Makine Öğrenmesi ile Radyolojik Görüntülerden Kemik Yaşı Tahmini

Öz

Kemik yaşı tahmini, endokrinolojik sorunların ve adli sorunların tanısında önemlidir. Greulich ve Pyle (GP) yöntemi kemik yaşı tahmini için yaygın olarak kullanılmaktadır. Ancak, gözlemcinin kendisi ve gözlemciler arası nispeten yüksek bir değişkenliğe sahiptir. Bu nedenle, kemik yaşının hesaplanmasında uzmanlardan bağımsız otomasyon tabanlı sistemler geliştirilmeye başlanmıştır. Bu çalışmada, makine öğrenimine dayalı sınıflandırma yöntemlerinin kemik yaşı tahmin performanslarını karşılaştırmayı amaçladık. Çalışmaya 12-108 aylık 388 erkek ve 387 kız dahil edildi. Cohort el bilek grafilerinde kemik alanın tüm el bilek alanına oranı her olgu için hesaplandı ve olgular üçer aylık intervaller ile sınıflandırıldı. Bu, veri tabanı olarak kabul edilip test verisi bu veri tabanı ile test edildi. Kemik yaşı tahmini için makine öğrenmesine (ML) dayanan tahmin modellerini kullandık. Weka ara yüzü kullanılarak oluşturulan modellerin tahmini performansları kronolojik yaş ile karşılaştırıldı. Ayrıca yöntemlerin öngörücü performansı arasında istatistiksel olarak anlamlı bir fark olup olmadığı Friedman testi ile test edilmiştir. Sonuç olarak, kız çocukları için ML yöntemleriyle yapılan kemik yaşı tahmininin kronolojik yaş ile anlamlı derecede ilişkili olduğu gözlenmiştir. GP ve kronolojik yaş arasında anlamlı bir fark bulundu. Bu çalışmadan elde edilen sonuçlar, ML tabanlı sınıflandırma yöntemlerinin kemik yaşını tahmin etmede yüksek başarı gösterdiğini göstermiştir. Bu nedenle, ML sınıflandırma modellerinin kemik yaşını tahmin etmek için kullanılabileceğini önermekteyiz.

Anahtar Kelimeler

• Bilek Radyografisi
• Kemik Yaşı Tahmini
• Makine Öğrenmesi

Proje Numarası

Project Grant Number: 17/217

Estimation of Bone Age from Radiological Images with Machine Learning

Öz

Bone age estimation (BAE) is important in the diagnosis of endocrinological problems and forensic issues. Greulich and Pyle (GP) method is widely used for BAE. But it has relatively high intraobserver and interobserver variability. For this reason, automation-based systems independent of experts have started to be developed in estimating bone age. We aimed to compare bone age estimation performances of machine learning based classification methods. A total of 388 boys and 387 girls between the age of 12-108 months were included in the study. In Cohort wrist radiographs, the ratio of bone area to the entire wrist area was calculated for each case, and the cases were classified with quarterly intervals. This is considered as a database and the test data has been tested with this database. We used the estimation models which are based on Machine learning (ML) for BAE. The predicted performances of the models created by using Weka interface were compared with chronological age. Moreover, whether there is a statistically significant difference between the predictive performance of the methods was tested by the Friedman test. As a result, it was observed that bone age estimation performed with ML methods for girls was significantly correlative with chronological age. A significant difference was found between GP and chronological age. The results obtained from this study showed that ML-based classification methods have high success in predicting bone age. Therefore, we suggest that ML classification models can be used to predict bone age.

Anahtar Kelimeler

• Bone Age Estimation
• Machine Learning
• Wrist Radiography

Destekleyen Kurum

This paper has been granted by the Mugla Sıtkı Kocman University Research Projects Coordination Office.

Proje Numarası

Project Grant Number: 17/217

Kaynakça

1. 1. Gilsanz V and Ratib O. Hand Bone Age: A Digital Atlas of Skeletal Maturity. 2005; 98. Springer Science & Business Media, Heidelberg.
2. 2. Maggio A, Flavel A, Hart R, et al. Skeletal age estimation in a contemporary Western Australian population using the Tanner-Whitehouse method. Forensic Sci Int. 2016;63:1-8.
3. 3. Pinchi V, De Luca F, Ricciardi F, et al. Skeletal age estimation for forensic purposes: A comparison of GP, TW2 and TW3 methods on an Italian sample. Forensic Sci Int. 2014;238:83-90.
4. 4. Cantekin K, Çelikoğlu M, Miloglu O, et al. Bone Age Assessment: The Applicability of the Greulich-Pyle Method in Eastern Turkish Children. J Forensic Sci. 2012;57(3):679-82.
5. 5. Öztürk F, Karataş OH, Mutaf IH, et al. Bone age assessment: comparison of children from two different regions with the Greulich–Pyle method In Turkey. Aust J Forensic Sci. 2016;48(6):694-703.
6. 6. Büken B, Şafak AA, Yazıcı B, et al. Is the assessment of bone age by the Greulich–Pyle method reliable at forensic age estimation for Turkish children? Forensic Sci Int. 2007;173:146-53.
7. 7. Berst MJ, Dolan L, Bogdanowicz MM, et al. Effect of knowledge of chronologic age on the variability of pediatric bone age determined using the Greulich and Pyle standards. AJR Am J Roentgenol. 2001;176(2):507-10.
8. 8. King DG, Steventon DM, O'sullivan MP, et al. Reproducibility of bone ages when performed by radiology registrars: an audit of Tanner and Whitehouse II versus Greulich and Pyle methods. Br J Radiol. 1994;67(801):848-51.

