1275

Deep Learning Enables 60% Accelerated Volumetric Brain MRI While Preserving Quantitative Performance – A Prospective, Multicenter Trial
Suzie Bash1, Long Wang2, Chris Airriess3, Sara Dupont2, Greg Zaharchuk4, Enhao Gong2, Tao Zhang2, Ajit Shankaranarayanan2, and Lawrence Tanenbaum5
1Neuroradiology, RadNet, Woodland Hills, CA, United States, 2Subtle Medical, Menlo Park, CA, United States, 3Cortechs.ai, San Diego, CA, United States, 4Stanford University, Stanford, CA, United States, 5RadNet, New York, NY, United States

Synopsis

In this prospective, multireader, multicenter study, we explore the impact of deep learning (DL) enhancement of 60% accelerated 3D T1 weighted brain MR image acquisitions. We found that the DL processed images demonstrated high volumetric quantification accuracy, matched clinical disease status predictability, and provided what readers perceived as superior image quality when compared with the longer standard-of-care exams, suggesting good generalizability, accuracy, and potential utility of DL enhancement in routine clinical settings. The results of this trial support the use of DL enhancement to shorten clinical MR brain examinations, even when additional quantitative tools such as volumetric analysis are applied.

PURPOSE

In this prospective, multicenter, multireader study, we evaluate the impact on both image quality and quantitative image analysis consistency (NeuroQuant™) of 60% accelerated volumetric MRI scans processed with a commercially available, vendor agnostic, DICOM-based, deep learning (DL) tool (SubtleMR™) compared to that of standard of care (SOC).

MATERIALS AND METHODS

40 subjects underwent brain MRI exams on 6 scanners from 5 institutions. SOC and accelerated (FAST) datasets were acquired for each subject and FAST scans were enhanced with DL processing (FAST-DL). Both SOC and FAST-DL were subjected to NeuroQuant™ quantitative analysis and classified by a neuroradiologist into clinical disease categories. Concordance of SOC and FAST-DL biomarker measurements were assessed. To evaluate image quality, randomized, side-by-side, multiplanar datasets (360 series) were presented blinded to 2 neuroradiologists and rated for apparent signal-to-noise ratio (SNR), image sharpness, artifacts, lesion conspicuity, image contrast, and gray-white differentiation.

RESULTS

FAST-DL was statistically superior to SOC for perceived quality across all imaging features despite a 60% scan time reduction. Both FAST-DL and SOC were superior to FAST for all features. There was no difference in quantitative volumetric biomarkers or clinical classification for SOC and FAST-DL datasets.

CONCLUSION

DL reconstruction allows 60% scan time reduction while maintaining high volumetric quantification accuracy, consistent clinical classification, and what radiologists perceive as superior image quality when compared with SOC. This trial supports the reliability, efficiency, and utility of DL based enhancement for quantitative imaging. Shorter scan times may boost utilization of volumetric quantitative MRI in routine clinical settings.

Acknowledgements

No acknowledgement found.

References

1. Bash S. Eye on AI: Enhancing Neuroimaging with Artificial Intelligence. Appl Radiol. 2020; 49(1):20-21. https://appliedradiology.com/articles/enhancing-neuroimaging-with-artificial-intelligence

2. Tanenbaum LN, Bash S, Davis M. Appl Radiol (AR Connect Expert Discussions). AI: Clinical Applications. In: Proceedings of RSNA 2019, Chicago, IL. Dec 2020. https://www.appliedradiology.com/Communities/Artificial-Intelligence

3. Lunderwold AS, Lunderwold A. An overview of deep learning in medical imaging focusing on MRI. Elsevier 2019; 29:102-127. DOI: https://doi.org/10.1016/j.zemedi.2018.11.002

4. Tian C, Xu Y, Li Z, et. al. Attention-guided CNN for image denoising. Neural Netw 2020;124: 117-129. DOI: 10.1016/j.neunet.2019.12.0241

Figures

FIG 1. Linear Regression Results. The plot graphs demonstrate linear distribution without scatter, indicating consistent concordance between SOC and FAST-DL in quantitative assessment of HOC, HV, SLV volume and ILV volume.

FIG 2. Bland-Altman Results. The plot graphs demonstrate linear distribution without significant scatter, indicating consistent concordance between SOC and FAST-DL in quantitative assessment of HOC, HV, SLV volume and ILV volume.

FIG 3. Representative 3D T1W multiplanar images with volumetric segmentation on a 3T scanner. [Left to right]: Sagittal, coronal, axial T1W images with SOC (scan time 5:01 min) on the top row and FAST-DL (scan time 2:37 min) on bottom row.

FIG 4. Representative axial 3D T1W images on a 3T scanner. [Left to right]: SOC (scan time 9:13 min), FAST (scan time 4:36 min), FAST-DL (scan time 4:36 min).

Table 1: Wilcoxon Rank Sum Test Results. Both readers combined. SOC is superior to FAST for all criteria (p-value <0.05). FAST-DL is superior to SOC for all criteria (p-value <0.05). FAST-DL is superior to FAST for all criteria (p-value <0.05).

Table 2: Paired T-test. There was excellent agreement between SOC and FAST-DL for quantitative assessment of HOC, HV, SLV volume and ILV volume.

Proc. Intl. Soc. Mag. Reson. Med. 29 (2021)
1275