Date of Award

12-2025

Document Type

Thesis

Degree Name

Master of Science in Computer Science

Department

School of Computer Science and Engineering

First Reader/Committee Chair

Dajani, Khalil

Abstract

This thesis presents a novel application of deep learning to the estimation of pulmonary vein coordinates using X-ray image pairs from a FORBILD Thorax phantom derived motion dataset. A Siamese neural network was developed to predict the 3D coordinates of one pulmonary vein at a time, specifically the Right Superior Pulmonary Vein (RSPV), Left Superior Pulmonary Vein (LSPV), Left Inferior Pulmonary Vein (LIPV), or Right Inferior Pulmonary Vein (RIPV), based on two-dimensional projection images.

The input data consisted of over 1.6 million grayscale X-ray image pairs across 1331 virtual patients, each annotated with ground truth 3D coordinates. To manage memory constraints, the dataset was partitioned, and model training was conducted iteratively in batches. Testing was performed on a held-out group of 121 patients, with accuracy evaluated by measuring the Euclidean distance between predicted and true coordinates.

Among the trained models, the best-performing one achieved a Mean Absolute Error (MAE) of 2.49 cm in predicting 3D vein coordinates, which, while currently not clinically sufficient, indicates the network’s potential. The limited performance is possibly due to the short training duration (10 epochs) and overfitting. Significant improvements may be possible with extended training over hundreds of epochs. These findings demonstrate the viability of the Siamese network architecture for anatomical localization tasks in 3D medical imaging and provide a baseline for further refinement.

Download

  Contact Author

Share

COinS