AiMIFY (1.x)

AiMIFY (1.x) is an image processing software that employs a deep learning-based convolutional neural network to enhance brain MRI images acquired with standard-dose gadolinium contrast agents. Its scientific basis lies in algorithmic improvement of key image metrics such as contrast-to-noise ratio, contrast enhancement, and lesion-to-brain ratio by processing T1 pre- and post-contrast DICOM inputs to improve visualization of enhancing tissues for radiologists.

AiMIFY (1.x) is a software as a medical device that utilizes machine learning to enhance Magnetic Resonance Imaging (MRI) images. Its primary function is to improve the contrast-to-noise ratio (CNR), contrast enhancement percentage (CEP), and lesion-to-brain ratio (LBR) in T1-weighted brain MRI images acquired with a standard dose of gadolinium-based contrast agent (GBCA). The system operates as a post-processing tool, taking T1 pre-contrast and T1 post-contrast DICOM images as input. It employs a convolutional neural network (CNN) based algorithm, specifically a UNet style network processing data in a 2.5D fashion, to generate an enhanced output image. AiMIFY integrates into existing clinical workflows by sending the enhanced DICOM images to PACS or other DICOM nodes for review by radiologists in imaging centers, clinics, or hospitals, and notably uses the SubtleMR device as a denoising module. Extensive testing demonstrated AiMIFY's efficacy. Quantitative assessments on 95 diverse brain cases showed significant average improvements: CNR by 559.94% (831.70% for lesion-only cases), LBR by 62.07%, and CEP by 133.29% when compared to standard post-contrast images. A qualitative reader study involving three board-certified neuro-radiologists confirmed that AiMIFY-processed images offered statistically significant improvements in the perceived visibility of lesion contrast enhancement, border delineation, and internal morphology (p<0.0001). Perceived image quality was not worse and often rated better, with artifact presence not negatively impacting diagnosis. While AiMIFY images exhibited a "high-dose-like" appearance and generally good concordance in radiomic features, increased vessel conspicuity was noted in approximately 11.58% of cases, though it was deemed unlikely to impact diagnosis in 90% of instances. The device has been validated for lesion sizes ranging from 0.02 cc to 59.83 cc.

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 AiMIFY's 2.5D UNet-style convolutional neural network architecture balance computational load with the volumetric nature of MRI data for optimal enhancement?; What are the specific image characteristics or patient conditions that most contribute to the 11.58% of cases where vessel conspicuity in AiMIFY-enhanced images was rated as potentially impacting diagnosis, and how will the planned vessel suppression feature address these?; Considering the reported 93% increase in 'dose concentration' appearance, what studies have been or could be conducted to correlate this appearance with actual diagnostic yield compared to true high-dose GBCA protocols, especially concerning subtle or atypical lesions?; The radiomics analysis showed concordance correlation coefficients (CCC) of 0.68 to 0.89 for lesion tissue. Are there specific lesion types or radiomic features within this range that approach the lower bound, and what is the implication for quantitative tumor assessment or therapy response monitoring?; Given that AiMIFY uses standard-dose GBCA input, what is the underlying mechanism by which the algorithm achieves significant increases in CNR, LBR, and CEP, particularly the non-linear scaling of contrast uptake, and how robust is this across varied MRI scanner hardware and software versions?