Improving prediction of skeletal growth problems for age evaluation using hand X-rays

• Published in: Multimedia tools and applications 2023-10, Vol.83 (33), p.80027-80049
• Main Authors: Farooq, Hina, Umer, Muhammad, Saidani, Oumaima, Almuqren, Latifah, Distasi, Riccardo
• Format: Article
• Language: English
• Subjects: 1238: Recent Advances in Biometrics Based on Biomedical Information; Abnormalities; Age; Artificial neural networks; Chronology; Computer Communication Networks; Computer Science; Data Structures and Information Theory; Joints (anatomy); Machine learning; Medical imaging; Misalignment; Multimedia Information Systems; Pediatrics; Predictions; Special Purpose and Application-Based Systems
• ISSN: 1573-7721; 1380-7501; 1573-7721
• DOI: 10.1007/s11042-023-17364-9

Skeletal age estimation using X-ray images is a widely employed clinical method for identifying anomalies in bone growth in infants and newborns. Pediatric bone abnormalities can arise from a spectrum of conditions, including wounds, infections, or tumors. Damage to the growth plate, stemming from factors like inadequate blood supply, separation from bone components, or minor misalignment, can impede bone development, distort joint structure, and potentially result in lasting joint injuries. Divergence between chronological and assessed ages can serve as an indicator of growth-related problems, as accurate bone age assessment mirrors the actual progression of growth. Skeletal age estimation plays a pivotal role in identifying endocrine disorders, genetic abnormalities, and growth irregularities. In our effort to address the challenge of bone age assessment, this study utilizes the Radiological Society of North America’s Pediatric Bone Age Challenge dataset, comprising 12,600 radiological images of patients’ left hands, along with their gender and bone age data. We propose a robust bone age evaluation system grounded in hand skeleton guidelines for the precise detection of hand bone maturation. The proposed approach for bone age assessment centers on a tailored convolutional neural network (CNN), which attains an accuracy rate of 97%. Moreover, this research analyzes growth rate prediction using six transfer learning models, offering valuable insights into the predictive capabilities of these models. This study not only contributes to advancing bone age estimation techniques but also underscores the potential of the proposed CNN-based approach in achieving highly accurate results, further enhancing diagnostic precision in pediatric medicine.