Graduate Students Seminar

Speaker 1: Ehsan Shakeri

Title

Geant4-Based Simulation and Image Reconstruction of Tc-99m for Radiopharmaceutical Imaging

Abstract

We present a framework for simulating and reconstructing radiopharmaceutical images using a Compton camera (CC), with emphasis on U-Net–based enhancement for improving image quality. A Geant4-based Monte Carlo model of a two-stage CC with a Tc-99m source was developed to simulate gamma-ray emissions and interactions across multiple camera orientations. The resulting synthetic data were reconstructed using physics-based algorithms but suffered from noise and blurring due to limited angular sampling.

To mitigate these limitations, we incorporated a U-Net convolutional neural network for denoising the reconstructed images. The U-Net, trained with a hybrid mean squared error (MSE) and structural similarity index measure (SSIM) loss, effectively suppressed noise while preserving fine structural details. These results highlight the promise of combining physics-based reconstruction with U-Net–based post-processing to improve Compton camera image quality and support clinical applications in personalized dosimetry.

Speaker 2: Jhilam Sur

Title

A new coefficient of correlation – Chatterjee's Correlation

Abstract

Usually we compute correlation using Pearson Correlation coefficient or Spearman's rank correlation. These are used to capture linear relationship between explanatory and response variable. Chatterjee's correlation coefficient captures non-linear relationship between explanatory and response variable. It is simple to compute, does not require assumption on the distribution of the variables and has asymptotic property.