Age and Gender Estimation from Clavicle CT Images

As my undergraduate senior thesis, I participated in a comprehensive deep learning research project on automatic age and gender estimation from clavicle CT images for forensic identification.

The research was conducted as a two-person team in collaboration with the Embedded Systems Laboratory and researchers from Kocaeli University Faculty of Medicine using an ethically approved clinical CT dataset. Rather than building a single machine learning model, our objective was to design and evaluate an end-to-end research pipeline covering every stage of the experimentation process, from literature review and data preparation to model development, experimentation, and scientific evaluation.

The research began with an extensive literature review on forensic age estimation methods, skeletal maturation, and recent advances in medical computer vision. Based on these findings, I designed a reproducible experimentation workflow that included medical image preprocessing, ROI extraction around the medial clavicular epiphysis, data augmentation strategies, dataset balancing, experiment tracking, and systematic evaluation.

Throughout the project, I investigated multiple approaches including 2D, 2.5D, and volumetric 3D representations while comparing different deep learning architectures and training strategies. Various preprocessing methods, ROI sizes, augmentation techniques, and model configurations were evaluated through ablation studies to better understand how each design decision affected overall performance.

The project evolved into a complete research framework rather than a single model implementation. Every experiment was documented, evaluated, and compared using quantitative metrics such as accuracy, AUC, precision, recall, F1-score, confusion matrices, and cross-validation results to ensure reproducibility and scientific rigor.

Beyond the machine learning pipeline, the project also involved continuous collaboration with academic researchers on experimental methodology, interpretation of results, and preparation of an academic publication. The resulting work serves both as my undergraduate senior thesis and as the foundation for an upcoming medical AI research paper.