy et al.[9] demonstrated that the Greulich and Pyle atlas had a large inter-observer error, which is problematic in the assessment of skeletal maturity during pubertal growth. Little et al.[10] stated that the use of the Greulich and Pyle atlas could not improve accuracy in the prediction of limb-length inequality. Hence, there comes a necessity for alternative methods for skeletal BAA. Next to the above wrist x-ray methods, the elbow based methods are of much significance, especially during the spurt of puberty. The elbow is characterized by clear developmental sequences of its ossification centers beginning at nine years of age in girls and eleven years in boys. Fusion of the elbow growth centers is complete at thirteen years in girls and fifteen years in boys. During the critical pubertal period, morphological changes of the olecranon apophysis (shown in Fig. 1) of the elbow are very characteristic. Alain et al. [11] conducted a study to determine the accuracy of the Sauvegrain method in 2005, and proved that certain elbow growth centers showed intermediate development morphology, which failed to correspond to the scores assigned by Sauvegrain method. They also suggested intermediate scores for such cases. The purpose of this work is to propose and develop a system to estimate the bone age during puberty, based on the ossification of the olecranon apophysis in the elbow, since the interpretation of skeletal age performed on the olecranon apophysis alone is a simplified but very practical method in clinical practice [12].

Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

30

SKELETAL BONE AGE ASSESSMENT FROM ELBOW RADIOGRAPHS

IS J AA

International Journal of Systems , Algorithms & Applications

Fig. 1. Lateral view of elbow joint [13]

1. Presence whether the apophysis of the bone is present or absent 2. Circularity whether the apophysis is circular in shape or not 3. Multiplicity whether the apophysis is single or multiple 4. Proximity whether both the nuclei of apophysis are closer or not 5. Merger whether the apophysis nuclei have merged or not. 6. Diameter whether the diameter of the apophysis is smaller than that of the ulna or not. 7. Fusion whether the apophysis has fused with the ulna or not. Each of the above features is extracted from the respective ROI

II. MATERIALS AND METHODS The proposed work consists of the following modules, as depicted in Fig. 2.
Image Pre-processing & Segmentation Feature Extraction ID3 Classification

2.1 Image Pre-processing& Segmentation The input image is enhanced by image smoothing to reduce the noise within the image or to produce a less pixilated image. A Gaussian filter is employed here to reduce noise and smoothen the image, followed by Canny edge detector to detect bone edges. The Canny edge detector [14],[15] is a significant and widely used contribution to edge detection techniques. This detector is optimal for step edges corrupted by white noise [16]. The Canny edge detector first smoothes the image to eliminate noise. It then finds the image gradient to highlight regions with high spatial derivatives. Segmentation using Boundary fill algorithm: The boundary fill algorithm fills a region with the given fill color until the given boundary color is found. The 8-point technique is used here to color a region. The algorithm is used recursively; it returns when the pixel to be colored is the boundary color or is already the fill color [17]. Fig. 3 provides a snapshot of the input, edge detected and the segmented images for three classes. 2.2 Feature Extraction Feature extraction is performed as a way to reduce the dimensionality of the data [18,19]. A long list of candidate features was calculated in order to form a powerful input vector. The features attempted to describe the morphology of the outline shape of the bones (ROIs). Once extracted and optimized, the vector would be used to train and validate the classifier. For the olecranon, the degree of maturity is based on the extent of ossification of its apophysis, shown in Fig. 4. So, we consider the following features to be extracted from the olecranon apophysis, namely:
Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

Fig. 2. Steps in Bone Age Estimation

(a) (b) (c) (a) Fig. 3. (a) Input (b) Edge detected and (c) Segmented images for the Sauvegrain stages 2, 4 and 6.5 respectively.

31

SKELETAL BONE AGE ASSESSMENT FROM ELBOW RADIOGRAPHS

IS J AA

International Journal of Systems , Algorithms & Applications

Fig. 4. Ossification stages of Olecranon apophysis [7].

