NEUROShield

NEUROShieldTM is an AI-driven medical image processing software designed for the automated volumetric quantification of the Hippocampus from 3D MR images, intended to function as a clinical decision support tool. It utilizes DeepNet/U-Net methodologies, trained and validated on diverse MRI datasets against radiologist-derived ground truth, to deliver accurate segmentation and volume measurements of the hippocampal structure.

NEUROShieldTM is a medical image processing software designed as a clinical decision support tool for neurologists and neuroradiologists. Its primary function is the fully automated labeling, visualization, and volumetric quantification of the Hippocampus brain structure derived from 3D MR images. The system operates as a cloud-based platform that utilizes Artificial Intelligence, specifically DeepNet/U-Net methodologies, to analyze these images. This automated process replaces traditional, time-consuming manual segmentation by performing segmentation, volume calculation, and generating reports, thereby assisting physicians in developing optimal treatment plans. The algorithm employed by NEUROShieldTM is locked, with any proposed modifications subject to FDA review. The scientific foundation of NEUROShieldTM is rooted in brain volume quantification, an established methodology for the enhanced interpretation of medical images. The AI model was developed by training Deep Net Segmentation Models using a dataset of 186 cases from India, varying in Tesla strength, image quality, and equipment manufacturer, with expert manual segmentation of the Hippocampus serving as the ground truth. The device's performance was subsequently validated on an independent dataset of 280 subjects from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. Ground truth for this validation was established using the STAPLE algorithm, which combined manual segmentations from three US Board Certified Radiologists. Comprehensive statistical analyses, including Dice coefficient, Hausdorff distance, correlation analysis, Bland-Altman plots, and relative volume difference, confirmed NEUROShield's high accuracy, achieving an average Dice coefficient of 0.91 and a Hausdorff distance of 3.8 mm, and consistently meeting predefined pass/fail criteria across diverse subgroups such as clinical status, gender, MRI strength, slice thickness, and US geographical regions.

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 NEUROShield's performance on Hippocampus segmentation compare to its predicate device, NeuroQuant, particularly in terms of Dice coefficient and Hausdorff distance for this specific structure, given NeuroQuant's broader scope?; What is the clinical impact of using NEUROShield on patient management or diagnostic timelines for conditions associated with hippocampal atrophy, beyond general 'optimal treatment plans'?; The training dataset (186 cases from India) and validation dataset (280 ADNI cases from USA) have different demographic origins; how was potential bias from these differing populations addressed or evaluated in the algorithm's generalizability?; While the device passed subgroup error analysis, were there any notable performance variations within specific clinical subgroups (e.g., ADNI-AD vs. ADNI-control) that might warrant further investigation or specific clinical considerations?; The document mentions 'report generation'; what specific quantitative and qualitative information is included in the NEUROShield report, and how is it designed for optimal interpretation by neurologists and neuroradiologists?