MR imaging has been primarily a soft-tissue imaging modality where tissues such as ligaments, tendons, calcifications and cortical bone structures were assessed from the signal void in the image. Such tissues have extremely short T2 relaxation times such that the MR signal has decayed at the time of readout with conventional MR pulse sequences. With recent hardware and software improvements, MR bone imaging is now possible clinically with oZTEo, a zero echo time (ZTE) technique that is a non-Cartesian radial sequence void of echo time (TE). oZTEo allows for differentiation of extremely short T2 species (e.g., cortical bone, calcifications) from other short T2 species (e.g., tendons, ligaments) and other longer T2 species (e.g., muscle, fat). In addition to the extremely short TE needed to retain the signal from extremely short T2 species, ZTE images use low flip angles and are inherently low in T1 and T2 contrast (mostly PD), enabling MR imaging with CT-like bone contrast. With the oZTEo MR bone imaging application, the ZTE MR signal is automatically inverted such that signal from cortical bone and calcified lesions are bright, while all other soft tissue species have homogeneous gray signal.
Initially released in early 2020, oZTEo has seen robust use in musculoskeletal (MSK) applications (e.g., fractures, calcification of tendons, pars defect, shoulder instability) and in pediatric and oncology patients (e.g., osteolytic or sclerotic lesions).
The addition of AIR™ Recon DL
As previously discussed, oZTEo has inherently low SNR (due to the low flip angle used), which can be overcome with the use of additional NEX (oversampled spokes in k-space) to increase SNR. However, this comes at the price of longer scan times. As many MSK MR imaging sites today aim to perform a full knee exam in five minutes, adding a three-minute oZTEo sequence might not be feasible in a fast-paced MR imaging environment. To address these issues, we turned to AIR Recon DL, GE HealthCare’s deep-learning-based reconstruction algorithm for MR imaging that increases SNR and image sharpness, and enables shorter scan times. AIR Recon DL, initially compatible with Cartesian sequences, was released in 2020. The core deep-learning model is being retrained with a new algorithm to ensure compatibility with non-Cartesian, ZTE-based applications like oZTEo to significantly improve SNR.‡ By improving SNR, it will be possible to reduce the NEX—or the oversampling—and substantially reduce the scan time (Figure 1).
Figure 1.
(A) oZTEo with conventional reconstruction and (B) oZTEo with AIR Recon DL reconstruction, NEX=1.5, 2:20 min. SNR and image sharpness are improved with the AIR Recon DL reconstruction.
Reducing chemical shift artifacts
Due to oZTEo’s radial sampling, phase accumulation that occurs during the readout period will translate to chemical shift artifacts. These chemical shift artifacts are particularly prominent at tissue interfaces, such as fat and water or tissue muscle and fat, and in high-resolution protocols where the readout period is long (Figure 2).
Figure 2.
Effect of bandwidth on chemical shift artifacts in ZTE (without DL chemical shift reduction) in images acquired with bandwidth (A) +/-50 kHz, (B) +/-62 kHz and (C) +/-83 kHz. (A) Low bandwidth acquisition results in longer readout duration, which will increase the phase accumulation for fat signal, resulting in more chemical shift artifacts. (C) High bandwidth acquisition results in fewer chemical shift artifacts.
We investigated the use of a deep-learning approach to mitigate these artifacts.1 By leveraging the known chemical shift between fat and water for the particular field strength and acquisition protocol, the DL model has been designed to make the appropriate correction in oZTEo images to reduce the artifacts. The model was trained using a large database of tens of thousands of images and validated in both phantom and in vivo experiments. After applying the correction, the images demonstrate a reduced spurious edge enhancement (Figure 3), indicating a more accurate image for better cortical bone and calcification depiction (Figure 4) and are expected to facilitate easier 3D volume rendering and segmentations (Figure 5).
Figure 3.
Comparison between (A) conventional reconstruction, (B) AIR Recon DL reconstruction, and (C) AIR Recon DL with chemical shift reduction reconstruction. (B, C) While AIR Recon DL improves SNR, (C) the addition of chemical shift correction further improves visualization of spinous process and pedicles.
Figure 4.
(A) Reference CT image, (B) oZTEo with conventional reconstruction and (C) oZTEo with AIR Recon DL and chemical shift reduction. (C) oZTEo with AIR Recon DL and chemical shift reduction results in a more accurate depiction of bony islands (arrows) within the proximal femurs, when compared to (A) reference CT. Images courtesy of Prof. F. E. Lecouvet, Cliniques Universitaires Saint Luc, Brussels, Belgium.
Figure 5.
(A) Conventional oZTEo image, (B) oZTEo with AIR Recon DL and (C) oZTEo with AIR Recon DL and chemical shift reduction. (D-F) 3D reconstructions of (A-C) the oZTE images. (C and F) oZTEo with AIR Recon DL and chemical shift reduction reduces artifact in the muscle/fat junction and increases SNR, resulting in higher quality 3D volume rendering.
oZTEo: background suppression
In the oZTEo application, inverting the ZTE signal also results in a bright background air
signal. This can be distracting to radiologists in a reading room and, more importantly, it can make 3D volume rendering more challenging. Suppressing this background signal is an essential step in enabling easier volume rendering and improving the ease of reading oZTEo images.
