Deep learning for magnetic resonance vessel wall image: image reconstruction, stenosis diagnosis and plaque calculation

Motivation: Studies have rarely investigated image reconstruction, stenosis detection and plaque calculation from magnetic resonance vessel wall images(MRVWI) because accurate interpretation is error prone and labor-intensive. Goal(s): To develop an automated CNN-based method for accurate image reconstruction, stenosis detection and plaque calculation in MRVWI and compare its performance with radiologists. Approach: A deep learning algorithm was constructed and trained using MRVWI were collected from four tertiary hospitals. Results: The overall reconstruction achieves a qualification rate of 92.3%. This algorithm reduces the diagnosis time of radiologists from 22.08min to 12.79 min.There was high agreement between the algorithm and radiologists on plaque calculation (kappa=.8) Impact: A deep learning algorithm for magnetic resonance vessel wall interpretation accurately determined image reconstruction, vessel stenosis and plaque calculation, which achieved automatic postprocessing and had equivalent diagnostic performance when compared with experienced radiologists.