Bias Reduction Using Expectation Maximization in the Optimization of an AI-Assisted Muon Tomography System

Muon scattering tomography is a non-invasive imaging technique that exploits cosmic ray muons to visualise the interior of dense objects. We implemented an Expectation Maximisation (EM) algorithm within a modern muography reconstruction framework and compared its performance with established methods (PoCA, BCA, ASR). Using simulated data, EM produced reconstructions with clearer density contrast in the XY projection and demonstrated potential for bias reduction. However, the method exhibited systematic deviations in high-density regions, poor performance in XZ and YZ projections, and high computational cost. These findings highlight both the promise of EM for enhanced imaging and the need for further optimisation, including improved parameter tuning, noise mitigation, and integration with real experimental data.