[Paper Review] Predictive Modelling of Bone Age through Classification and Regression of Bone Shapes

Anthony Bagnall, J. Lucian Davis|arXiv (Cornell University)|Jun 18, 2014

Forensic Anthropology and Bioarchaeology Studies4 citations

TL;DR

This paper proposes a feature-based, modular approach to automated bone age assessment using shape descriptors derived from hand radiograph bone outlines. By classifying bone stages and regressing chronological age using summary shape features—particularly from just three bones—it achieves accuracy matching expert human assessors, while enabling transparency, demographic analysis, and population-specific model adaptation.

ABSTRACT

Bone age assessment is a task performed daily in hospitals worldwide. This involves a clinician estimating the age of a patient from a radiograph of the non-dominant hand. Our approach to automated bone age assessment is to modularise the algorithm into the following three stages: segment and verify hand outline; segment and verify bones; use the bone outlines to construct models of age. In this paper we address the final question: given outlines of bones, can we learn how to predict the bone age of the patient? We examine two alternative approaches. Firstly, we attempt to train classifiers on individual bones to predict the bone stage categories commonly used in bone ageing. Secondly, we construct regression models to directly predict patient age. We demonstrate that models built on summary features of the bone outline perform better than those built using the one dimensional representation of the outline, and also do at least as well as other automated systems. We show that models constructed on just three bones are as accurate at predicting age as expert human assessors using the standard technique. We also demonstrate the utility of the model by quantifying the importance of ethnicity and sex on age development. Our conclusion is that the feature based system of separating the image processing from the age modelling is the best approach for automated bone ageing, since it offers flexibility and transparency and produces accurate estimates.

Motivation & Objective

• To develop an accurate, transparent, and modular system for automated bone age assessment using image-based shape features.
• To compare three approaches—direct outline representation, shapelet-based classification, and summary shape features—for predicting bone age stages.
• To evaluate the performance of regression models using shape features in predicting chronological age.
• To investigate the influence of demographic factors such as sex and ethnicity on bone age development.
• To enable flexible, explainable, and locally adaptable models for diverse clinical populations.

Proposed method

• Segmented hand radiographs to isolate individual bone outlines and verified their segmentation using geometric constraints.
• Extracted three types of features: one-dimensional outline representations, shapelets (discriminatory subsequences), and summary shape features based on Tanner-Whitehouse (TW) descriptors.
• Trained classifiers (including Elastic Ensemble and shapelet-based models) on the three feature types to predict TW bone stage categories.
• Constructed linear and non-linear regression models using shape features to predict chronological age from bone outlines.
• Applied stepwise regression and hypothesis testing to assess the significance of demographic variables (sex, ethnicity) in age prediction.
• Evaluated model performance using MAE, RMSE, and R² metrics, comparing results to human expert estimates.

Experimental results

Research questions

• RQ1 Which feature representation—raw outline, shapelets, or TW-based summary features—yields the most accurate bone age stage classification?
• RQ2 Can a regression model based on shape features predict chronological age with accuracy comparable to expert human assessors?
• RQ3 How do demographic factors such as sex and ethnicity influence bone age development, and can these be quantitatively modeled?
• RQ4 To what extent can accurate age prediction be achieved using only three bones, and which features are most predictive?
• RQ5 Can a modular, feature-based system offer greater transparency and adaptability than end-to-end deep learning or atlas-based methods?

Key findings

• Models based on TW-derived summary shape features outperformed those using one-dimensional outline representations and shapelets on at least one bone.
• Regression models using just three bones (phalanges, epiphysis) achieved a mean absolute error (MAE) of 0.98 and root mean square error (RMSE) of 1.67, matching the accuracy of expert human assessors.
• A two-variable linear model using epiphyseal width and distance to phalanx explained approximately 80% of the variance in chronological age (R² ≈ 0.80).
• Ethnicity significantly affected bone age development, with Asian ethnicity being a highly significant predictor; the model with sex and ethnicity had an RMSE of 0.855, slightly better than human raters (RMSE 0.89).
• Sex was a significant predictor in regression models, with interaction terms and stepwise selection confirming its importance, though the sex coefficient was not significantly different from zero in the full model.
• The feature-based modular system enabled transparent, explainable, and adaptable models, supporting future integration of local demographic data and hypothesis testing via confidence intervals.