Search Results

The Heartflow Analysis is an AI-based medical device software for the clinical quantitative and qualitative analysis of previously acquired Computed Tomography DICOM data for adult patients (ages 22 years and older) with suspected coronary artery disease. It provides anatomic data, plaque localization and characterization, as well as the calculations of FFRCT, a coronary physiological simulation, computed from simulated pressure, velocity and blood flow information obtained from a 3D computer model generated from static coronary CT images. The Heartflow Analysis is intended to support the risk assessment and functional evaluation of coronary artery disease.

The Heartflow Analysis is provided to support qualified clinicians to aid in the evaluation and risk assessment of coronary artery disease. The Heartflow Analysis is intended to be used by qualified clinicians in conjunction with the patient's clinical history, symptoms, and other diagnostic tests, as well as the clinician's professional judgment.

The Heartflow Analysis is an AI-based medical device software developed for the clinical quantitative and qualitative analysis of CT DICOM data. It is a tool for the analysis of CT DICOM-compliant cardiac images and data, to assess the anatomy and function of the coronary arteries in the risk stratification and evaluation of coronary artery disease.

The software displays coronary anatomy and functional information using graphics and text, including computed and derived quantities of percent stenosis, plaque volumes, blood flow, pressure and velocity, to aid the clinician in the assessment and treatment planning of coronary artery disease.

The Heartflow Analysis is performed on previously physician-acquired image data and is unrelated to acquisition equipment and clinical workstations.

Here's a breakdown of the acceptance criteria and the study that proves the device meets them, based on the provided FDA 510(k) Clearance Letter for HeartFlow Analysis:

1. Table of Acceptance Criteria and Reported Device Performance:

Note: The submission states that the subject device's plaque localization sensitivity was superior to the predicate, and its plaque quantification showed less volume error. The exact mean volume error for the predicate is not provided, making a direct comparison difficult beyond the stated significance. The reported 8.1 mm³ is the difference in mean volume error, implying the subject device's error was lower.

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

• Sample Size: 100 distinct patients, resulting in 583 unique lesions and 60,555 unique ground truth areas.
• Data Provenance:
  • Retrospective: The validation dataset consisted of "previously completed commercial cases that are maintained in a restricted library."
  • Country of Origin: The data was sourced from 67 different institutions across the United States, as detailed by state distribution in Table 6.
  • Scanner Manufacturers: Cases from various manufacturers were included (Siemens, GE Medical Systems, Toshiba, Philips, Canon Medical Systems, Siemens Healthineers, Fujifilm, Hitachi).

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

• Number of Experts: The document does not explicitly state the number of experts used to establish the ground truth for the test set. It mentions that plaque annotations and segmentations were collected from "expert physicians."
• Qualifications: The document describes them as "expert physicians" without providing specific details regarding their years of experience or specializations (e.g., radiologist, cardiologist).

4. Adjudication Method for the Test Set:

• Method: The document does not explicitly state an adjudication method like 2+1 or 3+1. It mentions that ground truth was established by "applying the plaque annotations as a mask over the segmentations, where both annotation and segmentation datapoints were collected from expert physicians." This suggests that the "expert physicians" provided the annotations and segmentations that formed the ground truth, but the process for resolving discrepancies among multiple experts (if more than one was involved per case) is not detailed.

5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study was Done and Effect Size of Human Improvement with AI vs. Without AI Assistance:

• MRMC Study: No, the provided FDA letter does not describe an MRMC comparative effectiveness study involving human readers with and without AI assistance. The study described focuses on the standalone performance of the AI algorithm (HeartFlow Analysis subject) compared to a previous version of the algorithm (HeartFlow Analysis predicate) and against expert annotations as ground truth.
• Effect Size of Human Improvement: Not applicable, as no MRMC study involving human readers' performance improvement with AI was conducted for this specific submission.

6. If a Standalone (Algorithm Only Without Human-in-the-Loop Performance) Study was Done:

• Yes, a standalone study was done. The entire "Summary of Performance Data" section describes the validation of the AI-based algorithm's performance (HeartFlow Analysis subject) against established ground truth, comparing it to an older version of the algorithm (HeartFlow Analysis predicate). This is a pure algorithm-only performance assessment. The device's intended use also states it "supports qualified clinicians to aid in the evaluation," indicating it's a tool, implying standalone performance evaluation is critical.

7. The Type of Ground Truth Used:

• Expert Consensus/Annotation: The ground truth was established using "plaque annotations as a mask over the segmentations, where both annotation and segmentation datapoints were collected from expert physicians." This indicates that the ground truth was derived from direct expert physician interpretation and labeling of the CT DICOM images.

8. The Sample Size for the Training Set:

• The document does not provide the sample size for the training set. It only states that the core technology "continues to be trained using deep learning (AI and machine learning) since 2015, to incorporate learnings from the volumes of CT data and studies."

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

• The document does not explicitly state how the ground truth for the training set was established. It notes that the algorithm "continues to be trained using deep learning... to incorporate learnings from the volumes of CT data and studies." While it describes the ground truth for the validation set as expert annotations/segmentations, it doesn't detail the training data's ground truth methodology. It's generally assumed that similar expert-derived ground truth would be used for training, but this is not explicitly confirmed in the provided text.

Ask a specific question about this device