DL-driven DLCSC reconstruction improves the reliability and accuracy of the ZTE sequence for detecting MM lesions, with accuracy ranging 80–93% and lesion detection increased by 25–30% compared to native ZTE and BB sequences.
Key Findings
Results
ZTE-DLCSC demonstrated better repeatability (intra-reader agreement) than native ZTE and BB sequences.
Repeatability was 'at least moderate' for ZTE (AC1 ≥ 0.45), 'good' for ZTE-DLCSC and BB (AC1 ≥ 0.60), and 'very good' for CT (AC1 ≥ 0.80).
Agreement was measured using Gwet's AC1 coefficient across all bone regions.
Ten bone regions per patient were assessed for lesion presence and number.
Three readers evaluated the sequences: a senior radiologist with 26 years' experience, a radiology fellow, and a resident.
Results
ZTE-DLCSC demonstrated better reproducibility (inter-reader agreement) than native ZTE and BB sequences.
Reproducibility was 'at least fair' for ZTE and BB (AC2 ≥ 0.20), 'good' for ZTE-DLCSC (AC2 ≥ 0.60), and 'very good' for CT (AC2 ≥ 0.80).
Agreement was measured using Gwet's AC2 coefficient across all bone regions.
CT served as the reference standard for all comparisons.
The study enrolled 10 patients aged 67 ± 12 years (mean ± standard deviation) with multiple myeloma.
Results
ZTE-DLCSC achieved substantially higher diagnostic accuracy than native ZTE for all reader levels.
Accuracy of ZTE-DLCSC ranged 80–93% across readers.
Compared to native ZTE, accuracy increased by 23% for one junior reader (p = 0.010) and by 25% for the other junior reader (p = 0.002).
Accuracy increased by 32% for the senior reader compared to native ZTE (p < 0.001).
CT was used as the reference standard for accuracy calculations.
Results
ZTE-DLCSC detected more osteolytic myeloma lesions than both native ZTE and gradient-echo black bone sequences.
ZTE-DLCSC detected 30% more lesions than native ZTE (p = 0.011).
ZTE-DLCSC detected 25% more lesions than the BB sequence (p = 0.024).
ZTE-DLCSC had fewer false positives and false negatives than both ZTE and BB sequences.
Lesion counts were assessed per region across 10 patients imaged at 3-T whole-body MRI covering lumbar spine, pelvis, and proximal femurs.
Methods
ZTE-DLCSC images were reconstructed from raw ZTE data using deep learning combined with chemical shift correction.
ZTE-DLCSC reconstruction was applied to raw data acquired during standard ZTE acquisition, requiring no additional scan time.
The study compared three MRI sequences: native ZTE, ZTE-DLCSC, and gradient-echo black bone (BB).
Patients underwent whole-body 18F-FDG PET/CT as clinical reference, followed by 3-T whole-body MRI.
The study was prospective, single-center, and registered at ClinicalTrials.gov (NCT05381077).
Results
Native ZTE showed the lowest repeatability and reproducibility among the MRI sequences evaluated.
Native ZTE repeatability was only 'at least moderate' (AC1 ≥ 0.45), compared to 'good' for ZTE-DLCSC and BB.
Native ZTE reproducibility was only 'at least fair' (AC2 ≥ 0.20), compared to 'good' for ZTE-DLCSC.
The gradient-echo black bone sequence showed the same level of repeatability as ZTE-DLCSC (AC1 ≥ 0.60) but lower reproducibility (AC2 ≥ 0.20, same as native ZTE).
CT consistently outperformed all MRI sequences with 'very good' repeatability and reproducibility (AC1 ≥ 0.80; AC2 ≥ 0.80).
What This Means
This research suggests that a new way of processing a type of MRI scan — called Zero Echo Time MRI with deep learning reconstruction and chemical shift correction (ZTE-DLCSC) — can significantly improve the detection of bone damage caused by multiple myeloma, a blood cancer that often destroys bone tissue. The study compared this enhanced MRI technique to standard versions of the same scan (native ZTE) and another MRI method (gradient-echo black bone), using CT scans as the gold standard reference. Ten myeloma patients were scanned at a single hospital, and three radiologists with different levels of experience reviewed the images.
The results showed that ZTE-DLCSC was substantially more accurate than the standard ZTE scan, improving accuracy by 23–32% depending on the reader, and it detected 25–30% more lesions than the other MRI techniques. It also produced more consistent results — both when the same reader reviewed images twice (repeatability) and when different readers compared their findings (reproducibility). This is important because inconsistency between readers is a common challenge in radiology, and a more reliable technique could reduce missed or incorrectly identified lesions.
This research suggests that applying deep learning reconstruction and chemical shift correction to ZTE MRI data — without requiring additional scanning time — could make MRI a more effective tool for monitoring myeloma bone disease. This could be particularly valuable as MRI avoids the radiation exposure associated with CT scans, and improvements in lesion detection could influence treatment decisions for myeloma patients. The study was small (10 patients), so larger studies would be needed to confirm these findings across broader clinical settings.
Lepot D, Chabot C, Duchêne G, Mandava S, Fung M, Poujol J, et al.. (2026). Zero echo time MRI with deep learning reconstruction and chemical shift correction for detecting osteolytic myeloma lesions.. European radiology experimental. https://doi.org/10.1186/s41747-026-00734-x