Search Filters

Search Results

Found 2 results

510(k) Data Aggregation

    K Number
    K222938
    Device Name
    Ablation-fit
    Manufacturer
    Date Cleared
    2023-09-12

    (351 days)

    Product Code
    Regulation Number
    892.2050
    Reference & Predicate Devices
    Why did this record match?
    Reference Devices :

    K212896

    AI/MLSaMDIVD (In Vitro Diagnostic)TherapeuticDiagnosticis PCCP AuthorizedThirdpartyExpeditedreview
    Intended Use

    Ablation-fit is a medical imaging application available for use with liver ablation procedures.

    Ablation-fit is used to assist physicians in planning, permitting the graphical display of anatomy involved in the procedure, ablation targets and ablation needle placement.

    Ablation-fit is used to assist physicians in confirming ablation zones during follow-up.

    The software is not intended for diagnosis. The software is not intended to predict ablation volumes or predict ablation success.

    Device Description

    Ablation-fit is a stand-alone medical imaging software that integrates Reconstruction, Segmentation, Registration and Visualization algorithms into a user interface to support physicians during liver ablation treatments planning and follow-up.

    Ablation-fit allows to perform the entire workflow from DICOM (Digital Imaging and COmmunications in Medicine) images to 3D reconstruction of volume of interests, ablation probe placement and treatment outcome verification.

    Specifically, Ablation-fit main functionalities include:

    • Image loading from different supports (including PACS),
    • DICOM images handling and visualization in axial, sagittal, coronal views,
    • image segmentation,
    • tools for manual edit of segmentations.
    • 3D visualization,
    • virtualization of ablation probe placement,
    • pre- and post-treatment images registration.

    The software permits segmentation and 3D reconstruction of volumes of interest. The software contours all of this anatomic information not only in axial, sagittal, and coronal planes for 2D visualization, but also three-dimensionally. Every computed segmentation can be manually modified in the 2D axial visualization and consequently the three-dimensional mapping of the scan changes accordingly.

    Ablation-fit let the user simulate the virtual needle insertion and shows the desired ellipsoid of ablation.

    Once the ablation procedure has been performed, pre- and post-treatment scans are registered. Consequently, the software can verify whether the ablation zones entirely surrounds the lesion and the safety margin.

    AI/ML Overview

    Here's a breakdown of the acceptance criteria and the study details for the Ablation-fit device, based on the provided FDA 510(k) summary:

    Acceptance Criteria and Device Performance

    The document doesn't present a direct table of specific acceptance criteria with corresponding performance metrics in a single, clear format. However, it states that "software testing using retrospective image data of ablation procedures" was conducted to evaluate "the accuracy of Ablation-fit in assessing the outcome of lesion percutaneous thermal ablations and the accuracy of the automatically performed segmentations." It also mentions "Bench tests that compare the output of all segmentation and registration processes with ground truth annotated by qualified experts show that the algorithms performed as expected."

    Based on these statements, we can infer the following general acceptance criteria and reported performance:

    Acceptance Criteria CategoryReported Device Performance
    Segmentation AccuracyAlgorithms performed as expected (compared to qualified expert-annotated ground truth). Retrospective evaluation showed accuracy in automatically performed segmentations.
    Registration AccuracyAlgorithms performed as expected (compared to qualified expert-annotated ground truth).
    Assessment of Ablation Outcome AccuracyRetrospective evaluation showed accuracy in assessing the outcome of lesion percutaneous thermal ablations.
    Measurement AccuracyMeasurement Accuracy Test performed to evaluate the accuracy of measurements carried out with Ablation-fit software on CT images. (Specific metrics not provided in this summary).
    Functionality (User Interaction)All semi-automatic functionalities underwent testing by three radiologists to account for variability resulting from user interaction, and the system satisfied user demands and requirements. (Specific metrics not provided, but implies satisfactory performance with user variability accounted for).
    Compliance with Standards & RequirementsDesigned and developed according to ANSI AAMI IEC 62304:2006/A1:2016. Software verification and validation testing conducted according to FDA guidance. User acceptance test performed according to ANSI AAMI IEC 62366-1:2015+AMD1:2020.
    Safety and Effectiveness EquivalenceFunctions at least as safely and effectively as the designated predicate device and is essentially equivalent to it. Does not introduce any new potential safety risks.

