(202 days)
Vantage Galan 3T systems are indicated for use as a diagnostic imaging modality that produces crosssectional transaxial, coronal, sagittal, and oblique images that display anatomic structures of the head or body. Additionally, this system is capable of non-contrast enhanced imaging, such as MRA.
MRI (magnetic resonance imaging) images correspond to the spatial distribution of protons (hydrogen nuclei) that exhibit nuclear magnetic resonance (NMR). The NMR properties of body tissues and fluids are:
·Proton density (PD) (also called hydrogen density) ·Spin-lattice relaxation time (T1) ·Spin-spin relaxation time (T2) ·Flow dynamics ·Chemical Shift
Depending on the region of interest, contrast agents may be used. When interpreted by a trained physician, these images yield information that can be useful in diagnosis.
The Vantage Galan (Model MRT-3020) is a 3 Tesla Magnetic Resonance Imaging (MRI) System, previously cleared under K222387. This system is based upon the technology and materials of previously marketed Canon Medical Systems and is intended to acquire and display crosssectional transaxial, coronal, sagittal, and oblique images of anatomic structures of the head or body.
The provided FDA 510(k) summary (K230355) describes the Canon Medical Systems Vantage Galan 3T, MRT-3020, V9.0 with AiCE Reconstruction Processing Unit for MR. This submission primarily concerns modifications to a previously cleared device (K222387), introducing several new software features.
The document discusses various tests for these new features, demonstrating their safety and effectiveness. Importantly, for features involving AI/ML components (like PIQE, NeuroLine+, IMC, and Free Breathing Dynamic DLR), the summary details some of the acceptance criteria and the studies performed to prove the device meets them.
Below is a structured summary of the acceptance criteria and study details based on the provided text, focusing on the AI/ML-related features where specific studies are described.
Acceptance Criteria and Device Performance Study for Vantage Galan 3T, MRT-3020, V9.0 with AiCE Reconstruction Processing Unit for MR
This document outlines the testing and validation of several new software features for the Vantage Galan 3T MRI system. Specifically, for features that incorporate Deep Learning or Machine Learning (PIQE, NeuroLine+, IMC, Free Breathing Dynamic DLR), detailed studies were conducted to ensure their performance meets pre-defined acceptance criteria.
1. Acceptance Criteria and Reported Device Performance
| Feature Tested | Acceptance Criteria | Reported Device Performance |
|---|---|---|
| PIQE (Precise IQ Engine) for MR | Bench Testing (Phantom): Maintain or improve SNR and signal intensity profiles for ringing and sharpness, demonstrate ability to generate higher in-plane matrix from lower matrix images, contribute to ringing artifact reduction and increase of sharpness. |
Bench Testing (Clinical - Brain & Knee): PIQE images should have sharper edges while mitigating smoothing and ringing effects, and maintain similar or better contrast and SNR compared to conventional methods.
Clinical Image Review: Images reconstructed with PIQE (new method) should be scored at or above "clinically acceptable" (3 or above on a modified 5-point Likert scale) in various clinically-relevant categories (ringing, sharpness, SNR, overall IQ, feature conspicuity) by board-certified radiologists, compared to the conventional method (matrix expansion with Fine Reconstruction and processing with GA Filter). Reviewers should exhibit strong agreement (at "good" and "very good" level) for all IQ metrics. PIQE should demonstrate: (a) ability to generate higher spatial in-plane resolution images (e.g., tripling matrix dimensions, i.e., 9x factor), (b) contribution to ringing artifact reduction, denoising, and increased sharpness, (c) ability to accelerate scanning by reducing acquisition matrix while maintaining clinical matrix size and image quality, (d) benefits on regular clinical protocols without acquisition parameter adjustment. | Bench Testing (Phantom): Confirmed PIQE generates higher in-plane matrix from lower matrix image, and contributes to ringing artifact reduction and increase of sharpness.
Bench Testing (Clinical - Brain & Knee): Confirmed PIQE generates images with sharper edges while mitigating smoothing and ringing effects and maintaining similar or better contrast and SNR compared to zeropadding interpolation and GA filter.
