A convolutional neural network to filter artifacts in spectroscopic MRI.

摘要: PURPOSE Proton MRSI is a noninvasive modality capable of generating volumetric maps in vivo tissue metabolism without the need for ionizing radiation or injected contrast agent. Magnetic resonance spectroscopic imaging has been shown to be viable studying several neuropathologies. However, key hurdle routine clinical adoption presence spectral artifacts that can arise from number sources, possibly leading false information. METHODS A deep learning model was developed identifying and filtering out poor quality spectra. The core used tiled convolutional neural network analyzed frequency-domain spectra detect artifacts. RESULTS When compared with panel MRS experts, our achieved high sensitivity specificity an area under curve 0.95. visualization scheme implemented better understand how made its judgement on single-voxel multivoxel MRSI, embedded into pipeline producing whole-brain MRI volumes real time. CONCLUSION fully automated method assessment provides valuable tool support studies use fields such as adaptive therapy planning.