(27 days)
The Swoop Portable MR Imaging System is a portable, ultra-low field magnetic resonance imaging device for producing images that display the internal structure of the head where full diagnostic examination is not clinically practical. When interpreted by a trained physician, these images provide information that can be useful in determining a diagnosis.
The Swoop system is portable, ultra-low field MRI device that enables visualization of the internal structures of the head using standard magnetic resonance imaging contrasts. The main interface is a commercial off-the-shelf device that is used for operating the system, providing access to patient data, exam setup, exam execution, viewing MRI image data for quality control purposes, and cloud storage interactions. The system can generate MRI data sets with a broad range of contrasts. The Swoop system user interface includes touch screen menus, controls, indicators, and navigation icons that allow the operator to control the system and to view imagery. The Swoop System image reconstruction algorithm utilizes deep learning to provide improved image quality for T1W, T2W, FLAIR, and DWI sequences.
The subject Swoop System described in this submission includes software modifications related to the pulse sequences and image processing.
Here's a summary of the acceptance criteria and the studies that prove the device meets them, based on the provided FDA 510(k) clearance letter for the Swoop® Portable MR Imaging® System:
1. Table of Acceptance Criteria and Reported Device Performance
| Study Component | Acceptance Criteria | Reported Device Performance |
|---|---|---|
| Performance Analysis | NMSE (Normalized Mean Squared Error) should be reduced and SSIM (Structural Similarity Index) should be improved for Advanced Reconstruction test images compared to Linear Reconstruction test images. Advanced Reconstruction must preserve the presentation of motion and zipper artifacts and no unexpected output should be observed. | For all models and all test datasets, NMSE was reduced and SSIM was improved for Advanced Reconstruction test images compared to Linear Reconstruction test images. Advanced Reconstruction preserved the presentation of motion and zipper artifacts, and no unexpected output was observed. |
| Contrast-to-Noise Ratio (CNR) Validation | The mean CNR of Advanced Reconstruction was required to be greater than the mean CNR of the baseline Linear Reconstruction at a statistical significance level of 0.05 for each sequence type. The study result must demonstrate that Advanced Reconstruction does not unexpectedly modify, remove, or reduce the contrast of pathology features. | In all cases, CNR of Advanced Reconstruction was greater than or equal to Linear Reconstruction for both hyper- and hypo-intense pathologies. This demonstrates that Advanced Reconstruction does not unexpectedly modify, remove, or reduce the contrast of pathology features. |
| Advanced Reconstruction Image Validation | Advanced Reconstruction was required to perform at least as well as Linear Reconstruction in all categories (median score ≥0 on a Likert scale) and perform better (median score ≥1 on a Likert scale) in at least one of the quality-based categories (noise, sharpness, contrast, geometric fidelity, artifact, and overall image quality) when reviewed by external ABR-certified radiologists. | Advanced Reconstruction achieved a median score of 2 (the most positive rating scale value) in all categories. This indicates reviewers found Advanced Reconstruction improved image quality while maintaining diagnostic consistency relative to Linear Reconstruction. |
2. Sample Sizes and Data Provenance
The provided document does not explicitly state the country of origin for the data or whether it was retrospective or prospective for the training or test sets.
