Reconstruction of 3D knee MRI using deep learning and compressed sensing: a validation study on healthy volunteers

European Radiology Experimental

Table 1 Acquisition parameters for the different three-dimensional sequences and results for changes in the scan time

Sequence	Standard SENSE	CS 10 CS-AI 10	CS 13 CS-AI 13	CS 15 CS-AI 15	CS 17 CS-AI 17
Echo time [ms]	170	170	170	170	170
Repetition time [ms]	1,300	1,300	1,300	1,300	1,300
Flip angle [degrees]	90	90	90	90	90
Field of view [mm]	140 × 159 × 160	140 × 159 × 160	140 × 159 × 160	140 × 159 × 160	140 × 159 × 160
Slice thickness [mm]	3D sequence, therefore voxel size as volume
Acquisition voxel size [mm]	0.63 × 0.62 × 0.63	0.63 × 0.62 × 0.63	0.63 × 0.62 × 0.63	0.63 × 0.62 × 0.63	0.63 × 0.62 × 0.63
Reconstruction voxel size [mm]	0.30 × 0.30 × 0.63	0.30 × 0.30 × 0.63	0.30 × 0.30 × 0.63	0.30 × 0.30 × 0.63	0.30 × 0.30 × 0.63
Turbo factor/echo train length	63	63	63	63	63
Sense/CS factor	2/ −	− /10	− /13	− /15	− /17
Scan time [s]	384	178	137	121	105
Saved scan time [s]		206	247	263	279
Scan time reduction [%]		54	64	68	73

Please note that only the acceleration factors were changed between the different sequences to keep them as comparable as possible. Also note that the acquisition times are the same for the CS and CS-AI sequences with the same acceleration factor. CS Acceleration using compressed sensing, CS-AI Acceleration using compressed sensing followed by AI image reconstruction using a deep learning-based algorithm