Clinical Image Review: All resulting reconstructions (conventional and new) were scored at, or above, clinically acceptable by three board-certified Radiologists per anatomy. The reviewers exhibited a strong agreement at the "good" and "very good" level for all IQ metrics. Confirmed (a) PIQE generates higher spatial in-plane resolution images (up to 9x factor), (b) PIQE contributes to ringing artifact reduction, denoising, and increased sharpness, (c) PIQE can accelerate scanning while maintaining image quality, and (d) PIQE benefits obtained on regular clinical protocols without requiring acquisition parameter adjustment. |
| NeuroLine+ | Bench Testing (Clinical): Successful scan alignment (offset and angle within acceptable error defined as typical interrater variability) greater than 80% of the time, and similar or better angular error compared to the conventional method. The results should support that NeuroLine+ is a clinically acceptable option for slice-alignment in head examination. | Bench Testing (Clinical): For angular error, NeuroLine+ met acceptance criteria, being similar or better than the conventional method. For auto-positioning, yielded 96.0% success, meeting the acceptance criteria (≥80%). The results support the conclusion that NeuroLine+ is a clinically acceptable option for slice-alignment in head examination. |
| Iterative Motion Correction (IMC) | Bench Testing (Clinical - simulated motion): Effective in reducing motion artifacts and meeting predetermined acceptance criteria based on peak SNR and structural similarity (SSIM).
Clinical Image Review (Volunteers): IMC corrected images should be the same as, or better than, images without IMC applied, as evaluated by three US board-certified radiologists specializing in neuro imaging across categories like SNR, tissue contrast, image sharpness, and diagnostic confidence (score of 3 or greater on a 5-point modified Likert scale considered clinically acceptable).
Clinical Image Review (Patients with Pathology): Diagnostic information in IMC images should be the same as or better than those without IMC applied. | Bench Testing (Clinical - simulated motion): Demonstrated IMC is effective in reducing motion artifacts and met predetermined acceptance criteria based on peak SNR and SSIM.
Clinical Image Review (Volunteers): Confirmed IMC performs as expected, significantly reducing motion artifacts, and improving overall image quality metrics (SNR, tissue contrast, image sharpness, diagnostic confidence). IMC corrected images are the same as, or better than, images without IMC applied.
Clinical Image Review (Patients with Pathology): Further confirmed diagnostic information in IMC images was the same or better than those without IMC applied. |
| Free Breathing Dynamic Deep Learning Reconstruction (DLR) | Bench Testing (Clinical - arterial phase detection): Automatic arterial phase detection should yield a success rate (automatically proposed phases include gold standard phase) greater than or equal to 80%.
Clinical Image Review: The average of visual scores for overall IQ, feature conspicuity, and diagnostic confidence should meet the acceptance criteria (score of 3 or higher on a modified 5-point Likert scale considered clinically acceptable). The results should support that Free Breathing Dynamic is a clinically acceptable option for free-breathing contrast-enhanced dynamic liver exams, providing acceptable diagnostic confidence. | Bench Testing (Clinical - arterial phase detection): Yielded 94.4% success, meeting the acceptance criteria (≥80%).
Clinical Image Review: The average of visual scores for overall IQ, feature conspicuity, and diagnostic confidence met the acceptance criteria. The results support the conclusion that Free Breathing Dynamic is a clinically acceptable option for the acquisition of free-breathing contrast-enhanced dynamic liver exams, providing acceptable diagnostic confidence. |
Note: For other features like Exsper 3D, Slice Shim, UTE CG Recon, DSD Filter, Ringing Correction, Auto Scan Assist, and Ceiling Camera, the summary mentions "Testing confirmed that..." the feature met its intended purpose or acceptance criteria, but specific quantitative criteria are not consistently detailed in table format. These typically fall under broader system performance tests.
2. Sample Sizes and Data Provenance
- PIQE (Precise IQ Engine):
- Test Set: 17 unique subjects (from USA, Europe, and Japan) providing a total of 292 scans (neuro and knee) in multiple orientations and contrast weightings.
- Data Provenance: USA, Europe, and Japan. Data were acquired separately from the training data.
- NeuroLine+:
- Test Set: 13 clinical cases (1 male, 12 female) from France.
- Data Provenance: France. Data was newly collected and entirely separate from the training group.
- IMC (Iterative Motion Correction):
- Test Set (Simulated Motion): 12 clinical datasets (without subject motion, with mathematically simulated motion added).
