(122 days)
NeuroQuant is intended for automatic labeling, visualization and volumetric quantification of segmentable brain structures and lesions from a set of MR images. Volumetric measurements may be compared to reference percentile data.
NeuroQuant is a fully automated MR imaging post-processing software medical device that provides automatic labeling, visualization, and volumetric quantification of brain structures and lesions from a set of MR images and returns segmented images and morphometric reports.
NeuroQuant provides morphometric measurements of brain structures based on a 3D T1 MRI series. The optional use of the T2 FLAIR MR series and T2* GRE/SWI series allows for additional quantification of T2 FLAIR hyperintense lesions and T2* GRE/SWI hypointense lesions.
The device is used by medical professionals in imaging centers, hospitals, and other healthcare facilities as well as by clinical researchers. When used clinically, the output must be reviewed by a radiologist or neuroradiologist. The results are typically forwarded to the referring physician, most commonly a neurologist. The device is a "Prescription Device" and is not intended to be used by patients or other untrained individuals.
From a workflow perspective, the device is packaged as a computing appliance that is capable of supporting DICOM standard input and output. NeuroQuant supports data from all major MRI manufacturers and a variety of field strengths. For best results, scans should be acquired using specified protocols provided by CorTechs Labs.
As part of processing, the data is corrected by NeuroQuant for image acquisition artifacts, including gradient nonlinearities and bias field inhomogeneity, to improve overall image quality.
Next, image baseline intensity levels for gray and white matter are identified and corrected for scanner variability. The scan is then aligned with the internal anatomical atlas by a series of transformations. Probabilistic methods and neural network models are then used to label each voxel with an anatomical structure based on location and signal intensities.
Output of the software provides values as numerical volumes, and images of derived data as grayscale intensity maps and as color overlays on top of the anatomical image. The outputs are provided in standard DICOM format as image series and reports that can be displayed on many commercial DICOM workstations.
The software is designed without the need for a user interface after installation. Any processing errors are reported either in the output series error report or system log files.
The software can provide data on age and gender-matched normative percentiles. The default reference percentile data for NeuroQuant comprises normal population data.
The device provides DICOM Storage capabilities to receive MRI series in DICOM format from an external source, such as an MRI scanner or PACS server. The device provides transient data storage only. If additional scans from other time points are available, the software can perform change analysis.
Here's a breakdown of the acceptance criteria and the study details for the NeuroQuant device, based on the provided FDA 510(k) summary:
1. Table of Acceptance Criteria and Reported Device Performance
| Model | Acceptance Criteria | Reported Device Performance | Metric |
|---|---|---|---|
| Brain Segmentation Model | Performance against predicate device (meets accuracy and reproducibility criteria) | Meets acceptance criteria for accuracy and reproducibility (details not explicitly stated beyond "meets acceptance criteria") | Dice Similarity Coefficient (DSC) |
| FLAIR Lesion Segmentation Model | Mean DSC ≥ 0.50 and standard deviation ≤ 0.18 | Mean DSC of 0.70 with a standard deviation of 0.14 | Dice Similarity Coefficient (DSC) |
| MCH Detection Model | Median F1 Score ≥ 0.51 | Median F1 Score of 0.60 | F1 Score |
2. Sample Sizes Used for the Test Set and Data Provenance
- Brain Segmentation Model:
- Test Set Size: 30 patients
- Data Provenance: Curated to represent diverse patient population across the United States. Type of study (retrospective/prospective) and specific countries of origin within the US are not specified, but it implies retrospective data collection from diverse institutions within the US.
- FLAIR Lesion Segmentation Model:
- Test Set Size: 63 patients
- Data Provenance: Curated to represent diverse patient population across the United States. Type of study (retrospective/prospective) not specified, but implies retrospective data collection from diverse institutions within the US (data acquired across Philips, GE, and Siemens scanners).
- MCH Detection Model:
- Test Set Size: 117 patients
- Data Provenance: Curated to represent diverse patient population across the United States. Type of study (retrospective/prospective) not specified, but implies retrospective data collection from diverse institutions within the US (data acquired across Philips, GE, and Siemens scanners).
3. Number of Experts Used to Establish Ground Truth for the Test Set and Qualifications of Those Experts
The document does not specify the number of experts used or their detailed qualifications (e.g., radiologist with 10 years experience) for establishing the ground truth of the test sets. It broadly states that the software was validated against "known ground truth values" and "gold standard - computer-aided expert manual segmentation," but provides no specifics on the human experts involved in generating this ground truth for the test sets.
4. Adjudication Method for the Test Set
The document does not specify any adjudication method (e.g., 2+1, 3+1) for the ground truth of the test sets. It only refers to a "gold standard - computer-aided expert manual segmentation."
5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study Was Done, and Effect Size
The document does not mention a multi-reader multi-case (MRMC) comparative effectiveness study, nor does it quantify how much human readers improve with AI vs. without AI assistance. The study focuses on the standalone performance of the algorithms.
6. If a Standalone (Algorithm Only Without Human-in-the-Loop Performance) Was Done
Yes, standalone performance was done. The performance metrics (Dice Similarity Coefficient, F1 Score) are measurements of the algorithm's output compared to a reference ground truth, indicating a standalone analysis. The document states that the results "must be reviewed by a trained physician," implying the device is a tool to assist, but the evaluation of the device itself focuses on its automated output.
7. The Type of Ground Truth Used
The ground truth for the test sets was established using "known ground truth values" and the "gold standard - computer-aided expert manual segmentation." This implies that human experts, potentially assisted by software tools, manually segmented or labeled the structures to create the reference standard for evaluation.
8. The Sample Size for the Training Set
- Brain Segmentation Model: Trained on 1,473 3D T1-weighted MRI series.
- FLAIR Lesion Segmentation Model: Developed using a training set of 340 T1 and FLAIR MRI series.
- MCH Detection Model: Trained on 463 2D T2*GRE/SWI MRI series.
9. How the Ground Truth for the Training Set Was Established
The document does not explicitly detail how the ground truth for the training sets was established. It describes the data sources (diverse MRI series from various institutions) and mentions the use of "probabilistic methods and neural network models" for labeling in the device's processing, which implies that these models learn from some form of labeled or pre-segmented data. Given the "computer-aided expert manual segmentation" mentioned for ground truth in performance testing, it's highly probable that similar methods were used for generating labels for the training data, but this is not explicitly stated.
§ 892.2050 Medical image management and processing system.
(a)
Identification. A medical image management and processing system is a device that provides one or more capabilities relating to the review and digital processing of medical images for the purposes of interpretation by a trained practitioner of disease detection, diagnosis, or patient management. The software components may provide advanced or complex image processing functions for image manipulation, enhancement, or quantification that are intended for use in the interpretation and analysis of medical images. Advanced image manipulation functions may include image segmentation, multimodality image registration, or 3D visualization. Complex quantitative functions may include semi-automated measurements or time-series measurements.(b)
Classification. Class II (special controls; voluntary standards—Digital Imaging and Communications in Medicine (DICOM) Std., Joint Photographic Experts Group (JPEG) Std., Society of Motion Picture and Television Engineers (SMPTE) Test Pattern).