Deep Learning-Based Sanders Classification

Koç Ö., HANCI N. B., ERDEM O., TUNA F.

28th IEEE Signal Processing: Algorithms, Architectures, Arrangements, and Applications, SPA 2025, Poznan, Polonya, 17 - 19 Eylül 2025, ss.279-284, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.23919/spa65537.2025.11215099
  • Basıldığı Şehir: Poznan
  • Basıldığı Ülke: Polonya
  • Sayfa Sayıları: ss.279-284
  • Anahtar Kelimeler: ConvNeXt, convolutional neural networks, deep learning, medical image analysis, pediatric bone age, Sanders classification, skeletal maturity assessment
  • Trakya Üniversitesi Adresli: Evet

Özet

This study presents a deep learning-based framework for automated Sanders classification, which is a clinically established method for assessing skeletal maturity from hand radiographs. A custom dataset comprising 1.057 annotated pediatric hand X-ray images was constructed by an expert physiatrist, based on the RSNA Pediatric Bone Age Challenge. To perform multi-class classification across eight Sanders stages, a ConvNeXt-small convolutional neural network was fine-tuned using advanced data augmentation techniques, the use of weighted sampling and class-balanced loss functions to address dataset imbalance. The model achieved a validation accuracy of 77.2% and a test accuracy of 66.1%, with particularly strong performance in late-stage ossification patterns. The proposed system demonstrates the feasibility of applying state-of-the-art deep learning architectures to support medical specialists in making consistent, rapid, and objective assessments of skeletal development, potentially aiding diagnosis and treatment planning in pediatric field.