NS-HGlio

NS-HGlio is an advanced software medical device employing deep learning algorithms to facilitate the semi-automatic labeling, visualization, and volumetric quantification of high-grade brain gliomas from multimodal MRI data for clinicians. Its demonstrated high accuracy in tumor segmentation and volume measurement, validated against expert neuroradiologists on diverse pre- and post-operative imaging datasets, supports its utility as a supplementary tool in neuro-oncological assessment.

NS-HGlio is a non-invasive software as a medical device (SaMD) designed for the semi-automatic labeling, visualization, and volumetric quantification of pathologically confirmed high-grade brain gliomas (WHO grade 3 astrocytoma, WHO grade 4 astrocytoma, and WHO grade 4 glioblastoma). Intended for patients 18 years or older, it processes standard MRI sequences (T1 pre-contrast, T1 post-contrast, T2, and FLAIR) using a deep learning methodology to identify and measure tumor subcomponents from DICOM images. The device provides clinicians with an additional tool for assessing disease conditions from pre- and post-operative MRI images, displaying results on Neosoma viewing software or connecting to existing PACS systems, and allows for comparison with past measurements if available; it is not intended for primary diagnosis. The performance of NS-HGlio was evaluated using a dataset of 33 subjects (132 MRIs) acquired from Siemens, GE, and Toshiba scanners, distinct from the training data to ensure generalizability. The ground truth for comparison was established by three U.S. board-certified neuroradiologists. The device demonstrated a mean Dice Similarity Coefficient (DSC) of 0.88 (95% CI: 0.86-0.90) for preoperative imaging and 0.80 (95% CI: 0.77-0.83) for postoperative imaging, exceeding the average expert DSCs of 0.84 and 0.74 for the same tasks, respectively. Furthermore, the mean Intraclass Correlation Coefficient (ICC) for volumetric measurements was 0.98 (95% CI: 0.97-0.99), indicating high reliability in quantification.

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 NS-HGlio deep learning model differentiate between various subcomponents of high-grade gliomas, such as enhancing tumor, necrotic core, and peritumoral edema, and what are the individual performance metrics for each subcomponent?; What were the specific MRI acquisition parameters (e.g., field strength, sequence parameters, slice thickness) for the 132 MRIs in the test dataset, and how robust is the NS-HGlio's performance across different scanner manufacturers and imaging protocols not explicitly listed?; The document mentions the device achieved a mean DSC of 0.88 on preoperative and 0.80 on postoperative imaging. What is the range or standard deviation of these DSC values, and were there specific cases or tumor characteristics where performance was notably lower or higher?; How does the 'semi-automatic' labeling feature practically integrate into a clinical workflow, what is the typical time saved compared to fully manual segmentation, and what level of user expertise is assumed for effective manual correction?; Were longitudinal data from the same patients included in the testing, and how does NS-HGlio perform in tracking changes in glioma volume and subcomponents over time compared to serial evaluations by neuroradiologists?