Deep learning reconstruction enhances bone visualization in zero echo time MRI for cervical spondylosis: A...

Area of Science:

• Medical Imaging
• Artificial Intelligence in Medicine
• Radiology

Background:

• Cervical spondylosis assessment often requires detailed visualization of osseous and soft tissue structures.
• Conventional MRI techniques may face limitations in balancing image quality and scan time for comprehensive evaluation.

Purpose of the Study:

• To evaluate the efficacy of deep learning-based reconstruction (DLR) in zero echo time (ZTE) MRI for improving image quality and reducing scan duration in patients with cervical spondylosis.

Main Methods:

• Forty-three patients with cervical spondylosis underwent ZTE MRI with varying numbers of excitations (NEX).
• Images were reconstructed conventionally and using DLR (NEX3-DL).
• Quantitative (SNR, CNR) and qualitative (bone depiction, artifacts, noise, overall quality) assessments were performed by radiologists, with statistical analysis including non-inferiority and superiority testing.

Main Results:

• DLR-enhanced ZTE MRI (NEX3-DL) significantly improved SNR and CNR compared to conventional reconstruction (NEX3), achieving performance comparable to higher NEX scans (NEX8).
• NEX3-DL scans were 62% shorter than NEX8 scans, with both NEX3-DL and NEX8 showing superior subjective image quality over NEX3.
• High inter- and intra-reader reliability (κ or ICC > 0.6) was observed for all assessments.

Conclusions:

• Deep learning-based reconstruction effectively enhances ZTE MRI for cervical spondylosis, improving bone visualization and overall image quality.
• DLR-enhanced ZTE MRI offers a time-efficient, radiation-free solution for comprehensive osseous and soft tissue evaluation.
• This technique is suitable for integration into routine MRI protocols, streamlining clinical workflow.