    Study Details

    Here's the breakdown of the study information:

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

    • Sample Size for Test Set: The exact number of cases or images in the "retrospective image data of ablation procedures" used for software testing is not specified in this document.
    • Data Provenance: The data used for testing was "retrospective image data of ablation procedures." The country of origin for the data is not specified.

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

    • Number of Experts: "Qualified experts" were used to annotate ground truth for segmentation and registration processes. In addition, "three radiologists" performed testing for semi-automatic functionalities to account for user interaction variability.
    • Qualifications of Experts: The specific qualifications (e.g., years of experience, subspecialty) of the "qualified experts" and the "three radiologists" are not specified beyond their profession.

    4. Adjudication Method for the Test Set

    • The document implies that ground truth was "annotated by qualified experts." For the "semi-automatic functionalities," testing involved "three radiologists" to account for user variability. There is no explicit mention of an adjudication method (e.g., 2+1, 3+1 consensus) for establishing the ground truth, particularly expert consensus for discrepancies. It's possible annotation by a single "qualified expert" was considered ground truth, or an unstated consensus method was used.

    5. If a Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study Was Done

    • No, an MRMC comparative effectiveness study was not explicitly mentioned or described. The testing involved "three radiologists" testing "semi-automatic functionalities to account for variability resulting from user interaction," but this appears to be part of validating the device's interaction and robustness, not a comparative effectiveness study of human readers with vs. without AI assistance.
    • Effect Size of Human Readers Improvement with AI vs. without AI assistance: This information is not provided as an MRMC study was not described.

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

    • Yes, a standalone performance assessment was conducted for some aspects. The document states: "Bench tests that compare the output of all segmentation and registration processes with ground truth annotated by qualified experts show that the algorithms performed as expected." This implies an evaluation of the algorithm's performance in these tasks independent of a human user's interaction in the final output generation.
    • Additionally, "the accuracy of the automatically performed segmentations" was evaluated, which is a standalone assessment.

    7. The Type of Ground Truth Used

    • The primary type of ground truth used was expert consensus / expert annotation. Specifically, "ground truth annotated by qualified experts" was used for segmentation and registration processes.

    8. The Sample Size for the Training Set

    • The document does not specify the sample size used for the training set.

    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 only discusses the ground truth for the "test set" or "bench tests."
    Ask a Question

    Ask a specific question about this device

    K Number
    K211713
    Manufacturer
    Date Cleared
    2022-04-28

    (329 days)

    Product Code
    Regulation Number
    892.2050
    Reference & Predicate Devices
    Why did this record match?
    Reference Devices :

    K212896

    AI/MLSaMDIVD (In Vitro Diagnostic)TherapeuticDiagnosticis PCCP AuthorizedThirdpartyExpeditedreview
    Intended Use

    The AngioCloud Service enables visualization and measurement of cerebral blood vessels for preoperational planning and sizing for neurovascular interventions and surgery.

    General functionalities are provided such as:

    • Segmentation of neurovascular structures
    • · Centerline calculation
    • · Visualization of 3D vascular images
    • · Measurement and annotation tools
    • · Case sharing and reporting tools

    Information provided by the software is not intended in any way to eliminate, replace, or substitute for, in whole or in part, the healthcare provider's judgment and analysis of the patient's condition.

