Search Results

SenoBright HD is an extension of the existing indication for diagnostic mammography with Senographe Pristina. The SenoBright HD application shall enable contrast enhanced breast imaging a dual energy technique. This imaging technique can be used as an adjunct following mammography and ultrasound exams to help localize a known or suspected lesion.

This submission is proposing a software update to SenoBright HD consisting of an improvement of the existing "recombination" algorithm with the New Image Recombination Algorithm (NIRA) by adding a local estimation of breast thickness in the images recombination to account for the non-uniformity of the breast thickness, and by compensating for potential patient movement between the 2 CESM acquisitions (Low Energy and High Energy).

SenoBright HD (K172404) is the name of Senographe Pristina FFDM system allowing to perform Contrast Enhanced Spectral Mammography (CESM) application.

The CESM acquisition technique consists in acquiring two images (one High Energy and one Low Energy) in sequence and under the same breast compression after patient injection with an iodinated contrast media. The two images are then recombined through a post-processing algorithm.

This design change is a software and labeling only option, compatible with SenoBright HD installed base and does not require any hardware modification on the Senographe Pristina platform.

Here's a breakdown of the acceptance criteria and study details for the SenoBright HD device with the New Image Recombination Algorithm (NIRA), based on the provided document:

1. Table of Acceptance Criteria & Reported Device Performance

The acceptance criteria are not explicitly stated in a quantitative manner (e.g., "must achieve X% sensitivity"). Instead, the document focuses on demonstrating non-inferiority or improvement compared to the predicate device (SenoBright HD K172404) and showing clinical acceptability. The performance is reported in terms of assessments made by radiologists.

2. Sample Size and Data Provenance

• Test Set Sample Size:
  • Clinical Image Evaluation (Initial Acceptability): 10 images
  • Clinical Image Evaluation (Comparative): 50 clinical images
• Data Provenance: The document does not explicitly state the country of origin for the clinical data used in the studies. Given GE Healthcare's presence in France (as per the submitter's address), it is plausible the data could originate from there or other international sites. The data is described as "clinical images," implying retrospective data from a clinical setting. It is not stated whether it was prospective or retrospective, but the description "clinical images" often implies retrospective collection for such comparative studies.

3. Number of Experts and Qualifications

• Number of Experts: 3 independent MQSA-qualified radiologists used for both clinical image evaluations.
• Qualifications: "MQSA-qualified radiologists." MQSA (Mammography Quality Standards Act) qualification indicates that these radiologists meet specific federal standards for interpreting mammograms in the United States, including training, experience, and continuing education requirements. The specific years of experience for each expert are not provided.

4. Adjudication Method for the Test Set

The document does not explicitly describe an adjudication method (like 2+1, 3+1). It states that the evaluations were "performed by 3 independent MQSA-qualified radiologists." This suggests that each radiologist independently assessed the images, and the reported percentages (97%, 99%, 98%) likely represent the proportion of cases where at least two out of three (or possibly all three) agreed on the assessment, or an aggregatetion of individual assessments, but the specific consensus/adjudication rule is not detailed.

5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study and Effect Size

• A comparative effectiveness study was performed as a form of reader study, but it wasn't a "human readers improve with AI vs without AI assistance" MRMC study in the sense of AI assisting the human reader.
• Instead, this study compared the quality of images generated by the new algorithm (NIRA) against the quality of images generated by the previous algorithm (Predicate SenoBright HD), both assessed by human readers.
• Therefore, there's no direct "effect size of how much human readers improve with AI vs without AI assistance" as the AI (NIRA) is the image generation method being evaluated, not an assistance tool for the human reader's diagnostic performance. The human readers are evaluating the output of the AI.

6. Standalone Performance (Algorithm Only)

The document primarily focuses on the output quality of the algorithm as perceived by human readers, rather than a quantifiable "standalone" diagnostic performance (e.g., sensitivity/specificity for detecting lesions). The phantom testing is a form of standalone performance evaluation for image quality metrics, but not for diagnostic accuracy in a clinical context.

7. Type of Ground Truth Used

The ground truth for the clinical image evaluations was effectively the expert consensus/assessment of the image quality metrics (contrast uptake visibility, artifact visibility, overall image quality) by the 3 MQSA-qualified radiologists. There is no mention of pathology or outcomes data being used as the clinical ground truth for lesion presence/absence for diagnostic performance evaluation, as the study was focused on image quality assessment.

8. Sample Size for the Training Set

The document does not provide any information regarding the sample size used for training the New Image Recombination Algorithm (NIRA). This is typical for premarket notifications where the focus is on verification and validation of changes rather than the internal development details of the algorithm itself.