9. 9. Guraksin GE, Uguz H, Baykan OK. Bone age determination in young children (newborn to 6 years old) using support vector machines. Turk J Elec Eng&Comp Sci. 2016;24:1693-708.
10. 10. Gertych A, Zhang A, Sayre J, et al. Bone age assessment of children using a digital hand atlas. Comp Med Imaging Graph. 2007;31(4-5):322-31.
11. 11. Pietka E, Pospiech-Kurkowskaa S, Gertych A, et al. Integration of computer assisted bone age assessment with clinical PACS. Comput Med Imaging Graph. 2003;27(2-3):217-28.
12. 12. Seok J, Hyun B, Kasa-Vubu J, et al. Automated Classification System for Bone Age X-ray Images. IEEE International Conference on Systems, Man, and Cybernetics (SMC). IEEE. 2012;208-13.
13. 13. Tristan-Vega A, Arribas JI. A Radius and Ulna TW3 Bone Age Assessment System. IEEE Trans Biomed Eng. 2008;55(5):1463-76.
14. 14. Liu J, Qi J, Liu Z, et al. Automatic bone age assessment based on intelligent algorithms and comparison with TW3 method. Comput Med Imaging Graph. 2008;32(8):678-84.
15. 15. Hasaltın E, Beşdok E. El-bi̇lek röntgen görüntüleri̇nden radyoloji̇k kemi̇k yaşı tespi̇ti̇nde yapay si̇ni̇r ağları kullanımı. National Conference of Electrical, Electronics and Computer Engineering. 2004;8-12.
16. 16. Thangam P, Mahendiran TV. Tetrolets-based System for Automatic Skeletal Bone Age Assessment. Int J Eng Res Sci. 2015;1:21-33.
17. 17. Darmawan MF, Yusuf SM, Abdul Kadir MR, et al. Comparison on three classification techniques for sex estimation from the bone length of Asian children below 19 years old: An analysis using different group of ages. Forensic Sci Int. 2015;247:130.e1-11.
18. 18. Lee JH, Kim KG. Applying Deep Learning in Medical Images:The Case of Bone Age Estimation. Healthc Inform Res. 2018;24(1):86-92.
19. 19. Iglovikov I, Rakhlin A, Kalinin AA, et al. Paediatric bone age assessment using deep convolutional neural networks, Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support. 2018; 300-8. Springer, Cham.
20. 20. Hyunkwang L, Tajmir S, Lee J, et al. Fully Automated Deep Learning System for Bone Age Assessment. J Digit Imaging. 2017;30(4):427-41.
21. 21. Spampinatoa C, Palazzoa C, Giordano D, et al. Deep learning for automated skeletal bone age assessment in X-ray images. Med Image Anal. 2017;36:41-51.
22. 22. Thodberg HH, Kreiborg S, Juul A, et al. The BoneXpert method for automated determination of skeletal maturity. IEEE Trans Med Imaging. 2009;28 (1):52-66.
23. 23. Predicting Skeletal Age avaible at: https://www.16bit.ai/bone-age
24. 24. Haykin S. Neural networks: a comprehensive foundation. Prentice Hall PTR, 1994.
25. 25. Rumelhart DE, Geoffrey EH, Ronald JW. Learning internal representations by error propagation. No. ICS-8506. California Univ San Diego La Jolla Inst for Cognitive Science, 1985.
26. 26. Quinlan JR. Simplifying decision trees. Int J Man Mach Stud. 1987;27:221-34.
27. 27. Friedman N, Geiger D, Goldszmidt M. Bayesian network classifiers. Mach Learn. 1997;29:131-63.
28. 28. Hosmer DW, Stanley JL, Sturdivant RX. Applied logistic regression. 2013;398. John Wiley & Sons.
29. 29. Weka 3: Data Mining Software in Java avaible at: https://www.cs.waikato.ac.nz/ml/weka/
30. 30. Friedman M. The use of ranks to avoid the assumption of normality implicit in the analysis of variance. J Am Stat Assoc. 1937;32:675-701.
31. 31. Friedman M. A comparison of alternative tests of significance for the problem of m rankings. Ann Math Stat. 1940;11:86-92.
32. 32. Korting TS. C4. 5 algorithm and multivariate decision trees. Image Processing Division, National Institute for Space Research–INPE Sao Jose dos Campos–SP, Brazil 2006.
33. 33. Friedman JH, Hastie T, Tibshirani R. Additive logistic regression: a statistical view of boosting (with discussion and a rejoinder by the authors). Ann Stat. 2000;28:337-407.
34. 34. Godbole S, Sarawagi S, Chakrabarti S. Scaling multi-class support vector machines using inter-class confusion. Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining. 2002.
35. 35. Koc A, Karaoglanoglu M, Erdogan M, et al. Assessment of bone ages: is the Greulich‐Pyle method sufficient for Turkish boys? Pediatr Int. 2001;43(6):662-5.