Sonic DL

Sonic DL is an advanced software feature utilizing a deep learning algorithm to reconstruct high-quality cardiac MR images from significantly under-sampled data. This methodology enables highly accelerated acquisition protocols, including single-heartbeat and free-breathing cine imaging, thereby reducing scan times while maintaining diagnostic accuracy for functional cardiac assessments on GE 1.5T and 3.0T systems.

Sonic DL is a software feature developed by GE Healthcare for their 1.5T and 3.0T MR systems, designed to significantly accelerate cardiac cine examinations. Its core technology involves a deep learning-based reconstruction algorithm that processes highly under-sampled magnetic resonance data, enabling much faster image acquisition compared to conventional methods like ASSET acceleration. The software is an optional, key-activated component integrated into the MR system, intended for cardiac imaging across all patient age groups, including those requiring free-breathing scans, with the goal of reducing scan times while maintaining image quality. The performance of Sonic DL was validated through extensive nonclinical and clinical testing. Nonclinical studies using simulated phantom data demonstrated its ability to reconstruct high-quality images with good temporal sharpness from under-sampled data, comparable to fully sampled images. Clinical studies, involving over 100 image series from patients (including those with various cardiac pathologies) and healthy volunteers, confirmed these findings. Reader studies showed that Sonic DL images maintained diagnostic quality and yielded functional cardiac measurements (e.g., left ventricular volumes, ejection fraction, cardiac output) with variability smaller than or comparable to conventional techniques, while significantly reducing scan times. Notably, Sonic DL enables acceleration factors as high as 12, facilitating single-heartbeat imaging even during free breathing, with results showing close agreement to traditional breath-hold methods and, in some cases, improved image quality over conventional free-breathing techniques.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the deep learning model in Sonic DL generalize across diverse patient populations with varying cardiac pathologies not extensively covered in the 57-subject clinical trial?; What are the specific computational requirements (e.g., processing time, hardware) for the Sonic DL reconstruction algorithm, and how do these impact workflow efficiency in a clinical setting compared to conventional reconstruction?; Beyond the evaluated metrics (PSNR, RMSE, SSIM, MAE, functional measurements), are there any subtle image artifacts or alterations introduced by the deep learning reconstruction that might affect the diagnosis of rare or atypical cardiac conditions?; How does the performance of Sonic DL vary between 1.5T and 3.0T MR systems, particularly concerning image quality and achievable acceleration factors for different cardiac sequences?; What is the long-term robustness and stability of the Sonic DL algorithm, and what mechanisms are in place for retraining or updating the model as new data or imaging techniques emerge?