NeuroQuant

NeuroQuant is an advanced medical image processing software that utilizes automated algorithms, including probabilistic atlases and deep learning models, to perform volumetric quantification of brain structures and lesions from MRI data. Its purpose is to provide clinicians and researchers with precise, objective measurements and visualizations, such as segmented images and morphometric reports, to aid in the assessment of neurological conditions by analyzing T1-weighted, T2 FLAIR, and T2* GRE/SWI sequences.

NeuroQuant is an automated MR imaging post-processing software designed for the quantitative analysis of brain structures and lesions. Its primary purpose is to provide automatic labeling, visualization, and volumetric measurements from various MR image series, including 3D T1, T2 FLAIR, and T2* GRE/SWI sequences. The software functions as a computing appliance, ingesting DICOM-formatted MRI scans from various manufacturers and correcting for image acquisition artifacts like gradient nonlinearities and bias field inhomogeneity. It then identifies and corrects baseline intensity levels for gray and white matter to account for scanner variability, aligns the scan with an internal anatomical atlas, and uses probabilistic methods alongside neural network models to label each voxel anatomically. The output, delivered in DICOM format, includes segmented images, morphometric reports with numerical volumes, grayscale intensity maps, and color overlays, which can be compared to age and gender-matched normative percentile data. The scientific basis of NeuroQuant involves leveraging MR image intensity patterns with a dynamic probabilistic neuroanatomical atlas, which is age and gender-specific. A significant enhancement includes the integration of static deep-learning technologies, featuring three distinct models: Brain Segmentation, FLAIR Lesion Segmentation, and Microhemorrhage (MCH) Detection. The Brain Segmentation model demonstrated high accuracy and reproducibility using the Dice Similarity Coefficient (DSC). The FLAIR Lesion Segmentation model, designed to identify hyperintensities indicative of conditions like multiple sclerosis, achieved a mean DSC of 0.70, exceeding its performance target. The MCH Detection model, aimed at detecting cerebral hypointensities associated with blood products (e.g., microbleeds, ARIA-H), surpassed its acceptance criteria with a median F1 Score of 0.60. These models were trained and validated on diverse datasets, ensuring robustness and generalizability for clinical and research applications.

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 new deep-learning models for brain segmentation, FLAIR lesion segmentation, and MCH detection compare specifically to other commercially available or research-grade algorithms not mentioned as predicates, particularly in diverse patient populations with varying degrees of pathology?; What is the sensitivity and specificity of the MCH Detection Model for different sizes and locations of cerebral microbleeds, and how does its performance vary across different T2* GRE/SWI sequence parameters from various MRI vendors?; Given the "static" nature of the deep-learning technologies, what is the planned update cycle or strategy for retraining these models to incorporate new data, evolving imaging techniques, or to improve performance on underrepresented pathologies or demographics?; How does NeuroQuant's artifact correction for gradient nonlinearities and bias field inhomogeneity impact the accuracy of volumetric measurements and segmentation, especially in scans of lower quality or from older MRI systems?; What are the specific quantitative differences in volumetric measurements and lesion detection when using only the 3D T1 sequence versus the combined use with T2 FLAIR and T2* GRE/SWI sequences, and how does this impact clinical interpretation for specific neurological conditions?