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 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.

The subject devices, MAGNETOM Aera (including MAGNETOM Aera Mobile), MAGNETOM Skyra, MAGNETOM Prisma, MAGNETOM Prisma™, MAGNETOM Vida, MAGNETOM Lumina with software syngo MR XA60A, consist of new and modified software and hardware that is similar to what is currently offered on the predicate device, MAGNETOM Vida with syngo MR XA50A (K213693).

This FDA 510(k) summary describes several updates to existing Siemens Medical Solutions MRI systems (MAGNETOM Vida, Lumina, Aera, Skyra, Prisma, and Prisma fit), primarily focusing on software updates (syngo MR XA60A) and some modified/new hardware components. The document highlights the evaluation of new AI features, specifically "Deep Resolve Boost" and "Deep Resolve Sharp."

Here's an analysis of the acceptance criteria and the study details for the AI features:

1. Table of Acceptance Criteria and Reported Device Performance

The document provides a general overview of the evaluation metrics used but does not explicitly state acceptance criteria in a quantitative format (e.g., "Deep Resolve Boost must achieve a PSNR of X" or "Deep Resolve Sharp must achieve Y SSIM"). Instead, it describes the types of metrics used and qualitative assessments.

• Impact characterized by PSNR and SSIM. | The impact of the network has been characterized by several quality metrics such as peak signal-to-noise ratio (PSNR) and structural similarity index (SSIM). Most importantly, the performance was evaluated by visual comparisons to evaluate e.g., aliasing artifacts, image sharpness and denoising levels. |
  | Deep Resolve Sharp | - Preservation of image quality (image sharpness) compared to original.
• Impact characterized by PSNR, SSIM, and perceptual loss.
• Verification and validation by visual rating and evaluation of image sharpness by intensity profile comparisons. | The impact of the network has been characterized by several quality metrics such as peak signal-to-noise ratio (PSNR), structural similarity index (SSIM), and perceptual loss. In addition, the feature has been verified and validated by inhouse tests. These tests include visual rating and an evaluation of image sharpness by intensity profile comparisons of reconstructions with and without Deep Resolve Sharp. |

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

• Deep Resolve Boost: The document doesn't explicitly state a separate "test set" size. It mentions the "Training and Validation data" which includes:
  • TSE: more than 25,000 slices
  • HASTE: pre-trained on the TSE dataset and refined with more than 10,000 HASTE slices
  • EPI Diffusion: more than 1,000,000 slices
  • Data Provenance: The data covered a broad range of body parts, contrasts, fat suppression techniques, orientations, and field strength. No specific country of origin is mentioned, but the manufacturer (Siemens Healthcare GmbH) is based in Germany, and Siemens Medical Solutions USA, Inc. is the submitter. The data was "retrospectively created from the ground truth by data manipulation and augmentation."
• Deep Resolve Sharp: The document doesn't explicitly state a separate "test set" size. It mentions "Training and Validation data" from "on more than 10,000 high resolution 2D images."
  • Data Provenance: Similar to Deep Resolve Boost, the data covered a broad range of body parts, contrasts, fat suppression techniques, orientations, and field strength. Data was "retrospectively created from the ground truth by data manipulation." No specific country of origin is mentioned.

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

Not specified. The document states that the acquired datasets "represent the ground truth." There is no mention of expert involvement in establishing ground truth for the test sets. The focus is on technical metrics (PSNR, SSIM) and "visual comparisons" or "visual rating" which implies expert review, but the number and qualifications are not provided.

4. Adjudication Method for the Test Set

Not explicitly stated. The document mentions "visual comparisons" for Deep Resolve Boost and "visual rating" for Deep Resolve Sharp. This suggests subjective human review, but no specific adjudication method (like 2+1 or 3+1 consensus) is detailed.

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

No MRMC comparative effectiveness study is described for the AI features. The studies mentioned (sections 8 and 9) focus on evaluating the technical performance and image quality of the AI algorithms themselves, not on their impact on human reader performance.

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

Yes, standalone performance evaluation of the algorithms was conducted. The "Test Statistics and Test Results Summary" for both Deep Resolve Boost and Deep Resolve Sharp detail the evaluation of the network's impact using quantitative metrics (PSNR, SSIM, perceptual loss) and qualitative assessments ("visual comparisons," "visual rating," "intensity profile comparisons"). This represents the algorithm's performance independent of a human reader's diagnostic accuracy.

7. The Type of Ground Truth Used

The ground truth used for both Deep Resolve Boost and Deep Resolve Sharp was the acquired datasets themselves, representing the original high-quality or reference images/slices.

• For Deep Resolve Boost, input data was "retrospectively created from the ground truth by data manipulation and augmentation," including undersampling k-space lines, lowering SNR, and mirroring k-space data. The original acquired data serves as the target "ground truth" for the AI to reconstruct/denoise.
• For Deep Resolve Sharp, input data was "retrospectively created from the ground truth by data manipulation," specifically by cropping k-space data to create low-resolution input, with the original high-resolution data serving as the "output / ground truth" for training and validation.

8. The Sample Size for the Training Set

• Deep Resolve Boost:
  • TSE: more than 25,000 slices
  • HASTE: pre-trained on the TSE dataset and refined with more than 10,000 HASTE slices
  • EPI Diffusion: more than 1,000,000 slices
• Deep Resolve Sharp: more than 10,000 high resolution 2D images.

