(119 days)
Amide Proton Transfer weighted imaging (3D APT) is a software option intended for use on Ingenia 3.0T and Ingenia 3.0T CX MR Systems. 3D APT is indicated for use in magnetic resonance imaging of the brain. 3D APT consists of an acquisition and reconstruction technique employing frequency-selective magnetization transfer effects to derive images reflecting the spatial distribution of amide protons, and thereby protein density. 3D APT images may assist a trained physician in diagnosis and therapy planning. APTW can be combined with multi-coil acceleration approaches (SENSE).
3D Amide Proton Transfer (3D APT) is an extension of off-resonance magnetization transfer imaging.
3D APT images display a spatial distribution of amide protons. The RF-shimmed saturation is generated by alternating excitation from the 2 channels of the MultiTransmit system.
3D APT images are obtained by subtracting the saturated image intensity at the 2 mirrored frequency offsets relative to the water frequency (±3.5 ppm). Parameters are optimized to null the difference signal for normal brain tissue.
The provided text, a 510(k) summary for the Philips 3D APT device, does not contain the detailed clinical study information typically required to fully answer the request regarding acceptance criteria and performance studies for an AI/ML medical device.
The document focuses on demonstrating substantial equivalence to a predicate device (Ingenia 1.5T/3.0T R5.3) based on design features, fundamental scientific technology, and indications for use. It mentions "non-clinical performance (verification and validation) tests" that complied with standards and guidance, but these are typically engineering and system validation tests, not clinical performance studies involving a test set, ground truth experts, or MRMC studies that would be relevant for an AI/ML product's clinical performance claims.
The 3D APT device is described as "an acquisition and reconstruction technique employing frequency-selective magnetization transfer effects to derive images reflecting the spatial distribution of amide protons, and thereby protein density." It's a software option that "may assist a trained physician in diagnosis and therapy planning." This description suggests it's a tool for imaging acquisition and processing, rather than a diagnostic AI algorithm that independently generates classifications or predictions based on image analysis. Therefore, the type of "acceptance criteria" and "study" would be different than for an AI diagnostic device.
Given these limitations from the provided text, I can only address parts of your request based on what's available and infer what might be relevant for such a device in the absence of explicit AI/ML performance data.
Here's an attempt to answer based on the provided document, highlighting what is not present:
Device: Philips 3D APT (Amide Proton Transfer weighted MRI)
The Philips 3D APT is a software option for Philips Ingenia 3.0T and Ingenia 3.0T CX MR Systems, indicated for use in magnetic resonance imaging of the brain. It's an acquisition and reconstruction technique that derives images reflecting the spatial distribution of amide protons and protein density, intended to assist trained physicians in diagnosis and therapy planning.
Based on the 510(k) summary, the acceptance criteria and study described are not for an AI/ML diagnostic system with a specific performance metric like sensitivity/specificity against a ground truth. Instead, the submission focuses on demonstrating substantial equivalence to existing predicate MRI systems (Ingenia 1.5T/3.0T R5.3) and ensuring the new image acquisition and reconstruction technique functions as intended and safely.
The document mentions "non-clinical performance (verification and validation) tests, which complied with the requirements specified in the international and FDA-recognized consensus standards and device-specific guidance." This typically refers to technical validation, image quality assessment (e.g., signal-to-noise ratio, spatial resolution, artifacts), safety (e.g., SAR limits), and functional performance within engineering specifications, rather than clinical efficacy studies in the context of diagnostic accuracy.
Therefore, the requested tables and details on AI/ML-specific performance metrics (e.g., sensitivity, specificity, AUC) and associated clinical study designs (test sets, ground truth methodology, expert adjudication, MRMC studies) are not present in this 510(k) summary. The device appears to be a novel MRI sequence/acquisition method, not a standalone AI diagnostic algorithm performing image analysis.
1. A table of acceptance criteria and the reported device performance
Since this is an MR image acquisition and reconstruction technique, not an AI diagnostic algorithm, the acceptance criteria would relate to imaging quality, consistency, and safety, not diagnostic performance metrics like accuracy, sensitivity, or specificity against a clinical ground truth. The document does not provide a specific table of acceptance criteria and reported performance for these non-clinical tests. It only states: "The results of these tests demonstrate that the proposed 3D Amide Proton Transfer (3D APT) meets the acceptance criteria and is adequate for its intended use."
2. Sample size used for the test set and the data provenance
- Test Set Sample Size: Not specified in terms of clinical cases for diagnostic performance. The "non-clinical performance tests" would likely involve phantoms and potentially a limited number of healthy volunteers for image quality assessment, but details are not provided.
- Data Provenance: Not specified. Again, for non-clinical performance, this would refer to the characteristics of phantoms or human subjects (if any) used for technical validation, not a clinical patient cohort.
3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts
Not applicable in the context of this 510(k) summary, as it describes an MRI acquisition technique, not a diagnostic AI that requires expert-established ground truth for performance evaluation on clinical cases. The "ground truth" for non-clinical performance would be based on physical measurements from phantoms or known parameters of the MR system.
4. Adjudication method for the test set
Not applicable. There's no indication of a clinical test set requiring expert adjudication for ground truth.
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
Not applicable. This is not an AI-assisted reading device in the context of MRMC studies. It's a new imaging contrast technique.
6. If a standalone (i.e. algorithm only without human-in-the-loop performance) was done
This device is used by humans (trained physicians) for diagnosis and therapy planning. Its "standalone" performance would be its ability to generate the 3D APT images reliably and with sufficient quality, which is covered by the general "non-clinical performance tests" but without detailed metrics in the summary. It's not a standalone diagnostic algorithm in the sense of providing an automated diagnosis.
7. The type of ground truth used (expert consensus, pathology, outcomes data, etc)
For a new imaging sequence like 3D APT, the "ground truth" (in the context of showing it works as intended) would typically involve:
- Physical principles: Confirmation that the sequence correctly applies RF pulses, gradient fields, and signal processing to generate images based on amide proton transfer effects.
- Phantom studies: Using phantoms with known properties to assess image quality, quantitative accuracy (if applicable), and consistency.
- Known physiological effects: Observing expected signal changes in human subjects in areas known to have certain protein densities, though this is for validation of the physics of the sequence, not diagnostic accuracy.
8. The sample size for the training set
Not applicable. This document describes an acquisition and reconstruction technique, not a machine learning algorithm that requires a "training set" in the conventional sense. The "training" for such a system comes from engineering design, physics principles, and potentially iterative refinement based on phantom and healthy volunteer studies.
9. How the ground truth for the training set was established
Not applicable, as there is no "training set" for an AI/ML algorithm described in this application.
In conclusion, the 510(k) summary for Philips 3D APT focuses on establishing substantial equivalence and general safety/performance of a new MRI acquisition sequence, rather than the clinical performance metrics of a diagnostic AI/ML algorithm that would involve a detailed test set, expert ground truth, and comparative studies with human readers.
§ 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.