RECOMPOSE – Reproducing DECOMPOSE Using Susceptibility Maps Acquired for Clinical Research

Here, we reproduced the results of the DECOMPOSE susceptibility separation model using new $T₂^*$-weighted data independently acquired using three clinically applicable sequences and processed with different QSM pipelines. This allowed us to investigate the sensitivity of DECOMPOSE to various dipole inversion algorithms. Good susceptibility source separation results were achieved using a 5-echo GRE acquisition, but maps of diamagnetic and paramagnetic sources from a highly accelerated 5-echo EPI sequence were noisy. When the input susceptibility maps exhibited artefacts, these were exacerbated by DECOMPOSE. Care must be taken not to lose local structural information when using (highly) regularised input susceptibility maps.