Bone Age Estimation Using X-ray Images with Gender-Aware Preprocessing

Muradnar, Rohan Balasaheb (2025) Bone Age Estimation Using X-ray Images with Gender-Aware Preprocessing. Masters thesis, Dublin, National College of Ireland.

Preview	PDF (Master of Science) Download (910kB) | Preview
Preview	PDF (Configuration Manual) Download (1MB) | Preview

Abstract

In paediatric endocrinology and orthopaedics, reliable estimates of skeletal-maturity are necessary but current Greulich-Pyle and Tanner-Whitehouse atlases are time-consuming, subjective and demographically-biased. As a result, we created a two-stage, gender-sensitive deep-learning pipeline where sex is a routing variable as opposed to optional metadata. In Stage 1, a lightweight convolutional neural network (CNN) classifies sex. In Stage 2, all the radiographs are passed to a sex-specific, transfer-learned InceptionV3 regressor. An auxiliary preprocessing engine adjusts CLAHE contrast, segmentation margins and crop heuristics separately in boys and girls to conserve dimorphic ossification cues. The balanced QU hand-X-ray cohort and an external RSNA audit set were used to conduct experiments. Baselines included (i) a single gender-agnostic regressor and (ii) late-fusion InceptionV3 that appends the predicted sex to the features. Evaluation encompassed mean absolute error (MAE), mean-squared error, R², proportions within ±6/±12 months, calibration curves, Bland–Altman bias and bootstrap tests of paired errors. The suggested framework also reduced MAE, enhanced calibration, and, more importantly, reduced sex-stratified error disparities relative to both baselines by half. Ablation tests verified that both the sex gate and gender-tuned preprocessing made a positive input to the gains. All code, configuration files and logs are under version control, and Grad-CAM maps demonstrate anatomically plausible focus by sex. At preprocessing and architectural levels, therefore, embedding domain knowledge on sexual dimorphism increases the accuracy and fairness of automated bone-age assessment.

Item Type:	Thesis (Masters)
Supervisors:	Name Email Khan, Sallar UNSPECIFIED
Uncontrolled Keywords:	Bone age assessment; deep learning; gender-aware preprocessing; sex classification; transfer learning; fairness; pediatric radiology; convolutional neural networks; calibration
Subjects:	R Medicine > RJ Pediatrics Q Science > QH Natural history > QH301 Biology > Methods of research. Technique. Experimental biology > Data processing. Bioinformatics H Social Sciences > HQ The family. Marriage. Woman > Gender R Medicine > Healthcare Industry
Divisions:	School of Computing > Master of Science in Data Analytics
Depositing User:	Ciara O'Brien
Date Deposited:	26 Aug 2026 09:11
Last Modified:	26 Aug 2026 09:11
URI:	https://norma.ncirl.ie/id/eprint/9647

Actions (login required)

View Item