Back to the list
Congress: ECR25
Poster Number: C-18416
Type: Poster: EPOS Radiologist (scientific)
DOI: 10.26044/ecr2025/C-18416
Authorblock: O. N. Samuel, C. Saidlear, E. L. Twomey, *. Chi Temple St Group, *. Chi Crumlin Group, J. Cooke, M. Kelly; Dublin/IE
Disclosures:
Osahenre Noelle Samuel: Nothing to disclose
Colm Saidlear: Nothing to disclose
Eilish L. Twomey: Nothing to disclose
* Chi Temple St Group: Nothing to disclose
* Chi Crumlin Group: Nothing to disclose
Jennie Cooke: Nothing to disclose
Michael Kelly: Nothing to disclose
Keywords: Artificial Intelligence, MR physics, Paediatric, MR, Diagnostic procedure, Efficacy studies, Physics, Education and training, Image verification, Quality assurance
Purpose Magnetic resonance imaging (MRI) is the preferred imaging modality for paediatric medical conditions due to its high spatial resolution, soft tissue characterisation and lack of ionising radiation. However, its main limitation is long scan durations due to the intrinsic relationship between SNR, resolution and scan time[1,2]. The project assesses the performance of GE Healthcare’s deep learning-based reconstruction method, Air ReconTM DL (ARDL), for paediatric brain scanning. ARDL reduces image noise to the desired level: ‘Low’, ‘Medium’ or ‘High’, while improving...
Read more Methods and materials Study ParticipantsEthical approval was obtained through the Children’s Health Ireland Research Ethics Committee (REF: rec-398-24). MRI datasets from GE Signa Architect 3T MRI scanner were divided into 3 groups: (1) no evident pathology, (2) subtle pathology and (3) major pathology. T2 fast spin echo (FSE) brain scans, both ARDL reconstructed and the corresponding conventionally reconstructed version (ORIG), were anonymised and exported. ARDL images acquired were set to ‘High’ noise removal. DICOM headers were edited to allow blinding of the image...
Read more Results Quantitative AnalysisThe paired samples t-test was used to statistically analyse all quantitative results between the two reconstruction groups.Noise was observed to have decreased by a factor of 2.22. Signal was found to be slightly higher in conventionally reconstructed scans than in ARDL reconstructed scans. Resulting SNR values show an overall significant difference in means between reconstruction types, with ARDL reconstructed images exhibiting higher SNR, (Fig .5). [fig 5] Similarly, difference in CNR between reconstruction methods was statistically significant, with CNR being higher...
Read more Conclusion As there is an evident lack of studies regarding the use of deep learning-based reconstruction methods in the paediatric cohort, this study aimed to retrospectively assess GE Healthcare’s novel algorithm, Air ReconTM DL, offering a methodology to the wider radiology community for performing these validation studies through a combined quantitative and qualitative approach. Due to the nature of this retrospective study, a proper comparison between ARDL reconstructed images and conventional reconstructed images was not possible. The images which formed part...
Read more References [1] Lebel RM. Performance characterization of a novel deep learning-based MR image reconstruction pipeline. ArXiv. 2020;abs/2008.06559.[2] Harris H, Lawson S, Peters RD. The clinical benefits of AIR™ Recon DL for MR image reconstruction. GE Healthcare. 2020.[3] Allen TJ, Henze Bancroft LC, Unal O, Estkowski LD, Cashen TA, Korosec F, Strigel RM, Kelcz F, Fowler AM, Gegios A, et al. Evaluation of a Deep Learning Reconstruction for High-Quality T2-Weighted Breast Magnetic Resonance Imaging. Tomography. 2023; 9(5):1949-1964. https://doi.org/10.3390/tomography9050152.[4] Zochowski KC, Tan ET,...
Read more
GALLERY