Search Results

Cirrus Resting State fMRI Software (Cirrus) is a software solution that performs magnetic resonance image processing including the processing of blood oxygen level dependent (BOLD) functional magnetic resonance imaging (fMRI) data, resting state fMRI analysis, and output generation. Cirrus generates task-analogous motor, language, and vision resting state fMRI correlation maps, in DICOM® format for visualization and analysis external to Cirrus.

Cirrus maps have been found to vary in normal subjects tested by repeat MR acquisition under conditions where no functional mapping change was expected. Medical imaging processing systems intended for BOLD fMRI post-processing are adjunctive and not intended to replace direct functional mapping procedures.

Typical users of Cirrus are medical professionals, including but not limited to surgeons, radiologists, and other clinicians.

Cirrus Resting State fMRI Software (Cirrus) is software as a medical device (SaMD) that performs image processing, resting state functional magnetic resonance imaging analysis, and output generation. Cirrus is launched by a host computing system external of Cirrus.

Cirrus processes an individual patient's brain magnetic resonance imaging (MRI) dataset which includes blood oxygenation level dependent (BOLD) functional MRI (fMRI) data and structural MRI data. Software components making up Cirrus include (i) a suite of fMRI preprocessing tools; (ii) a voxel-wise resting state fMRI correlation map generator, (iii) a nonadaptive machine-learning based resting state network (RSN) membership scoring algorithm; and (iv) an RSN map output generator.

The output of Cirrus is a set of patient-specific, task-analogous motor, language, and vision resting state fMRI correlation maps. The output maps correspond to three canonical, predefined brain resting state networks:

• Sensorimotor network (SMN),
• Language network (LAN), and
• Vision network (VIS).

Output resting state network maps are provided in DICOM® format for visualization and analysis external of Cirrus and accompanied by a quality report.

Here is a summary of the acceptance criteria and the study that proves the device meets the acceptance criteria for the Cirrus Resting State fMRI Software, based on the provided FDA 510(k) clearance letter:

Cirrus Resting State fMRI Software (K251009)

This section outlines the acceptance criteria for the Cirrus Resting State fMRI Software and the studies conducted to demonstrate that the device meets these criteria. The device aims to generate task-analogous motor, language, and vision resting state fMRI correlation maps that are consistent with established methods and clinically relevant.

1. Table of Acceptance Criteria and Reported Device Performance

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

The primary validation studies are based on retrospective data.

• SMN and LAN Map Validation: Patient demographics for the specific analysis comparing Cirrus SMN and LAN maps to published literature are referred to the individual cited publications (Dierker et al. 2017, Mitchell et al. 2013, Park et al. 2020, and Roland et al. 2019). The data was acquired at a "US clinical research institute" and specifically mentioned for:
  • Dierker et al. (2017): Not explicitly stated in the 510(k) summary, but typically US-based research.
  • Mitchell et al. (2013): Not explicitly stated in the 510(k) summary, but typically US-based research.
  • Park et al. (2020): Not explicitly stated in the 510(k) summary, but typically US-based research.
  • Roland et al. (2019): Not explicitly stated in the 510(k) summary, but typically US-based research.
  • Overall, the data for these studies comprised "brain tumor and epilepsy adult, adolescent, and child patients (ages 3 to 71 years)." All data were acquired on 3T Siemens scanners.
• VIS Map Validation:
  • Sample Size: 26 subjects.
  • Data Provenance: Retrospective, comprised of brain tumor and epilepsy adult and pediatric patients. All data were acquired on 3T Siemens scanners.
• Validation for GE MRI Input (Cross-Scanner Type Consistency):
  • Sample Size: 8 healthy, normal subjects.
  • Data Provenance: Prospective (implied by "each imaged at 8 geographically diverse sites"), acquired at 8 geographically diverse sites within the US. Data was collected on both Siemens and GE 3T MR devices.

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

The provided document does not explicitly state the number of experts or their specific qualifications (e.g., years of experience as radiologist) used to establish the ground truth for the test set during the validation of Cirrus.

However, the "expert clinician's perspective" is mentioned in the context of demonstrating comparability, but no details are given for the direct evaluation of the Cirrus maps by experts for the acceptance criteria. The ground truth for SMN and LAN maps was based on "published literature," implying an established consensus from the scientific community and potentially expert consensus reflected in those studies. For the VIS map, it was compared to "same-patient task-activated vision maps," which typically serve as a clinical gold standard (functional truth) established and interpreted by expert radiologists/neurologists.

4. Adjudication Method for the Test Set

The document does not describe a formal adjudication method (e.g., 2+1, 3+1) for the test set results. The validation relied on quantitative metrics (spatial correlation, AUC) against established literature or task-based fMRI, and implicit expert interpretation from the published studies.

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

No specific Multi-Reader Multi-Case (MRMC) comparative effectiveness study comparing human readers with and without AI assistance is detailed in the provided 510(k) summary. The studies focused on the performance of the algorithm itself in generating maps comparable to existing scientific understanding and clinical standards.

6. Standalone Performance Study

Yes, the provided performance data primarily describes the standalone performance of the Cirrus software. The validation involved comparing the output of the Cirrus algorithm (RSN maps) against:

• Published scientific literature (for SMN and LAN maps).
• Task-activated fMRI maps (for VIS maps).
• Cross-scanner consistency tests.
  The results (spatial correlation scores, AUC, and confidence levels) refer to the algorithm's direct output.

7. Type of Ground Truth Used

• For SMN and LAN Maps:
  • Expert Consensus / Published Knowledge: The ground truth was based on "corresponding resting state network maps evaluated in the published literature" and assessed to be "consistent with those used in Dierker et al (2017), Mitchell et al. (2013), Park et al. (2020), and Roland et al. (2019)." This implies relying on established scientific understanding and expert-interpreted functional maps within the scientific community.
• For VIS Maps:
  • Functional Ground Truth (Task-based fMRI): The ground truth was "a reference of same-patient task-activated vision maps." Task-based fMRI is considered a clinical gold standard for functional localization.
• For Cross-Scanner Consistency:
  • Internal Consistency: The ground truth was based on the expectation that functional networks should remain stable within the same subject across different scanner types, effectively using the subject's own functional connectivity as a "ground truth" for anatomical consistency across acquisitions.

8. Sample Size for the Training Set

• N = 48: 19 males and 29 females.
• Data Provenance: Acquired at a US clinical research institute.
• Inclusion Criteria: All subjects were adults screened to exclude neurological impairment and psychotropic medications.

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

The provided text only states that a "nonadaptive machine-learning based resting state network (RSN) membership scoring algorithm" is part of Cirrus. It does not explicitly detail how the ground truth for this specific training set (N=48) was established. Given the nature of resting-state fMRI analysis for RSNs, the ground truth for training would typically involve:

• Unsupervised Learning/Data-Driven Approaches: Identifying intrinsic connectivity patterns within the resting-state data itself, often guided by established neuroscience principles about functional networks, rather than externally labeled "ground truth" for each voxel.
• Pre-defined Atlases/Templates: Using established brain atlases or templates of canonical resting-state networks (e.g., component analysis results from large datasets like the Human Connectome Project) as a reference for network identification and membership scoring.
• The overall context of the submission suggests that the algorithm was trained to identify patterns consistent with the "three canonical, predefined brain resting state networks: Sensorimotor network (SMN), Language network (LAN), and Vision network (VIS)." The "nonadaptive" nature implies that the core RSN definition might be fixed or learned from this initial N=48 dataset in an unsupervised manner, rather than requiring new, specific expert labels for each training example.

Ask a specific question about this device