Search Results

This device is a digital radiography/fluoroscopy system used in a diagnostic and interventional angiography configuration. The system is indicated for use in diagnostic and angiographic procedures for blood vessels in the heart, brain, abdomen and lower extremities.

αEvolve Imaging is an imaging chain intended for adults, with Artificial Intelligence Denoising (AID) designed to reduce noise in real-time fluoroscopic images and signal enhancement algorithm, Multi Frequency Processing (MFP).

Device Description

The Alphenix, INFX-8000V/B, INFX-8000V/S, V9.6 with αEvolve Imaging, is an interventional X-ray system with a floor mounted C-arm as its main configuration. An optional ceiling mounted C-arm is available to provide a bi-plane configuration where required. Additional units include a patient table, X-ray high-voltage generator and a digital radiography system. The C-arms can be configured with designated X-ray detectors and supporting hardware (e.g. X-ray tube and diagnostic X-ray beam limiting device). The Alphenix, INFX-8000V/B, INFX-8000V/S, V9.6 with αEvolve Imaging includes αEvolve Imaging, an imaging chain intended for adults, with Artificial Intelligence Denoising (AID) designed to reduce noise in real-time fluoroscopic images and signal enhancement algorithm, Multi Frequency Processing (MFP).

AI/ML Overview

Here's an analysis of the acceptance criteria and the study proving the device meets them, based solely on the provided FDA 510(k) summary:

Overview of the Device and its New Feature:

The device is the Alphenix, INFX-8000V/B, INFX-8000V/S, V9.6 with αEvolve Imaging. It's an interventional X-ray system. The new feature, αEvolve Imaging, includes Artificial Intelligence Denoising (AID) to reduce noise in real-time fluoroscopic images and a signal enhancement algorithm, Multi Frequency Processing (MFP). The primary claim appears to be improved image quality (noise reduction, sharpness, contrast, etc.) compared to the previous version's (V9.5) "super noise reduction filter (SNRF)."

1. Table of Acceptance Criteria and Reported Device Performance

The 510(k) summary does not explicitly state "acceptance criteria" with numerical thresholds for each test. Instead, it describes various performance evaluations and their successful outcomes. For the clinical study, the success criteria are clearly defined.

Acceptance Criteria (Inferred/Stated)	Reported Device Performance
Bench Testing (Image Quality)
1. Change in Image Level, Noise & Structure: AID to be better at preserving mean image intensity, improved denoising, and image structure preservation compared to SNRF.	AID determined to be better at preserving mean image intensity and suggested to have improved denoising and image structure preservation (using student's t-test).
2. Signal-to-Variance Ratio (SVR) and Signal-to-Noise Ratio (SNR): AID to show improved ability to preserve image signal while decreasing image noise compared to SNRF.	AID determined to have improved ability to preserve image signal while decreasing image noise (using student's t-test).
3. Modulation Transfer Function (MTF): Improved performance for low-to-mid frequencies and similar high-frequency region compared to SNRF.	Results showed improved performance for low-to-mid frequencies in all test cases, and high-frequency region of MTF curve was similar for AID and SNRF in majority of cases (using student's t-test).
4. Robustness to Detector Defects: Detector defects to be sufficiently obvious to inform clinician of service need, and image quality outside the defect area to remain visually unaffected, facilitating procedure completion.	Detector defects were sufficiently obvious, and image quality outside the area of the detector defect remained visually unaffected, facilitating sufficient image quality to finish the procedure.
5. Normalized Noise Power Spectrum (NNPS): AID to have smaller noise magnitude in the frequency range of ~0.1 cycles/mm to 1.4 cycles/mm, with negligible differences above 1.4 cycles/mm.	AID had a smaller noise magnitude in the frequency range of ~0.1 cycles/mm to 1.4 cycles/mm. Noise magnitudes above 1.4 cycles/mm were very small and differences considered negligible.
6. Image Lag Measurement: AID to perform better in reducing image lag compared to SNRF.	AID determined to perform better in reducing image lag (using student's t-test).
7. Contrast-to-Noise Ratio (CNR) of Low Contrast Object: AID to show significantly higher CNR for low-contrast elements compared to SNRF.	AID had a significantly higher CNR than images processed with SNRF for all elements and test cases (using student's t-test).
8. Contrast-to-Noise Ratio (CNR) of High Contrast Object: AID to show significantly higher CNR for high-contrast objects (guidewire, vessels) compared to SNRF.	AID had a significantly higher vessel and guidewire CNR than images processed with SNRF for all test cases (using student's t-test).
Clinical Study (Reader Study)
Overall Preference (Binomial Test): Image sequences denoised by AID chosen significantly more than 50% of the time over SNRF.	The Binomial test found that image sequences denoised by AID were chosen significantly more than 50% of the time (indicating overall preference).
Individual Image Quality Metrics (Wilcoxon Signed Rank Test): Mean score of AID images significantly higher than SNRF for sharpness, contrast, confidence, noise, and absence of image artifacts.	The mean score of AID imaging chain images was significantly higher than that of the SNRF imaging chain for sharpness, contrast, confidence, noise, and the absence of image artifacts.
Generalizability: Algorithm to demonstrate equivalent or improved performance compared to the predicate with diverse clinical data.	Concluded that the subject algorithm demonstrated equivalent or improved performance, compared to the predicate device, as demonstrated by the results of the above testing.

