Enhanced muscle MRI using deep learning: shorter acquisition time with improved image quality

This study aimed to evaluate the effectiveness of a k-space learning type deep learning (DL) reconstruction combined with fat-suppressed turbo spin-echo (TSE) T2-weighted imaging (T2WI) in improving image quality and reducing acquisition time for muscle magnetic resonance imaging (MRI).

In this prospective study, 98 controls (mean age 56.3 years) and 33 patients (mean age 45.2 years) underwent both DL reconstructed TSE (TSE_DL) and standard fat-suppressed TSE T2WI scans of the bilateral thigh using a 3T MRI scanner. Quantitative metrics, including noise, signal-to-noise ratio (SNR), and contrast-to-noise ratio (CNR), were measured. Two radiologists performed qualitative assessments using a 5-point Likert scale to evaluate image quality, anatomical structure visibility, and diagnostic confidence. Inter-reader agreement and statistical comparisons between groups were analyzed.

The acquisition times were 2 min 11s and 1 min 33s for standard TSE and TSE_DL, respectively. TSE_DL demonstrated significantly lower noise and higher SNR and CNR than standard TSE in both healthy volunteers and patients (p < 0.05). Qualitative assessments showed that TSE_DL provided superior overall image quality, better visualization of thigh muscles and femoral bones, sharper edges, improved contrast resolution, and higher diagnostic confidence compared to standard TSE (p < 0.01).

DL during image acquisition reconstruction, combined with fat-suppressed TSE T2WI, significantly enhances image quality and reduces acquisition time in muscle MRI, suggesting its potential for routine clinical use in musculoskeletal imaging.