- Test Set (Clinical Image Review - Volunteers): 21 volunteers (imaged at 3T in either 16-channel or 32-channel coil). 300 image volumes (100 per group: motion-free, motion without IMC, motion with IMC).
- Test Set (Clinical Image Review - Patients with Pathology): 34 additional image volumes in brain or cervical spine from typical clinical patients with pathology and motion.
- Data Provenance: "All testing data were acquired separately and independently from the training data after the machine learning training was completed." (Specific countries not stated for IMC volunteer/patient data, but inferred to be where studies were conducted, likely similar to PIQE study locations i.e. US based readers, so US based data).
- Free Breathing Dynamic DLR:
- Test Set (Arterial Phase Detection): 18 clinical cases (9 male, 9 female) from Japan.
- Test Set (Clinical Image Review): 29 contrast-enhanced Free Breathing Dynamic liver studies (50 slices per study, repeated for 23 dynamic phases) from patients.
- Data Provenance: Japan (for phase detection), United States, France, and Japan (for clinical image review). "All testing data were acquired separately and independently from the training data after the machine learning training was completed."
3. Number of Experts and Qualifications for Ground Truth
- PIQE (Precise IQ Engine):
- Number of Experts: 6 USA board certified radiologists (3 per anatomy: brain/neuro, 3 for knee/musculoskeletal).
- Qualifications: "USA board certified radiologists." Specific years of experience not mentioned.
- NeuroLine+:
- Number of Experts: 2 experienced ARRT licensed MR technologists. (To manually annotate angle and position of target planes for ground truth).
- Qualifications: "experienced ARRT licensed MR technologists."
- IMC (Iterative Motion Correction):
- Number of Experts: 3 US board certified radiologists, specializing in neuro imaging.
- Qualifications: "US board certified radiologists, specializing in neuro imaging." Specific years of experience not mentioned.
- Free Breathing Dynamic DLR:
- Number of Experts: 2 US board certified radiologists. (For clinical image review). For ground truth of arterial phase detection, "experienced radiologists" manually selected the gold standard phase, specific number not stated or whether these were the same 2 as for image review.
- Qualifications: "US board certified radiologists" (for image review), "experienced radiologists" (for arterial phase ground truth).
4. Adjudication Method for the Test Set
- PIQE (Precise IQ Engine): Images were scored by 3 reviewers per anatomy. The results state: "The reviewers exhibited a strong agreement at the 'good' and 'very good' level for all IQ metrics." This implies a consensus-based approach or agreement analysis, but a formal adjudication rule (e.g., 2+1, 3+1) is not explicitly detailed. The acceptance criteria noted that scores of 3 or above by these reviewers were considered clinically acceptable.
- NeuroLine+: Ground truth for accuracy of auto-detected angle/center position was established by two experienced ARRT licensed MR technologists manually annotating the angle and position. It's not explicitly stated if or how adjudication occurred if their annotations differed. Given "acceptable error defined as typical interrater variability," it suggests a comparison against a known range of acceptable differences rather than formal adjudication of conflicting reads.
- IMC (Iterative Motion Correction): Three US board certified radiologists read and scored the images. Similar to PIQE, a formal adjudication method (e.g., 2+1) is not explicitly described, but the conclusion is based on their evaluation where a score of 3 or greater was considered clinically acceptable.
- Free Breathing Dynamic DLR: Two US board certified radiologists read and scored cases. For the arterial phase detection where ground truth was selected by "experienced radiologists," a formal adjudication method for potential disagreements is not described.
5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was done, and effect size of human readers improvement with AI vs without AI assistance
- MRMC-like studies were conducted for PIQE and IMC (and to a lesser extent for Free Breathing Dynamic DLR):
- PIQE: A randomized, blinded clinical image review study was conducted with 6 USA board certified radiologists (3 per anatomy). It compared images reconstructed with PIQE (new method) against the conventional method. The study aimed to show non-inferiority or improvement of AI-reconstructed images in terms of quality metrics (ringing, sharpness, SNR, overall IQ, feature conspicuity). The results indicate AI-processed images were "at, or above, clinically acceptable" and that PIQE "contributes to ringing artifact reduction, denoising and increased sharpness." This suggests an improvement in image quality. The document does not explicitly present an "effect size" of how much human readers "improved" with AI assistance in terms of their diagnostic accuracy or efficiency with the AI generated images vs. non-AI generated. The focus is on the quality of the AI-processed images themselves enabling acceptable diagnosis.
