Search Results

The MAGNETOM system is indicated for use as a magnetic resonance diagnostic device (MRDD) that produces transverse, sagittal, coronal and oblique cross-sectional images, spectroscopic images and/or spectra, and that displays, depending on optional local coils that have been configured with the system, the internal structure and/or function of the head, body, or extremities. Other physical parameters derived from the images and/or spectra may also be produced. Depending on the region of interest, contrast agents may be used. These images and/or spectra and the physical parameters derived from the images and/or spectra when interpreted by a trained physician yield information that may assist in diagnosis.

The MAGNETOM system may also be used for imaging during interventional procedures when performed with MR compatible devices such as in-room displays and MR Safe biopsy needles.

MAGNETOM Flow.Ace and MAGNETOM Flow.Plus are 60cm-bore MRI systems with quench pipe-free, sealed magnets utilizing DryCool technology. They are equipped with BioMatrix technology and run on Siemens' syngo MR XA70A software platform. The systems include Eco Power Mode for reduced energy and helium consumption. They have different gradient configurations suitable for all body regions, with stronger configurations supporting advanced cardiac imaging. Compared to the predicate device, new hardware includes a new magnet, gradient coil, RF system, local coils, patient tables, and computer systems. New software features include AutoMate Cardiac, Quick Protocols, BLADE with SMS acceleration for non-diffusion imaging, Deep Resolve Swift Brain, Fast GRE Reference Scan, Ghost reduction, Fleet Reference Scan, SMS Averaging, Select&GO extension, myExam Spine Autopilot, and New Startup-Timer. Modified features include improvements for Pulse Sequence Type SPACE, improved Gradient ECO Mode Settings, and Inline Image Filter switchable for users.

The provided 510(k) clearance letter and summary describe the acceptance criteria and supporting studies for the MAGNETOM Flow.Ace and MAGNETOM Flow.Plus devices, particularly focusing on their AI features: Deep Resolve Boost, Deep Resolve Sharp, and Deep Resolve Swift Brain.

Here's a breakdown of the requested information:

1. Table of Acceptance Criteria and Reported Device Performance

The document uses quality metrics like PSNR, SSIM, and NMSE as indicators of performance and implicitly as acceptance criteria. Visual inspection and clinical evaluations are also mentioned.

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

The document uses "Training and Validation data" and often refers to the datasets used for both. It is not explicitly stated what percentage or how many cases from these datasets were strictly reserved for a separate "test set" and what came from the "validation sets." However, given the separation in slice count, the "Validation" slices for Deep Resolve Swift Brain might be considered the test set.

• Deep Resolve Boost:
  • TSE: >25,000 slices
  • HASTE: >10,000 HASTE slices (refined)
  • EPI Diffusion: >1,000,000 slices
  • Data Provenance: Retrospectively created from acquired datasets. Data covered a broad range of body parts, contrasts, fat suppression techniques, orientations, and field strength.
• Deep Resolve Sharp:
  • Data: >10,000 high resolution 2D images
  • Data Provenance: Retrospectively created from acquired datasets. Data covered a broad range of body parts, contrasts, fat suppression techniques, orientations, and field strength.
• Deep Resolve Swift Brain:
  • 1.5T Validation: 3,616 slices (This functions as a test set for 1.5T)
  • 3T Validation: 6,048 slices (This functions as a test set for 3T)
  • Data Provenance: Retrospectively created from acquired datasets.

The document does not explicitly state the country of origin for the data.

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

The document does not explicitly state the number or qualifications of experts used to establish the ground truth for the test sets. For Deep Resolve Swift Brain, it mentions "evaluated in clinical settings with collaboration partners," implying clinical experts were involved in the evaluation, but details are not provided. For Boost and Sharp, the "acquired datasets...represent the ground truth," suggesting the raw imaging data itself, rather than expert annotations on that data, served as ground truth.