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

The ground truth for the training set was established as the acquired, unaltered (or minimally altered, e.g., removal of k-space lines to simulate lower quality input from high quality ground truth) raw imaging data.

• For Deep Resolve Boost: "The acquired datasets (as described above) represent the ground truth for the training and validation. Input data was retrospectively created from the ground truth by data manipulation and augmentation." This implies that the original, high-quality scans were considered the ground truth, and the AI was trained to restore manipulated, lower-quality versions to this original quality.
• For Deep Resolve Sharp: "The acquired datasets represent the ground truth for the training and validation. Input data was retrospectively created from the ground truth by data manipulation. k-space data has been cropped such that only the center part of the data was used as input. With this method corresponding low-resolution data as input and high-resolution data as output / ground truth were created for training and validation." Similar to Boost, the original, higher-resolution scans served as the ground truth.

Ask a specific question about this device

The MAGNETOM systems are indicated for use as magnetic resonance diagnostic devices (MRDD) that produce transverse, sagittal, coronal and oblique cross sectional images, spectroscopic images and/or spectra, and that display 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 systems described above may also be used for imaging during interventional procedures when performed with MR compatible devices such as in-room display and MR-Safe biopsy needles.

The subject device, syngo MR E11C system software, is being made available for the following MAGNETOM MR Systems:

• MAGNETOM Aera,
• MAGNETOM Skyra, ●
• MAGNETOM Prisma and
• MAGNETOM Prisma™ ●

The syngo MR E11C SW includes new sequences. new features and minor modifications of already existing features.

The provided text describes a 510(k) premarket notification for new software (syngo MR E11C) for Siemens MAGNETOM MR systems. However, it does not contain the detailed information required to answer all aspects of your request regarding acceptance criteria and a study proving device performance as typically expected for AI/ML device submissions.

This submission is for a software update to existing Magnetic Resonance Diagnostic Devices (MRDDs), and the focus is on demonstrating substantial equivalence to previously cleared predicate devices. The "study" mentioned is primarily non-clinical performance testing and software verification/validation, rather than a clinical study with acceptance criteria for specific diagnostic outcomes.

Here's an attempt to extract and infer information based on the provided text, highlighting what is present and what is missing:

1. Table of acceptance criteria and the reported device performance

The document does not explicitly state quantitative acceptance criteria for diagnostic performance or specific metrics. Instead, it relies on demonstrating that the new software's features perform "as intended" and maintain "equivalent safety and performance profile" compared to predicate devices.

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

• Test Set Sample Size: "Sample clinical images were taken for particular new and modified sequences." The specific number or characteristics of these images (sample size) is not provided.
• Data Provenance: The document does not specify the country of origin of the data or whether it was retrospective or prospective. It only mentions "sample clinical images," suggesting clinical data was used for assessment.

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

• This information is not provided. The document states "Image quality assessments... were completed," but does not detail who performed these assessments or how ground truth was established for them. For a diagnostic device, interpretation by a "trained physician" is mentioned in the Indications for Use, but this is a general statement about the device's usage, not specific to the assessment of the new software.

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

• This information is not provided.

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

• No, an MRMC comparative effectiveness study was not done. The document explicitly states: "No clinical tests were conducted to support the subject device and the substantial equivalence argument..."
• This submission is not for an AI-enhanced diagnostic tool in the sense of providing automated interpretations or assisting human readers in a measurable way with specific diagnostic outcomes. It's an update to MR imaging acquisition software. Therefore, the concept of "how much human readers improve with AI vs without AI assistance" does not apply in this context.

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

• The device is a Magnetic Resonance Diagnostic Device (MRDD) software update. Its output is images and/or spectra that are "interpreted by a trained physician" to "assist in diagnosis." As such, it is inherently a human-in-the-loop system. The non-clinical tests involved "Image quality assessments" and "Acoustic noise measurements," which are performance evaluations of the acquisition capabilities, not a standalone diagnostic interpretation by the algorithm.
• Therefore, a standalone diagnostic performance evaluation (algorithm only) in the context of providing a diagnosis was not performed or described.

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

• For "Image quality assessments," the type of ground truth is not explicitly stated. It can be inferred that it would likely involve visual assessment by experts against what is considered normal or expected for an MR image, potentially comparing to images acquired with predicate software or known anatomical/pathological features. However, specific ground truth methods like pathology or long-term outcomes data are not mentioned.

8. The sample size for the training set

• The document does not mention a separate training set or details about its size. This submission focuses on software changes and their verification, not on the development of a new AI model that requires a distinct training phase.

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

• Since a separate training set is not mentioned, the method for establishing its ground truth is also not provided.

Summary of what's present and what's missing:

This 510(k) submission primarily focuses on demonstrating that new software features (like quiet diffusion imaging, improved fast TSE, simultaneous multi-slice imaging, and a short acquisition time brain examination protocol) for existing MR systems maintain the fundamental technological characteristics, safety, and effectiveness of predicate devices. The "study" here is a series of non-clinical tests (image quality review, acoustic noise measurements, software V&V) rather than a clinical trial measuring diagnostic accuracy or reader performance. The level of detail you're asking for, especially concerning clinical study design elements like sample size, expert reader qualifications, adjudication methods, and ground truth establishment for diagnostic output, is typically found in submissions for AI/ML diagnostic tools that directly interpret images or provide diagnostic assistance, which is not the primary claim of this particular device update.

Ask a specific question about this device