Deep Learning Image Reconstruction

The Deep Learning Image Reconstruction software employs a dedicated Deep Neural Network to process raw Computed Tomography data, aiming to produce high-quality 'TrueFidelity CT Images' with an appearance similar to traditional methods while matching or improving upon ASiR-V's performance in key image quality metrics. This technology, now ported to the Revolution EVO system and deployed via GE's Edison Platform, relies on advanced computational algorithms trained for optimal CT image generation across a range of clinical applications.

The Deep Learning Image Reconstruction (DLIR) software is an advanced image processing tool intended for use with GE's Computed Tomography (CT) systems, specifically ported to the Revolution EVO platform. It leverages a dedicated Deep Neural Network (DNN), meticulously trained to reconstruct CT images from raw X-ray transmission data. The primary objective of DLIR is to generate 'TrueFidelity CT Images' that offer an image appearance akin to traditional Filtered Back Projection (FBP) methods, while simultaneously matching or exceeding the performance of GE's ASiR-V reconstruction in key areas like image noise, low contrast detectability, high-contrast spatial resolution, and streak artifact suppression. The system provides users with three selectable strengths (Low, Medium, High) to adapt reconstruction to specific clinical needs for head, whole body, cardiac, and vascular CT scans across all age groups, and is deployed on GE's Edison Platform. The DLIR software's efficacy on the Revolution EVO system was validated through comprehensive non-clinical and clinical evaluations. Non-clinical bench testing compared DLIR and ASiR-V using identical raw datasets, assessing parameters such as low contrast detectability (LCD) with phantoms, image noise, high-contrast spatial resolution (MTF), streak artifact suppression, longitudinal spatial resolution, Noise Power Spectrum (NPS), CT number accuracy, contrast-to-noise ratio (CNR), artifact analysis, pediatric image quality, and low-dose lung cancer screening performance. A clinical reader study involving 60 retrospective cases had seven board-certified radiologists independently evaluate images reconstructed by both DLIR and ASiR-V. They assessed overall image quality and directly compared the methods on image noise texture, sharpness, and noise texture homogeneity, with results supporting the software's claimed performance benefits.

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 the three user-selectable strengths (Low, Medium, High) of Deep Learning Image Reconstruction compare quantitatively across different anatomical regions and patient populations, particularly in terms of noise reduction versus preservation of fine diagnostic details?; What specific architectural details of the Deep Neural Network (DNN) contribute to its ability to maintain ASiR-V's performance in key metrics while achieving an FBP-like image appearance, and how was the training dataset curated to achieve this specific outcome?; Beyond general streak artifact suppression, are there specific types of challenging artifacts (e.g., those arising from dense metal implants or significant patient motion) where DLIR demonstrates a more marked improvement or potential new limitations compared to ASiR-V or traditional FBP methods on the Revolution EVO platform?; How does deployment on GE's Edison Platform, as opposed to direct integration on the CT console, impact image reconstruction times, data handling, and the potential for remote updates or scalability of the DLIR algorithm across different scanners?; What were the specific quantitative scores or statistical differences observed in the clinical reader study for 'image noise texture,' 'image sharpness,' and 'image noise texture homogeneity' when comparing DLIR to ASiR-V, and did these preferences translate to improved diagnostic confidence or task performance for specific clinical indications?