[Paper Review] Child Gender Determination with Convolutional Neural Networks on Hand Radio-Graphs

Mumtaz Ali Kaloi, Kun He|arXiv (Cornell University)|Nov 13, 2018

Dermatoglyphics and Human TraitsBiochemistry, Genetics and Molecular Biology7 references3 citations

TL;DR

This paper proposes GDCNN, a convolutional neural network model that determines child gender using left-hand radiographs with 98% accuracy, leveraging class activation mapping (CAM) to identify the wrist (carpals) region as the primary area of attention. The method is the first to apply deep learning to child gender determination via hand X-rays, offering high accuracy even with incomplete skeletal development.

ABSTRACT

Motivation: In forensic or medico-legal investigation as well as in anthropology the gender determination of the subject (hit by a disastrous or any kind of traumatic situation) is mostly the first step. In state-of-the-art techniques the gender is determined by examining dimensions of the bones of skull and the pelvis area. In worse situations when there is only a small portion of the human remains to be investigated and the subject is a child, we need alternative techniques to determine the gender of the subject. In this work we propose a technique called GDCNN (Gender Determination with Convolutional Neural Networks), where the left hand radio-graphs of the children between a wide range of ages in 1 month to 18 years are examined to determine the gender. To our knowledge this technique is first of its kind. Further to identify the area of the attention we used Class Activation Mapping (CAM). Results: The results suggest the accuracy of the model is as high as 98%, which is very convincing by taking into account the incompletely grown skeleton of the children. The attention observed with CAM discovers that the lower part of the hand around carpals (wrist) is more important for child gender determination.

Motivation & Objective

• To develop an automated, deep learning-based method for gender determination in children when only partial skeletal remains are available.
• To address the limitation of traditional forensic methods that rely on skull and pelvis, which may be damaged or absent in child victims.
• To explore the potential of hand radiographs—particularly the left hand—as a reliable alternative for gender determination in pediatric forensic cases.
• To identify the most discriminative anatomical regions in hand X-rays for gender classification using attention visualization techniques.
• To establish a high-accuracy model suitable for real-world forensic applications involving children with incomplete remains.

Proposed method

• A custom convolutional neural network (GDCNN) is trained on a dataset of left-hand radiographs from children aged 1 month to 18 years.
• The model uses transfer learning and deep feature extraction to classify gender based on morphological differences in hand bones.
• Class Activation Mapping (CAM) is applied to visualize and localize the regions of the hand that contribute most to the prediction.
• The attention maps are analyzed across thousands of test images to identify consistent patterns in feature importance.
• The network is trained and validated using standard deep learning protocols, including cross-validation and accuracy evaluation on test splits.
• The model is evaluated on a diverse dataset of pediatric hand X-rays to ensure generalization across age groups.

Experimental results

Research questions

• RQ1 Can a deep learning model achieve high accuracy in determining the gender of children using only hand radiographs?
• RQ2 Which anatomical regions in the hand are most informative for child gender classification?
• RQ3 How does the attention mechanism of a CNN align with known anatomical differences in pediatric hand bones?
• RQ4 Can the model generalize across a wide age range (1 month to 18 years) despite incomplete skeletal development?
• RQ5 Does the model outperform traditional anthropometric and morphological techniques in accuracy and robustness for pediatric cases?

Key findings

• The GDCNN model achieves a gender classification accuracy of 98% on test data, significantly outperforming previous methods.
• Class Activation Mapping (CAM) reveals that the carpals (wrist) region is the primary area of attention for gender prediction in both male and female children.
• The model shows consistent attention on the ulna, radius, and carpals across over 1,400 test images, indicating strong regional focus.
• The bones in the carpals region of males are observed to be thicker and more tightly packed, while females show thinner bones with visible gaps—this pattern is captured by the model.
• The phalanges (finger bones) contribute minimally to gender prediction, according to attention maps.
• The model maintains high performance even when only the lower part of the hand (carpals) is available, suggesting robustness to partial image input.