Medical image upsampling algorithm to reduce errors in grid-based ultrasound simulation.

Abstract

Clinical imaging resolution is often insufficient for accurate ultrasound simulation in personalised treatment planning. Naive image upsampling algorithms introduce artifacts that reduce simulation accuracy. We develop a Gaussian smoothing and mesh-based upsampling method and evaluate upsampled image fidelity and ultrasound simulation accuracy using an ex vivo high resolution human cervical vertebra volume obtained with hierarchical phase-contrast tomography, against linear and nearest neighbour interpolation. The mesh-based method significantly and consistently reduced L2 and L∞ errors and reduced focal errors with inconsistent significance. Medical image upsampling for ultrasound simulation should be completed with edge-aware methods to improve accuracy.