Deep Learning Reconstruction of Accelerated MRI: False-Positive Cartilage Delamination Inserted in MRI Arthrography...

Area of Science:

• Radiological imaging
• Artificial intelligence in medicine
• Magnetic resonance imaging (MRI)

Background:

• The radiological imaging industry is increasingly adopting artificial intelligence (AI) software solutions.
• Deep learning reconstruction techniques offer potential for accelerating MRI data acquisition through undersampling.
• Reduced MRI acquisition times could enhance machine utility and operational cost-efficiency.

Purpose of the Study:

• To evaluate the impact of deep learning reconstruction on MRI data quality.
• To investigate potential artifacts introduced by AI-driven image reconstruction.

Main Methods:

• Case study involving magnetic resonance arthrography of a right hip joint.
• Utilized a 30-year-old male patient with no significant health issues.
• Employed deep learning reconstruction for undersampled MRI data.

Main Results:

• Deep learning reconstruction of undersampled MRI data can introduce false-positive findings.
• Specifically, false-positive cartilage delamination and diffuse cartilage defects were observed.

Conclusions:

• The precision of novel AI-based MRI reconstruction technologies requires rigorous validation.
• Systematic bias, particularly stemming from training data selection, must be thoroughly assessed in future evaluations.