FDA 510(k) Search - By Innolitics (SaMD and AI Experts)

LVIS NeuroMatch Software is intended for the review, monitoring and analysis of electroencephalogram (EEG) recordings made by EEG devices using scalp electrodes and to aid neurologists in the assessment of EEG. The device is intended to be used by qualified medical practitioners who will exercise professional judgement in using the information.
The Seizure Detection component of LVIS NeuroMatch is intended to mark previously acquired sections of adult EEG recordings from patients greater than or equal to 18 years old that may correspond to electrographic seizures, in order to assist qualified medical practitioners in the assessment of EEG traces. EEG recordings should be obtained with a full scalp montage according to the electrodes from the International Standard 10-20 placement.
The Spike Detection component of LVIS NeuroMatch is intended to mark previously acquired sections of adult EEG recordings from patients ≥18 years old that may correspond to spikes, in order to assist qualified medical practitioners in the assessment of EEG traces. LVIS NeuroMatch Spike Detection performance has not been assessed for intracranial recordings.
LVIS NeuroMatch includes the calculation and display of a set of quantitative measures intended to monitor and analyze EEG waveforms. These include Artifact Strength, Asymmetry Spectrogram, Autocorrelation Spectrogram, and Fast Fourier Transform (FFT) Spectrogram. These quantitative EEG measures should always be interpreted in conjunction with review of the original EEG waveforms.
LVIS NeuroMatch displays physiological signals such as electrocardiogram (ECG/EKG) if it is provided in the EEG recording.
The aEEG functionality included in LVIS NeuroMatch is intended to monitor the state of the brain.
LVIS NeuroMatch Artifact Reduction (AR) is intended to reduce muscle and eye movements, in EEG signals from the International Standard 10-20 placement. AR does not remove the entire artifact signal and is not effective for other types of artifacts. AR may modify portions of waveforms representing cerebral activity. Waveforms must still be read by a qualified medical practitioner trained in recognizing artifacts, and any interpretation or diagnosis must be made with reference to the original waveforms.
LVIS NeuroMatch EEG source localization visualizes brain electrical activity on a 3D idealized head model. LVIS NeuroMatch source localization additionally calculates and displays summary trends based on source localization findings over time.
This device does not provide any diagnostic conclusion about the patient's condition to the user.

Device Description

NeuroMatch is a cloud-based software as a medical device (SaMD) intended to review, monitor, display, and analyze previously acquired and/or near real-time electroencephalogram (EEG) data from patients greater than or equal to 18 years old. The device is not intended to substitute for real-time monitoring of EEG. The software includes advanced algorithms that perform artifact reduction, seizure detection, and spike detection.

The subject device is identical to the NeuroMatch device cleared under K241390, with exception of the following additional features:

Source localization;
Source localization trends;

Source localization and source localization trends are substantially equivalent to the Epilog PreOp (K172858). Apart from the proposed additional software changes and associated changes to the Indications for Use and labeling there are no changes to the intended use or to the software features that were previously cleared. Below is a description of the software functions that will be added to the cleared NeuroMatch Device.

1. Source Localization

The NeuroMatch Source Localization visualization feature is used to visualize recorded EEG activity from the scalp in an idealized 3D model of the brain. The idealized brain model is based on template MR images. Each single sample of EEG-measured brain activity corresponds to a single point/pixel referred to as a source localization node (i.e., "node"). Together, the source localization nodes form a 3D cartesian grid where EEG signals with higher standardized current density are depicted in red and signals with lower standardized current density are depicted in blue. Source localization can be performed for any selected segment of the EEG data. The maximum and minimum of the source localization values are the absolute maximum and minimum values across the selected EEG signal, respectively. Users can also set an absolute threshold for the minimum value of the source localization outputs.

2. Source Localization Trends

NeuroMatch provides three automatic source localization trends to assist physicians investigating the amplitude and the frequency of the signal of interest (e.g. seizure onset) at the source space. Two of the trends provide simple 3D views of the sources of the high amplitude / high frequency across the signal of interest. The third trend provides a similar 3D view of the high frequency source movement across time.

Maximum Amplitude Projection (MAP): This metric allows clinicians to readily determine which brain regions are active and have high amplitude source localization results. The metric is determined by iterating through each node within a specified analysis time window and outputting the maximum source localization amplitude at that node within the specified analysis time window. No value is reported for nodes which have not been identified as maximum at any time during the specified window. This metric can help show brain regions that have high amplitude during a seizure.
Node Visit Frequency (NVF): This metric is reported as the number of times that a node has been labeled as maximum over time. This metric can help clinicians identify which brain regions are frequently active during a seizure.
Node Transition Frequency (NTF): This metric allows clinicians to determine which brain regions are active in consecutive time frames over a selected time period. A node transition is defined as a transition from one maximum point to another over time, and the node transition frequency is calculated by iterating through all time points for a specified analysis window, counting the number of times a transition between two points occurs over that time, and then dividing it by the time window of analysis. This metric can help identify pairs of brain regions that are frequently active in sequential order.

AI/ML Overview

Here's an analysis of the acceptance criteria and study details for the NeuroMatch device, based on the provided FDA 510(k) clearance letter:

1. Table of Acceptance Criteria and Reported Device Performance

The FDA letter does not explicitly state "acceptance criteria" in the traditional sense of pre-defined thresholds for performance metrics. Instead, the study's primary objective for Source Localization was to demonstrate non-inferiority to a reference device (CURRY) and comparable performance to a predicate device (Epilog PreOp). Therefore, the "acceptance criteria" can be inferred from the study's conclusions regarding non-inferiority and comparability.