4. Adjudication Method for the Test Set

The document does not describe a formal adjudication method (e.g., 2+1, 3+1). For Deep Resolve Swift Brain, it mentions "visually inspected" and "evaluated in clinical settings with collaboration partners," suggesting a qualitative assessment, but details on consensus or adjudication are missing.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was Done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance

A formal MRMC comparative effectiveness study demonstrating human reader improvement with AI vs. without AI assistance is not described in the provided text. The studies focus on the AI's standalone performance in terms of image quality metrics and internal validation.

6. If a Standalone (i.e. algorithm only without human-in-the-loop performance) was Done

Yes, standalone performance was done for the AI features. The "Test Statistics and Test Results Summary" for Deep Resolve Boost, Deep Resolve Sharp, and Deep Resolve Swift Brain describe the evaluation of the algorithm's output using quantitative metrics (PSNR, SSIM, NMSE) and visual inspection against reference standards, which is characteristic of standalone performance evaluation.

7. The Type of Ground Truth Used

For Deep Resolve Boost, Deep Resolve Sharp, and Deep Resolve Swift Brain, the ground truth used was the acquired high-quality datasets themselves. The input data for training and validation was then retrospectively created from this ground truth by manipulating or augmenting it (e.g., undersampling k-space, adding noise, cropping, using only the center part of k-space). This means the original, higher-quality MR images or k-space data served as the reference for what the AI models should reconstruct or improve upon.

8. The Sample Size for the Training Set

• Deep Resolve Boost:
  • TSE: >25,000 slices
  • HASTE: pre-trained on the TSE dataset and refined with >10,000 HASTE slices
  • EPI Diffusion: >1,000,000 slices
• Deep Resolve Sharp: >10,000 high resolution 2D images
• Deep Resolve Swift Brain: 20,076 slices

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

For Deep Resolve Boost, Deep Resolve Sharp, and Deep Resolve Swift Brain, the "acquired datasets (as described above) represent the ground truth for the training and validation." This implies that high-quality, fully acquired MRI data was considered the ground truth. The input data used during training (e.g., undersampled, noisy, or lower-resolution versions) was then derived or manipulated from this original ground truth. Essentially, the "ground truth" was the optimal, full-data acquisition before any degradation was simulated for the AI's input.

Ask a specific question about this device

MAGNETOM Free.Max system is indicated for use as a magnetic resonance diagnostic device (MRDD) that produces transverse, sagittal, coronal, and oblique cross-sectional images that display the internal structure and/or function of the head. body, or extremities. Other physical parameters derived from the images may also be produced. Depending on the region of interest, contrast agents may be used. These images and the physical parameters derived from the images when interpreted by a trained physician vield information that may assist in diagnosis.

MAGNETOM Free.Max may also be used for imaging during interventional procedures when performed with MR-compatible devices such as MR Safe biopsy needles.

MAGNETOM Free.Star system is indicated for use as a magnetic resonance diagnostic device (MRDD) that produces transverse, sagittal, coronal, and oblique cross-sectional images that display the internal structure and/or function of the head, body, or extremities. Other physical parameters derived from the images may also be produced. Depending on the region of interest, contrast agents may be used. These images and the physical parameters derived from the images when interpreted by a trained physician yield information that may assist in diagnosis.

MAGNETOM Free.Max and MAGNETOM Free.Star with syngo MR XA60A include new and modified features compared to the predicate devices MAGNETOM Free.Max and MAGNETOM Free.Star with syngo MR XA50A (K220575, cleared on June 24, 2022).

Below is a high-level summary of the new and modified hardware and software features compared to the predicate devices MAGNETOM Free.Max and MAGNETOM Free.Star with syngo MR XA50A:

Hardware
New hardware features:

• Contour Knee coil
• Respiratory Sensor

Modified hardware features:

• myExam 3D Camera
• Host computer
• MaRS

Software
New Features and Applications:

• Injector coupling
• Respiratory Sensor Support
• myExam RT Assist (only for MAGNETOM Free.Max)
• myExam Autopilot Hip
• Deep Resolve Boost
• Complex Averaging
• HASTE_Interactive (only for MAGNETOM Free.Max)
• BEAT_Interactive (only for MAGNETOM Free.Max)
• Needle Intervention AddIn (only for MAGNETOM Free.Max)

Modified Features and Applications:

• Deep Resolve Sharp
• Deep Resolve Gain
• SMS Averaging

Other Modifications:

• Indications for Use (only for MAGNETOM Free.Max)
• MAGNETOM Free.Max RT Edition marketing bundle (only for MAGNETOM Free.Max)

The provided text describes information about the submission of the MAGNETOM Free.Max and MAGNETOM Free.Star MRI systems for FDA 510(k) clearance, and references a specific AI feature called "Deep Resolve Boost." However, it does not contain acceptance criteria or a detailed study proving the device meets specific performance criteria for the AI feature.