- IMC: A clinical image review was performed by three US board certified radiologists comparing images with and without IMC applied. The finding was that "IMC corrected images are the same as, or better than, images without IMC applied." A follow-up study with pathological cases confirmed "diagnostic information in IMC images was the same or better than those without IMC applied." Again, this focuses on the performance of the AI processing on the images themselves rather than a direct measure of human reader improvement in diagnostic tasks or efficiency. The study format (randomized and blinded image review) aligns with MRMC principles for evaluating image quality impacts.
6. If a Standalone (i.e., algorithm only without human-in-the-loop performance) was done
Yes, for several features, standalone performance was assessed through bench testing using phantoms or analyzed clinical images where the algorithm's output was quantified.
- PIQE: Bench testing used ACR phantom images to evaluate SNR, signal intensity profiles for ringing and sharpness, and clinical images for Edge Slope Width, Ringing Variable Mean, SNR, and Contrast Change Ratio. This evaluates the algorithm's direct impact on image characteristics.
- NeuroLine+: Performance was assessed by comparing the algorithm's auto-detected angle and center position against manually annotated ground truth by technologists, yielding a success rate for autopilot positioning. This is a standalone evaluation of the algorithm's
- IMC: Bench testing utilized clinical datasets with mathematically simulated motion to evaluate the algorithm's effectiveness in reducing motion artifacts using metrics like peak SNR and SSIM. This is a standalone evaluation of the algorithm's image processing capability.
- Free Breathing Dynamic DLR: Bench testing for automatic arterial phase detection assessed the algorithm's success rate in identifying the correct phases. This is a standalone performance metric for that specific AI function.
7. The type of Ground Truth used
- Expert Consensus/Opinion (Qualitative):
- For PIQE, IMC, and Free Breathing Dynamic DLR clinical image reviews, the ground truth for image quality and diagnostic acceptability was established by scores/preferences from multiple board-certified radiologists, based on a modified Likert scale. This is a form of expert consensus/opinion.
- Expert Annotation (Quantitative):
- For NeuroLine+, the ground truth for slice alignment accuracy (angle and position) was established by manual annotation by two experienced ARRT licensed MR technologists.
- For Free Breathing Dynamic DLR arterial phase detection, the gold standard phase was "manually selected by experienced radiologists."
- Known Conditions / Simulated Data:
- For IMC, some bench testing involved clinical datasets where motion was mathematically simulated, allowing for a known "truth" regarding motion artifacts.
- For PIQE, phantom images (ACR phantom) provided a controlled environment for objective image quality metrics.
8. The Sample Size for the Training Set
The document explicitly states for PIQE, IMC, and Free Breathing Dynamic DLR that "All testing data were acquired separately and independently from the training data after the machine learning training was completed." While it confirms the independence of training and testing data, it does not provide the specific sample size for the training set for any of the AI/ML features.
9. How the Ground Truth for the Training Set was Established
The document does not explicitly describe how the ground truth for the training set was established for any of the AI/ML features. It only confirms that the testing data were independent of the training data. For machine learning algorithms, ground truth for training data is typically established through a similar process of expert labeling, clinical data, or simulation, but the specifics are not detailed in this summary.
§ 892.1000 Magnetic resonance diagnostic device.
(a)
Identification. A magnetic resonance diagnostic device is intended for general diagnostic use to present images which reflect the spatial distribution and/or magnetic resonance spectra which reflect frequency and distribution of nuclei exhibiting nuclear magnetic resonance. Other physical parameters derived from the images and/or spectra may also be produced. The device includes hydrogen-1 (proton) imaging, sodium-23 imaging, hydrogen-1 spectroscopy, phosphorus-31 spectroscopy, and chemical shift imaging (preserving simultaneous frequency and spatial information).(b)
Classification. Class II (special controls). A magnetic resonance imaging disposable kit intended for use with a magnetic resonance diagnostic device only is exempt from the premarket notification procedures in subpart E of part 807 of this chapter subject to the limitations in § 892.9.