3D statistical shape models of patella for sex classification

Area of Science:

• Forensic Anthropology
• Medical Imaging
• Biometrics

Background:

• Accurate sex determination is crucial in forensic anthropology.
• Patella bones offer potential for sex identification, especially from incomplete remains.
• Existing methods may lack automation or require complete skeletal structures.

Purpose of the Study:

• To develop and validate a novel automated feature extraction technique for sex classification from patellae.
• To compare the efficacy of various statistical and neural network classification methods.
• To assess the potential of this method for forensic applications.

Main Methods:

• A dataset of 228 patellae (95 female, 133 male) was CT scanned.
• Automated extraction, normalization, and ranking of geometric and moment-based features.
• Construction of 45-dimensional feature vectors for classification using methods like Fuzzy C-Means (FCM), Linear Discriminant Function (LDF), and Feedforward Backpropagation Neural Networks (NN).

Main Results:

• Classification success rates varied: 83.77% (FCM), 90.3% (LDF).
• Feedforward backpropagation neural networks (NN) achieved high accuracy: 96.02% training and 93.51% testing rates.
• The automated feature extraction and classification approach demonstrated significant potential.

Conclusions:

• The proposed automated method for sex classification from patellae is effective and accurate.
• Neural network classification, particularly feedforward backpropagation, shows superior performance.
• This technique holds promise for sex determination in forensic anthropology, even with incomplete skeletal remains.