| Study Component | Sample Size (Test Set) | Data Provenance (Country, Retrospective/Prospective) |
|---|---|---|
| Performance Analysis | Total Subjects: 118 | |
| Total Unique Images: 378 | ||
| Per Model/Sequence Group: |
- T1, T2, FLAIR: 44 patients, 92 images
- DWI: 34 patients, 65 images | Not specified in the provided document. |
| Contrast-to-Noise Ratio (CNR) Validation | Patients: 43
Images: 95
ROIs (Regions of Interest): 316 | Not specified in the provided document. |
| Advanced Reconstruction Image Validation | Patients: 46
Images: 177
Per Sequence: At least 16 cases per sequence (with at least 4 cases per sequence-available image orientation) | Not specified in the provided document. |
3. Number of Experts and Qualifications for Ground Truth
| Study Component | Number of Experts | Qualifications of Experts |
|---|---|---|
| Performance Analysis | Not applicable for direct expert review; ground truth was generated synthetically or from high-field/synthetic contrast images. | N/A (reference-based metrics comparing reconstructed images to ground truth images). |
| Contrast-to-Noise Ratio (CNR) Validation | 2 | American Board of Radiology (ABR) certified radiologists. |
| Advanced Reconstruction Image Validation | 5 | External, American Board of Radiology (ABR) certified radiologists representing clinical users. |
4. Adjudication Method
| Study Component | Adjudication Method |
|---|---|
| Performance Analysis | Not applicable; objective metrics (NMSE, SSIM) compared reconstructed images to synthetic/derived ground truth. Qualitative assessment for motion and zipper artifacts. |
| Contrast-to-Noise Ratio (CNR) Validation | ROI annotations were reviewed by two ABR-certified radiologists, and inaccurate annotations were excluded. This implies a form of consensus or expert reconciliation for the ROIs. |
| Advanced Reconstruction Image Validation | Reviewers rated images using a five-point Likert scale. Individual ratings were used to derive a median score for each category. No explicit adjudication method (e.g., 2+1) for discrepant reader opinions is described beyond deriving a median. |
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
- A form of MRMC study was conducted for the "Advanced Reconstruction Image Validation" where five ABR-certified radiologists reviewed images.
- Effect Size with AI vs. without AI assistance: The study compared Advanced Reconstruction (which utilizes deep learning) to Linear Reconstruction (without advanced AI assistance). Advanced Reconstruction achieved a median score of 2 (the most positive rating scale value) in all categories (noise, sharpness, contrast, geometric fidelity, artifact, and overall image quality), indicating "improved image quality" relative to Linear Reconstruction. The Likert scale used was not detailed, but a score of 2 on a 5-point scale (where 0 might be "no difference" and higher values indicate improvement) suggests a significant positive effect.
6. Standalone Performance Study
- Yes, a standalone (algorithm only) performance study was conducted.
- The "Performance Analysis" section describes evaluating Advanced Reconstruction's ability to reproduce ground truth images using objective metrics (NMSE, SSIM) without human reader involvement for the primary comparison. The "Contrast-to-Noise Ratio Validation" also measured objective image characteristics (CNR) of the algorithm's output.
7. Type of Ground Truth Used
| Study Component | Type of Ground Truth |
|---|---|
| Performance Analysis | Reference-based metrics: A set of images including Swoop data, high field images, and synthetic contrast images was used as ground truth target images. Test input data (synthetic k-space generated from the target images) was reconstructed and compared to this ground truth. This is a form of derived/computed ground truth based on ideal or high-quality reference scans and synthetic generation. |
| Contrast-to-Noise Ratio (CNR) Validation | Pathologies in images were annotated and reviewed by two ABR-certified radiologists. The CNR was measured between these annotated pathologies and healthy white matter. This implicitly uses expert consensus/annotation for identifying and defining the regions of interest for ground truth comparison. However, the "ground truth" for the improvement in CNR is the Linear Reconstruction itself. |
| Advanced Reconstruction Image Validation | The reference standard for comparison was Linear Reconstruction. The "ground truth" here is human expert assessment (radiologists' ratings) of relative image quality and diagnostic consistency between the Advanced and Linear reconstructions, treating Linear Reconstruction as the baseline for comparison. |
8. Sample Size for the Training Set
- The document states that the test dataset was "entirely independent from the dataset used for model training."
- However, the specific sample size or characteristics of the training set are not provided in this document.
9. How the Ground Truth for the Training Set was Established
- The document states: "In all cases, models are trained and validated with MRI data and images as the only inputs and outputs" and "Advanced Reconstruction was performed using a test dataset entirely independent from the dataset used for model training."
- Similar to the training set sample size, the establishment of ground truth for the training set is not detailed in the provided document. It can be inferred that if high-field or synthetically derived images were used for validation, similar methods might have been used for training, but this is not explicitly stated.
§ 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.