The section titled "Test statistics and test results" for Deep Resolve Boost (Table 1, page 7) mentions that the impact of the network was characterized by quality metrics such as PSNR and SSIM, and visual inspection. It also states: "After successful passing of the quality metrics tests, work-in-progress packages of the network were delivered and evaluated in clinical settings with cooperation partners." This suggests internal testing and evaluation, but does not provide the specific numerical acceptance criteria or the detailed results of these tests.

Therefore, I cannot fully complete the requested table and answer all questions due to the lack of this specific information in the provided document.

However, I can extract the available information regarding the AI feature "Deep Resolve Boost" as much as possible:

1. Table of acceptance criteria and the reported device performance:

2. Sample size used for the test set and the data provenance:

• Test Set (Validation set for AI feature Deep Resolve Boost):
  • Sample Size: 1,874 2D slices.
  • Data Provenance: "in-house measurements and collaboration partners." The document does not specify the country of origin.
  • Retrospective or Prospective: Retrospectively created from ground truth by data manipulation and augmentation.

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts:

• Number of experts: Not specified.
• Qualifications of experts: The document states the "acquired datasets represent the ground truth for the training and validation," but it does not specify how this ground truth was established in terms of expert involvement for the test set. It mentions "clinical settings with cooperation partners" for evaluation, but this is distinct from ground truth establishment.

4. Adjudication method (e.g., 2+1, 3+1, none) for the test set:

• Not specified. The document states "acquired datasets represent the ground truth," suggesting pre-existing data or a different method of ground truth establishment than explicit reader adjudication for this AI feature.

5. If a multi-reader multi-case (MRMC) comparative effectiveness study was done, If so, what was the effect size of how much human readers improve with AI vs without AI assistance:

• The document states "No clinical tests were conducted to support substantial equivalence for the subject device" (page 10). It mentions that "work-in-progress packages of the network were delivered and evaluated in clinical settings with cooperation partners," but this is not described as an MRMC comparative effectiveness study, nor are any results on human reader improvement reported.

6. If a standalone (i.e. algorithm only without human-in-the loop performance) was done:

• The performance of the "Deep Resolve Boost" AI feature was characterized by "quality metrics such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM)" and visual inspection, which suggests a standalone evaluation of the algorithm's output against a reference standard. Specific results are not provided.

7. The type of ground truth used (expert consensus, pathology, outcomes data, etc.):

• For Deep Resolve Boost: "The acquired datasets represent the ground truth for the training and validation." Input data for training was "retrospectively created from the ground truth by data manipulation and augmentation." This implies that high-quality, likely clinical-grade, MRI scans acquired without the AI feature were considered the "ground truth" to which the AI-processed images were compared. It's not explicitly stated if this "ground truth" itself was established by expert consensus, but it infers it from high-quality clinical acquisition.

8. The sample size for the training set:

• For Deep Resolve Boost: 24,599 2D slices.

9. How the ground truth for the training set was established:

• "The acquired datasets represent the ground truth for the training and validation. Input data was retrospectively created from the ground truth by data manipulation and augmentation. This process includes further under-sampling of the data by discarding k-space lines, lowering of the SNR level by addition of noise and mirroring of k-space data."
• This indicates that "ground truth" was established by using full, high-quality MR images. The "input data" for the AI model (which the AI then "boosts") was intentionally degraded (under-sampled, noised) from this high-quality ground truth. The AI's task is to reconstruct the degraded input data back to resemble the original high-quality "ground truth."

Ask a specific question about this device