This background removal was designed using morphologic image processing algorithms. This ensures that the background air remains dark, while the foreground intensities are appropriately inverted to generate the CT-like bone contrast. We also developed a non-linear histogram adjustment such that the bone and tissue signal in an oZTEo image are closer to the typical CT Hounsfield Unit (HU) values, making these images more compatible with 3D volume rendering software that typically uses CT thresholds. This will facilitate a “one-click” 3D volume rendering for oZTEo images and streamline post-processing applications.2 This is potentially important for future use in advanced visualization and downstream solutions, such as surgical planning and navigation systems.
Figure 6.
(A-C) oZTEo with AIR Recon DL and background suppression demonstrating histogram adjustment, background air segmentation removal, and bone and muscle signal intensities scaled closer to CT HU ranges. (D-F) Corresponding 3D volume rendered images using GE HealthCare’s Advantage Workstation Volume Illumination tool without additional manual segmentation.
Summary
oZTEo is a novel MR imaging sequence enabling new opportunities for bone imaging that helps avoid ionizing radiation in vulnerable populations such as pediatrics, pregnant women or cancer patients, and offers a one-stop-shop MSK imaging exam. The SNR limitations of ZTE applications can be overcome with the addition of AIR Recon DL, which mitigates the need to oversample k-space, thus significantly lowering the scan time—by at least a factor of two—of ZTE-based applications like oZTEo.
These capabilities are important in difficult-to-image anatomies like the spine or small joints and in populations where higher resolution is desired, like pediatrics. It also gives a boost in SNR for centers with 1.5T MR systems so they can more confidently acquire higher quality imaging with short acquisition times.
‡ Technology in development that represents ongoing research and development efforts. Not for sale. Not CE marked. Not cleared or approved by the U.S. FDA or any other global regulator for commercial availability.
References
- Mandava, Sagar, Michael Carl, Florian Wiesinger, Maggie Fung, and R. Marc Lebel. 2024. “Deep learning based chemical shift artifact reduction in Zero Echo Time (ZTE) MRI.” Proceedings of the International Society for Magnetic Resonance in Medicine 32: 4428. https://archive.ismrm.org/2024/4428.html.
- Carl, Michael, Laura Carretero-Gomez, Sagar Mandava, and Maggie Fung. 2025. “1-Click Bone Volume Rendering with ZTE MRI.” Proceedings of the International Society for Magnetic Resonance in Medicine 33: 3118. https://archive.ismrm.org/2025/3118.html.
Bibliography
- Peng, Li, Xuelin Pan, Huilou Liang, et al. 2025. “Deep learning reconstruction enhances bone visualization in zero echo time MRI for cervical spondylosis: A prospective study.” European Journal of Radiology 191: 112310. https://doi.org/10.1016/j.ejrad.2025.112310.
- Yi, Jisook, Seok Hahn, Ho-Joon Lee, et al. 2025. “Deep learning reconstruction of zero echo time magnetic resonance imaging: diagnostic performance in axial spondyloarthritis.” European Radiology. https://doi.org/10.1007/s00330-025-11843-3.
- Lee, Chena, Joonsung Lee, Sagar Mandava, et al. 2025. “Deep learning image enhancement for confident diagnosis of TMJ osteoarthritis in zero-TE MR imaging.” DentoMaxilloFacial Radiology 54 (4): 302-306. https://doi.org/10.1093/dmfr/twae063.
- Kaniewska, Malwina, Fabio Zecca, Carina Obermüller, et al. 2025. “Deep learning reconstruction of zero-echo time sequences to improve visualization of osseous structures and associated pathologies in MRI of cervical spine.” Insights into Imaging 16 (29). https://doi.org/10.1186/s13244-025-01902-0.
- Carretero-Gómez, Laura, Maggie Fung, Florian Wiesinger, et al. 2025. “Deep learning-enhanced zero echo time MRI for glenohumeral assessment in shoulder instability: a comparative study with CT.” Skeletal Radiology 54 (6): 1263-1273. https://doi.org/10.1007/s00256-024-04830-0.
- Ensle Falko, Frederik Abel, Maelene Lohezic, Carina Obermüller, Roman Guggenberger R. 2024. “Deep learning reconstruction for optimized bone assessment in zero echo time MR imaging of the knee.” European Journal of Radiology 179: 111663. https://doi.org/10.1016/j.ejrad.2024.111663.
- Ensle, Falko, Malwina Kaniewska, Maelene Lohezic, Roman Guggenberger. 2024. “Enhanced bone assessment of the shoulder using zero-echo time MRI with deep-learning image reconstruction.” Skeletal Radiology 53 (12): 2597-2606. https://doi.org/10.1007/s00256-024-04690-8.