9. How Ground Truth for Training Set was Established

The document does not provide any information on how the ground truth for the training set (if any was used for supervised learning of NIRA) was established. As NIRA is described as an "evolution" of an existing recombination algorithm, it might involve engineering improvements rather than a machine learning model that requires a labeled training set in the typical sense. It states NIRA accounts for "non-uniformity of the breast thickness" and compensates for "potential patient movement," suggesting algorithmic improvements based on physical principles and image characteristics rather than purely data-driven supervised learning.

Ask a specific question about this device

SenoBright HD is an extension of the existing indication for diagnostic mammography with Senographe Pristina. The SenoBright HD application shall enable contrast enhanced breast imaging a dual energy technique. This imaging technique can be used as an adjunct following mammography and ultrasound exams to help localize a known or suspected lesion.

The subject of this submission is a modification of Senographe Pristina FFDM system (cleared in K162268) that will introduce a new imaging option called SenoBright HD. This imaging option has been previously cleared for Senographe Essential FFDM system in K103485,marketed as SenoBright Contrast Enhanced Spectral Mammography (CESM). The dual energy exposures will be done following an iodine based contrast injection of the patient and with a single breast compression. The new mode of operation for Senographe Pristina system is referred to as SenoBright HD Contrast Enhanced Spectral Mammography (CESM) due to the nature of taking an exposure with the x-ray spectrum optimized for general mammographic imaging and a second exposure with the x-ray spectrum optimized for the iodine based contrast image. The main modification of the Senographe Pristina system is the addition of software feature to control the low and high energy images and post processing and recombination of those images to create FFDM like image and recombined image. These two images allow the visualization of the breast tissue in a way that is typical and familiar for mammographic imaging and the x-ray contrast enhancement in the breast at the same time.

Here's an analysis of the provided text to extract information about the acceptance criteria and the study that proves the device meets them:

1. A table of acceptance criteria and the reported device performance

The document does not explicitly state quantitative acceptance criteria in a table format, nor does it provide specific reported device performance metrics against such criteria. Instead, it makes a general statement about "appropriate performance" and "good" image quality.

2. Sample size used for the test set and the data provenance (e.g. country of origin of the data, retrospective or prospective)

• Test Set Sample Size: The document only mentions a "Clinical image evaluation was performed" without specifying the number of images or cases in the test set.
• Data Provenance: Not specified (e.g., country of origin, retrospective or prospective).

3. Number of experts used to establish the ground truth for the test set and the qualifications of those experts (e.g. radiologist with 10 years of experience)

• Number of Experts: "expert radiologists" (plural) were used, but the exact number is not specified.
• Qualifications of Experts: Only "expert radiologists" is stated, without details on their specific experience (e.g., years of experience, board certifications).

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

The document does not specify an adjudication method for the test set. It only mentions that an evaluation was performed by "expert radiologists."

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 multi-reader multi-case (MRMC) comparative effectiveness study focusing on human readers improving with AI vs. without AI assistance was not done. The device description explicitly states: "The SenoBright HD application shall enable contrast enhanced breast imaging a dual energy technique. This imaging technique can be used as an adjunct following mammography and ultrasound exams to help localize a known or suspected lesion." This implies it is a diagnostic imaging tool, not an AI-assisted reader tool.

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

The device itself is an imaging system (Full-field digital mammography system with contrast enhancement), not an algorithm. Therefore, a "standalone algorithm only" performance study is not directly applicable in the way it would be for an AI-CAD device. The "Clinical image evaluation" mentioned was likely to assess the quality and utility of the images produced by the device, which would then be interpreted by human readers.

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

The document states "A Clinical image evaluation was performed showing that image quality of SenoBright HD in a clinical setting is good as assessed by expert radiologists." This implies that the ground truth for the image quality assessment was expert radiologist assessment/consensus. It does not mention pathology or outcomes data as ground truth for this specific evaluation, as the evaluation focused on image quality rather than diagnostic accuracy against a definitive truth for lesions.

8. The sample size for the training set

The document does not provide any information about a specific "training set" or its size. The device is a modification of an existing FFDM system, introducing new software features for image acquisition and processing. This suggests that the development likely involved engineering and image processing adjustments rather than deep learning model training in the conventional sense that would require a distinct "training set."

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

As no training set is mentioned for an AI model, this question is not applicable based on the provided text. The "ground truth" for the development of the image processing might have involved phantom studies and clinical images used for optimization, but these are not described as a formal "training set" with established ground truth in the context of machine learning.

Ask a specific question about this device