2.3 Classification This module is responsible for translating the features into corresponding skeletal bone age. The selected features take positions in the ID3 decision tree based on the order of ossification of the bones, which influence their contribution towards the bone age estimation procedure. 2.3.1 ID3 classifier: Studies reveal that there may be discrepancies in the results of classifications by different radiologists or at sometimes even between the classifications by the same radiologist in different moments [22]. This is because some stages are mistaken with the nearby stages easily. So it becomes difficult to strictly define the features. Instead, they have a certain degree of overlapping between them, since the growth pattern is gradual. Also the features can fall into only any one of the categories (i.e.) mutually exclusive. So the ID3 classifier is more suitable for our system. ID3 is the third series of Iterative Dichotomiser procedures, hence named so. They are intended for use with nominal inputs only and they have their number of levels equal to the number of input variables. The algorithm continues until all nodes are pure or there are no more variables on which to split [23]. Choosing the root is an important issue to be considered for efficient classification [24]. The features with greater discrimination power are to be placed above those with low values in the decision tree. The inputs to the ID3 tree (in Fig. 5) will be the 8 features modeled as Boolean values, and the output will be the skeletal class or category. The selected class is then mapped onto the skeletal bone age, based on the criteria shown in Table 1, which is different for girls and boys. 2.3.2 Training Sets: The ID3 decision tree was trained with 10 male and 10 female radiographs for each age group, thus with a total of 80 radiographs, 40 male and 40 female cases for the age group 10-13 years for girls and 12-15 years for boys. After the decision tree was built and fine-tuned, 50 radiographs from girls and 50 from boys were used to test it. The results showed that 42 out of 50 were correctly classified for the girls (84%) and 40 out of 50 were correctly classified for the boys (80%). The classification error was to assign a later class or a previous class. III. RESULTS AND DISCUSSION The performance of the computerized system in estimating the bone age was evaluated using a dataset consisting of 100 digital radiographs (50 for boys and 50 for girls). The quality of the image is a vital factor of influence in the estimation process. So much consideraVolume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

tion was imparted on obtaining quality radiographs. Also particular attention was dedicated to design and implement robust techniques for preprocessing and reliable feature extraction. During the preprocessing, the noise caused due to radiation and other external factors were eliminated. From the segmented ROI, 8 morphological features were extracted and modeled into formal features (Boolean values) suitable to be fed into the classifier. Finally, the ID3 classifier was used to estimate the final bone age. The accuracy of the classifier was measured in terms of two metrics, namely the Classification Rate CR and Average Relative Error ARE. Classification rate, CR is defined as the ratio of the number of correctly classified images NCC to the total number of test images input to the classifier, NIP.

CR =

N CC N IP

Incorrectly classified images include the images with under estimated or over estimated bone age when compared with the bone age estimated by a radiologist expert. The accuracy of classification is given in Table 2. Average Relative Error, ARE is the average of the relative errors between the estimated bone age and the predicted bone age by the expert. It is given by,

ARE =

(2) The ARE values of the classifier are provided in Table 3.


TABLE I
CRITERIA SELECTION

1 n ei n i =1

TABLE II CLASSIFICATION RATE

TABLE III AVERAGE RELATIVE ERROR

32

SKELETAL BONE AGE ASSESSMENT FROM ELBOW RADIOGRAPHS

IS J AA

International Journal of Systems , Algorithms &

Fig. 8. ID3 Decision Tree

IV. CONCLUSION In this paper, we have proposed an efficient bone age assessment approach to estimate the skeletal bone age from elbow radiograph. The method is advantageous in that it requires only small amount of data to train the classifier and from the results it is evident that the performance of the system is excellent during puberty when compared to the conventional BAA systems. The system was tested on a set of 100 radiographs (50 from girls and 50 from boys), achieving a success rate for bone age estimation of 84% for girls and 80% for boys, with a minimal error rate of 0.66% for girls and 0.68% for boys. Future work will be focused on extending the system to work on the age group below 10 years, broadening the system to include the further elbow bones such as Trochlea, Epicondyle etc. and integrating the system with the clinical PACS. ACKNOWLEDGEMENTS The authors would like to thank Dr.R.Saravanan, MD (Pediatrics), KKCT Hospital, Chennai for providing the database of hand radiographs and Dr.R.Shankar Anandh, MD(Radio Diagnosis) and Dr.S.Sanjitha DMRD(Radio Diagnosis) of Barnard institute of Radiology, Chennai for their valuable help in diagnosing the radiographs and validating the system. REFERENCES
[1] Vicente Gilsanz, and Osman Ratib, Hand Bone Age A Digital Atlas of Skeletal Maturity. Springer-Verlag, 2005.

