First experiments in surface-based mechanical property reconstruction of gelatine phantoms A. Peters, S. Wortmann, R. Elliott, M. Staiger, J.G. Chase, E.E.W. Van Houten
Digital Image-based Elasto- Tomography (DIET) aims to be a low- cost alternative to current breast cancer screening modalities Based on elastographic principles and low-cost digital imaging techniques Introduction The DIET System [1] Peters et. al, JSME Int. Journal, (2004) Four major steps in the DIET system Actuate Capture Process Reconstruct Simulation studies undertaken have proven the concept of surface-based mechanical property reconstruction [1]
Cylindrical tissue-approximating gelatine phantoms Actuation achieved using dSPACE TM, laser interferometer, linear voice-coil actuator with amplifier Methods Phantom Studies Motion captured using two consumer- level digital cameras Manually-applied dots on tracked on phantom surface Real motion approximated with a least-squares fitted ellipsoid
Finite Element (FE) model of cylinder created and meshed Actuated with same constraints as real gelatine phantom Sparse parallel direct matrix inversion and solution performed with MUMPS [2] and Goto BLAS [3] Methods FE Simulation Projecting a measured motion point back to the surface of a 3D mesh to allow motion comparison [2] Amestoy et. al, Parallel Computing, (2005) [3]
Forward FE simulation performed at small intervals over a range of homogeneous stiffness values Results Simulated Motion Sample displacement solutions at a range of stiffness values Testing showed 22k node mesh solutions were converged at 10kPa and above
Results Motion Error Sweep
Qualitative comparison made between actual motion and simulated phantom motion at 27kPa Results Direct Comparison Homogeneous gelatine phantom stiffness successfully identified using steady-state motion measurements and a FE model MEASURED SIMULATED 27kPa
Damping and phase Material non-linearity More advanced reconstruction Multiple parameters Gradient-descent Genetic algorithm/simulated annealing Tighter integration of motion capture and processing Acknowledgements PhD supervisors Data collection Jérôme Rouzé & Arnaud Milsant Edouard Ravini & Fabrice Jandet Conclusions Current Challenges