AIR Recon DL

AIR Recon DL is a deep learning-based software feature for GE Healthcare MRI systems that aims to improve image quality by reducing noise and truncation artifacts. Its methodology involves applying trained convolutional neural networks to raw MR data, resulting in enhanced signal-to-noise ratio, image sharpness, and the potential for significantly shorter scan times across various clinical applications.

AIR Recon DL is an optional software feature for GE Healthcare MR systems designed to enhance magnetic resonance image quality using a deep learning-based reconstruction technique. Its core function is to remove noise and ringing (truncation) artifacts from MR images, which can facilitate reduced scan times and improve diagnostic clarity. This version expands compatibility to PROPELLER and 3D Cartesian acquisitions and is applicable across GE Healthcare's 1.5T, 3.0T, and 7.0T MR systems for all anatomies and patient ages, with or without contrast agents. The underlying technology utilizes convolutional neural networks trained with methods and data similar to its predicate, but now supports a broader range of pulse sequences and field strengths. Performance evaluation included both nonclinical phantom testing and extensive clinical reader studies. Phantom tests confirmed improvements in signal-to-noise ratio (SNR) and image sharpness while preserving low contrast detectability and ensuring Apparent Diffusion Coefficient (ADC) maps were not adversely affected. The clinical study, involving 133 cases across various anatomies, pathologies, and MR systems, showed that AIR Recon DL provided images with equivalent or superior quality regarding apparent SNR, sharpness, and lesion conspicuity compared to conventional images. Radiologists indicated a preference for AIR Recon DL images in 99% of evaluations, and images acquired with shorter scan times using AIR Recon DL were rated as good as or better than those from longer conventional scans. The feature did not significantly alter the appearance of motion artifacts and maintained the accuracy of quantitative measurements such as contrast pharmacokinetics, lesion sizes, and brain volumetry.

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 performance of AIR Recon DL vary across the different field strengths (1.5T, 3.0T, and 7.0T), particularly for specific anatomies or subtle pathologies?; What were the specific architectural details and training data characteristics of the deep learning convolutional neural network used in AIR Recon DL that contribute to its robustness across diverse patient populations and MR sequences?; For the single case where lesion conspicuity was not maintained or improved with AIR Recon DL, what were the image characteristics or pathological features, and how does this inform the limitations of the algorithm?; How does AIR Recon DL handle complex or severe motion artifacts compared to conventional reconstruction, beyond the observation that it does not 'significantly change' their appearance?; What are the precise quantitative limits of agreement for measurements like contrast pharmacokinetics, lesion sizing, and brain volumetry when comparing AIR Recon DL processed images to conventionally reconstructed images, especially near the threshold of clinical decision-making?