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

The 510(k) summary provides the following information about the clinical test set:

Clinical Dataset Source: Patient image sequences were acquired from three hospitals:
- Memorial Hermann Hospital (Houston, Texas, USA)
- Waikato Hospital (Hamilton, New Zealand)
- Saiseikai Kumamoto Hospital (Kumamoto, Japan)
Data Provenance: The study used retrospective "patient image sequences" for side-by-side comparison. The summary does not specify if the acquisition itself was prospective or retrospective, but the evaluation of pre-existing sequences makes it a retrospective study for the purpose of algorithm evaluation.
Sample Size: The exact number of patient image sequences or cases used in the clinical test set is not specified in the provided document. It only mentions that the sequences were split into four BMI subgroups.

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

Number of Experts: The document states the clinical comparison was "reviewed by United States board-certified interventional cardiologists." The exact number of cardiologists is not specified.
Qualifications: "United States board-certified interventional cardiologists." No mention of years of experience or other specific qualifications is provided.

4. Adjudication Method for the Test Set

The document describes a "side-by-side comparison" reviewed by experts in the clinical performance testing section. For the overall preference and individual image quality metrics, statistical tests (Wilcoxon signed rank test and Binomial test) were used. This implies that the experts rated or expressed preference for both AID and SNRF images, and these individual ratings/preferences were then aggregated and analyzed.

The exact adjudication method (e.g., 2+1, 3+1 consensus) for establishing a ground truth or a final decision on image quality aspects is not explicitly stated. It seems each expert provided their assessment, and these assessments were then statistically analyzed for superiority rather than reaching a consensus for each image pair.

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

MRMC Study: Yes, a type of MRMC comparative study was conducted. The clinical performance testing involved multiple readers (US board-certified interventional cardiologists) evaluating multiple cases (patient image sequences).
Effect Size of Human Readers' Improvement with AI Assistance: The study directly compared AID-processed images to SNRF-processed images in a side-by-side fashion. It doesn't measure how much humans improve with AI assistance in a diagnostic task (e.g., how much their accuracy or confidence improves when using AI vs. not using AI). Instead, it measures the perceived improvement in image quality of the AI-processed images when evaluated by human readers.
- The study determined: "the mean score of the AID imaging chain images was significantly higher than that of the SNRF imaging chain with regard to sharpness, contrast, confidence, noise, and the absence of image artifacts."
- And for overall preference, "the Binomial test found that the image sequences denoised by AID were chosen significantly more than 50% of the time."
This indicates a statistically significant preference for and higher perceived image quality in AID-processed images by readers. However, it does not quantify diagnostic performance improvement with AI assistance, as it wasn't a study of diagnostic accuracy but rather image quality assessment. The "confidence" metric might hint at improved reader confidence using AID images, but it's not a direct measure of diagnostic effectiveness.

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

Yes, extensive standalone performance testing of the AID algorithm was conducted through "Performance Testing – Bench" and "Image Quality Evaluations." This involved objective metrics and phantom studies without human subjective assessment.

Examples include:

Change in Image Level, Noise and Structure
Signal-to-Variance Ratio (SVR) and Signal-to-Noise Ratio (SNR)
Modulation Transfer Function (MTF)
Robustness to Detector Defects (visual comparison, but the algorithm's output is purely standalone)
Normalizes Noise Power Spectrum (NNPS)
Image Lag Measurement
Contrast-to-Noise Ratio of a Low Contrast Object
Contrast-to-Noise Ratio of a High Contrast Object

7. The Type of Ground Truth Used

For Bench Testing: The ground truth for bench tests was primarily established through physical phantoms and objective image quality metrics. For example, the anthropomorphic chest phantom, low-contrast phantom, and flat field fluoroscopic images provided known characteristics against which AID and SNRF performance were measured using statistical tests.
For Clinical Study: The ground truth for the clinical reader study was established by expert opinion/subjective evaluation (preference and scores for sharpness, contrast, noise, confidence, absence of artifacts) from "United States board-certified interventional cardiologists." There is no mention of a more objective ground truth like pathology or outcomes data for the clinical image evaluation.

8. The Sample Size for the Training Set

The document does not provide any information about the sample size used for the training set of the Artificial Intelligence Denoising (AID) algorithm.

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

The document does not provide any information about how the ground truth for the training set was established. It describes the AID as "Artificial Intelligence Denoising (AID) designed to reduce noise," implying a machine learning approach, but details on its training are missing from this summary.

Ask a Question

Ask a specific question about this device

Page 1 of 1