a# Dataset for: Spectral X-ray Imaging using Deep Learning

**Author**: Raziye Kubra Kumrular  
**ORCiD**: [0000-0002-0976-3683](https://orcid.org/0000-0002-0976-3683)  
**Affiliation**: University of Southampton, Faculty of Engineering and Physical Science  
**Date of data collection**: March 2021 – November 2024
**Date of file creation**: June 2025  
**License**: CCBY  
**Location**: University of Southampton, UK

---

##  Description

This dataset supports the doctoral thesis:  
**"Spectral X-ray Imaging using Deep Learning"**

It includes simulated and reconstructed spectral X-ray data in `.npy` format, intended for testing the performance of deep learning models and iterative reconstruction algorithms. 
The data were generated using synthetic powder phantoms under various projection and exposure conditions, including reconstructions based on full and half projection subsets.

---

##  File Descriptions

- `PDHG_TGV_5sn_500it.npy`:  
  Reconstructed data using the PDHG algorithm with Total Generalized Variation (TGV) regularization. Acquired with **180 projections**, **5 seconds exposure per projection**, and **500 iterations**.

- `PDHG_TGV_30pr_30sn_500it.npy`:  
  Reconstructed data using the PDHG algorithm with TGV regularization. Acquired with **30 projections**, **30 seconds exposure per projection**, and **500 iterations**.

- `powder_phantom_ground_truth_ratio.npy`:  
  Ground truth ratio image of the synthetic powder phantom used for quantitative validation.

- `powder_phantom_synthetic_even_high_30pr_30sn.npy`:  
  Simulated sinogram with Poisson noise. Includes **30 projections**, each with **30 seconds exposure**.  
  FDK Reconstruction is performed using only the **even-numbered projections** (i.e., half of the data).

- `powder_phantom_synthetic_even_high_180pr_5sn.npy`:  
  Simulated sinogram with Poisson noise. Includes **180 projections**, each with **5 seconds exposure**.  
  FDK Reconstruction is performed using only the **even-numbered projections**.

- `powder_phantom_synthetic_odd_high_30pr_30sn.npy`:  
  Simulated sinogram with Poisson noise. Includes **30 projections**, each with **30 seconds exposure**.  
  FDK Reconstruction is performed using only the **odd-numbered projections**.

- `powder_phantom_synthetic_odd_high_180pr_5sn.npy`:  
  Simulated sinogram with Poisson noise. Includes **180 projections**, each with **5 seconds exposure**.  
 FDK Reconstruction is performed using only the **odd-numbered projections**.

---

## Usage Notes

- All files are saved in `.npy` format and can be loaded using NumPy:
  ```python
  import numpy as np
  data = np.load("filename.npy")