For Source Localization Trends, the acceptance criterion was functional correctness and clinician understanding.

Feature / Metric	Acceptance Criteria (Inferred)	Reported Device Performance
Source Localization
Non-Inferiority to CURRY (Reference Device)	Lower bound of one-sided 95% CI of success rate difference (NeuroMatch - CURRY) > pre-specified non-inferiority margin.	NeuroMatch success rate: 90.7% (39/43 concordant patients)
CURRY success rate: 86% (37/43 concordant patients)
Lower bound of one-sided 95% CI of success rate difference: -4.65% (greater than pre-specified non-inferiority margin).
This established non-inferiority.
Comparability to Epilog PreOp (Predicate Device)	Comparable success rate and 95% CI overlap.	NeuroMatch success rate: 91.7% (95% CI: 79.16, 100)
Epilog PreOp success rate: 91.7% (95% CI: 79.16, 100)
This indicates comparable performance.
Consistency across Gender (Source Localization)	No considerable gender-related differences, consistently non-inferior to CURRY.	Male: CURRY 81.3%, NeuroMatch 87.5%
Female: CURRY 88.9%, NeuroMatch 92.6%
Observation suggests no considerable gender-related differences.
Consistency across Age Groups (Source Localization)	Comparable performance to CURRY consistently across age groups.	Age [18, 30): CURRY 81.8%, NeuroMatch 81.8%
Age [30, 40): CURRY 91.7%, NeuroMatch 91.7%
Age [40, 50): CURRY 85.7%, NeuroMatch 92.9%
Age [50, 75): CURRY 83.3%, NeuroMatch 100.0%
Results suggest comparable performance across age groups.
Source Localization Trends	Functional correctness (passes all test cases).
Clinician understanding and perceived clinical utility.	All test cases passed, confirming trends functioned as intended and yielded expected results.
Clinical survey of 15 clinicians showed they were able to understand the function of each trend and provided information regarding clinical utility in their workflow.

2. Sample Size Used for the Test Set and Data Provenance

Sample Size (Source Localization Test Set): 43 patients.
Data Provenance: Collected from three independent and geographically diverse medical institutions:
- Two institutions in the United States.
- One institution in South Korea.
- The study utilized retrospective data, as it focused on "previously acquired sections" of EEG recordings and "normalized post-operative MRIs with distinctive resection regions," indicating these were historical cases with established outcomes.

3. Number of Experts Used to Establish the Ground Truth for the Test Set and Qualifications

Number of Experts: Three (3) US board-certified epileptologists.
Qualifications: "US board-certified epileptologists." (Specific years of experience are not mentioned, but board certification implies a high level of expertise in the field).

4. Adjudication Method for the Test Set

Adjudication Method: The three board-certified epileptologists independently completed a survey. They were presented with source localization results from each device (NeuroMatch, CURRY, PreOp) and normalized post-operative MRIs with resection regions.
Ground Truth Establishment: Each physician independently determined the resection region at the sublobar level and then assessed whether the SL output of each device had any overlap with this determined resection region. For each patient and device, they responded to a "Yes/No" question asking about concordance. The method doesn't explicitly mention a consensus or adjudication process between the three experts for the final ground truth, but rather their individual assessments were used to determine the concordance rate. However, implying the "resected brain areas" as the primary ground truth, their task was to evaluate if the SL output agreed with this established ground truth from the MRIs. The "concordance" was then aggregated across their individual assessments against the known resection region.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study

Yes, a form of MRMC study was done, but not in the traditional sense of measuring human reader improvement with AI assistance.
- The study involved multiple readers (3 epileptologists) assessing multiple cases (43 patients).
- However, the comparison was between AI algorithms (NeuroMatch vs. CURRY vs. PreOp), with the human readers acting as independent evaluators to establish concordance with a post-operative ground truth (resected brain areas).
- Effect Size of Human Reader Improvement with AI vs. Without AI Assistance: This specific metric was not assessed or reported. The study evaluated the standalone AI performance of NeuroMatch compared to other AI devices, using human experts to determine the "correctness" of the AI's output in relation to surgical outcomes. It did not measure how human readers' diagnostic accuracy or efficiency changed when using NeuroMatch as an aid.

6. Standalone (Algorithm Only Without Human-in-the-Loop Performance) Study

Yes, a standalone study was done for the Source Localization feature. The study directly compared the performance of the NeuroMatch algorithm against the CURRY reference device and the PreOp predicate device. The output was a "Yes/No" concordance with the resected brain area, as assessed by the experts. The experts evaluated the device's output, not their own performance using the device.

7. Type of Ground Truth Used

Source Localization: The ground truth used was the resected brain areas as identified on normalized post-operative MRIs. This is a form of outcomes data combined with anatomical pathology (surgical intervention). The epileptologists were tasked with identifying whether the source localization output (from the algorithms) "overlapped" with these resected regions.
Source Localization Trends: For the trends (MAP, NVF, NTF), the ground truth for functional correctness was EEG datasets with known solutions (i.e., simulated or carefully crafted data where the expected output of the algorithms was precisely predictable). For clinical utility, the ground truth was clinical feedback and perceived understanding from the 15 clinicians.

8. Sample Size for the Training Set

The document does not specify the sample size for the training set for any of the algorithms. It only details the test set used for validation.

9. How the Ground Truth for the Training Set Was Established

The document does not specify how the ground truth for the training set was established. Since the training set size is not provided, this information is also absent.

Summary

Regulation Number and Section

§ 882.1400 Electroencephalograph.

(a)
Identification. An electroencephalograph is a device used to measure and record the electrical activity of the patient's brain obtained by placing two or more electrodes on the head.(b)
Classification. Class II (performance standards).