[2] Acheson RM. The Oxford method of assessing skeletal maturity. Clin Orthop Relat Res. 1957;10:19-39. [3] Pyle SI, Hoerr NL. Radiographic atlas of skeletal development of the knee. Springfield: Charles C. Thomas; 1955. [4] Hoerr NL, Pyle SI, Francis CC. Radiographic atlas of skeletal development of the foot and ankle, a standard of reference. Springfield, IL: Charles C. Thomas; 1962. [5] J.M.Tanner, R.H.Whitehouse, Assessment of Skeletal Maturity and Prediction of Adult Height (TW2 method). Academic Press, 1975. [6] Greulich WW, Pyle SI. Radiographic atlas of skeletal development of the hand and wrist. 2nd ed. Stanford: Stanford University Press; 1959. [7] Sauvegrain J, Nahum H, Bronstein H. Study of bone maturation of the elbow. Ann Radiol (Paris). 1962;5:542-50. French. [8] Concetto Spampinato, Skeletal Bone Age Assessment. University of Catania, Viale Andrea Doria, 6 95125, 1995. [9] Cundy P, Paterson D, Morris L, Foster B. Skeletal age estimation in leg length discrepancy. J Pediatr Orthop. 1988;8:513-5. [10] Little DG, Nigo L, Aiona MD. Deficiencies of current methods for the timing of epiphysiodesis. J Pediatr Orthop. 1996;16:173-9. [11] Alain Dimeglio, Yann Philippe Charles, Jean-Pierre Daures, Vincenzo De Rosa, and Boniface Kabore, Accuracy of the Sauvegrain Method in Determining Skeletal Age During Puberty. The Journal of Bone and Joint Surgery, Vol. 87-A, No. 8, August 2005. [12] Dimeglio A. Growth in pediatric orthopaedics, J Pediatr Orthop. 2001; 21:549-55. [13] http://www.joint-pain-expert.net/ [14] Rafael C.Gonzalez, Richard E. Woods, Digital Image Processing. Third Edition, (Pearson, 2009). [15] Anil K. Jain, Fundamentals of Digital Image Processing. Prentice Hall, 1989. [16] Canny, A Computational Approach to Edge Detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 8, 679-698,1986. [17] Donald Hearn, M.Pauline Baker, Computer Graphics with OpenGL. Pearson Education, Third Edition, 2009. [18] Lefkaditis, D., Awcock, G. J. & Howlett, R. J., Intelligent Optical Otolith Classification for Species Recognition of Bony Fish. International Conference on Knowledge-Based & Intelligent Information & Engineering Systems, Bournemouth, UK: Springer-Verlag, pp. 1226-1233, 2006. [19] Mark S. Nixon and Alberto S. Aguado, Feature Extraction and Image Processing. Newnes, 2002. [20] Christopher J.C. Burges, Geometric Methods for Feature Extraction and Dimensional Reduction. Microsoft Research. [21] R. Tyrrell Rockafellar, Roger J-B Wets, Variational Analysis. Springer-Verlag, 2005. [22] Vinuela-Rueda B, Alberola-Lopez S, Inter and intra observer variability in bone age assessment using TW2-RUS . in: 26 Congreso Nacional Soc. Esp. Rad. Medica, Maspalomas, Spain; 2002. [23] Bing Liu, Yiyuan Xia, and Philip S. Yu, Clustering through decision tree construction. Proceedings of the ninth international conference on Information and knowledge management. ACM, New York, 2000. [24] Richard O.Duda, Peter E.Hart and David G.Stork, Pattern Classification. Second Edition, John Wiley, 2002

Volume 2, Issue ICTM 2011, February 2012, ISSN Online: 2277-2677 ICTM 2011|June 8-9,2011|Hyderabad|India

33