    Device Description

    The AngioCloud Service is a standalone web-application which intends to receive a 3D rotational angiography (3D-RA) DICOM dataset and provide interactive views to physicians for the visualization and measurement of cerebral vasculatures during preoperational planning. Physicians can query a 3D-RA dataset directly via the AngioCloud Service provided that they have an internet browser with access to the internet. General functionalities are provided such as:

    • Segmentation of neurovascular structures ●
    • . Centerline calculation
    • Visualization of 3D vascular images
    • . Measurement and annotation tools
    • Case sharing and reporting tools

    The AngioCloud Service runs as a web application on a standard Windows or Mac OS X based computer and can also be accessed on mobile web browsers, but with limited software functionalities enabled. The AngioCloud Service does not use any artificial intelligence or machine learning functionality and the main segmentation algorithm is based on level-set methods. The Visualization Toolkit (VTK) and Vascular Model Toolkit (VMTK) serve as important software libraries that the underlying algorithm of AngioCloud Service leverages for a number of computational operations such as 3D segmentation, geometric analysis, mesh generation, and surface data analysis for image-based modeling of blood vessels. The device does not contact the patient nor does it control any life-sustaining devices. Information provided by the AngioCloud Service is not intended in any way to eliminate, replace, or substitute for, in whole or in part, the healthcare provider's judgment and analysis of the patient's condition.

    AI/ML Overview

    The AngioCloud Service is a web application designed for the visualization and measurement of cerebral blood vessels for preoperational planning and sizing for neurovascular interventions and surgery. It includes functionalities such as segmentation of neurovascular structures, centerline calculation, visualization of 3D vascular images, measurement and annotation tools, and case sharing and reporting tools. The device does not use artificial intelligence or machine learning. The following describes the acceptance criteria and the study proving the device meets these criteria.

    Acceptance Criteria and Reported Device Performance

    The performance testing was conducted primarily to evaluate the AngioCloud Service's segmentation module. The acceptance criteria were based on two metrics: Hausdorff Distance (dH) and Dice Coefficient (DC). These metrics were used to compare the segmentations performed by the AngioCloud Service against a reference standard. While specific numeric acceptance thresholds for dH and DC are not provided in the document, the conclusion states that "Both the mean DC and mean dH for each group met the acceptance criteria."

    Metric / FunctionalityAcceptance Criteria (Implicit)Reported Device Performance
    Segmentation Module
    Hausdorff Distance (dH)Met pre-defined criteria (specific numeric thresholds not provided)Mean dH for each group met the criteria
    Dice Coefficient (DC)Met pre-defined criteria (specific numeric thresholds not provided)Mean DC for each group met the criteria
    Overall Software Functionalities
    DICOM images importationVerifiedTested and met
    Case managementVerifiedTested and met
    Auto-segmentation and manual segmentationVerifiedTested and met
    Visualization of 3D vascular images (desktop & mobile)VerifiedTested and met
    Centerline calculation (desktop & mobile)VerifiedTested and met
    Measurements and annotation tool (desktop & mobile)VerifiedTested and met
    Case sharing and reporting toolVerifiedTested and met

    Study Details

    1. Sample Size and Data Provenance:

      • Test Set Sample Size: The document states that "Anatomically relevant phantom models derived from clinical scans were acquired using Siemens (Artis Q with PURE) and Phillips (Allura Xper FD 20/20) Angio suites." The exact number of phantom models or clinical scans used to derive them is not specified.
      • Data Provenance: The data used for the test set were "phantom models derived from clinical scans" acquired using Siemens and Phillips Angio suites. The country of origin and whether the data was retrospective or prospective are not explicitly stated. However, the use of "phantom models" implies a controlled, non-patient-specific testing environment, likely based on retrospective clinical data used for model creation.
    2. Number of Experts and Qualifications for Ground Truth:

      • The document does not specify the number of experts used to establish the ground truth for the test set or their qualifications.
    3. Adjudication Method for the Test Set:

      • The document does not specify any adjudication method used for the test set.
    4. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study:

      • No MRMC comparative effectiveness study was mentioned. The device does not utilize AI/ML, and the testing focused on the performance of its segmentation module against a reference standard, not on human reader improvement with AI assistance.
    5. Standalone Performance:

      • Yes, a standalone performance evaluation was conducted for the segmentation module. The Hausdorff distance (dH) and DICE coefficient (DC) were used to evaluate the algorithm's segmentation performance against a reference standard. The study assesses the device's inherent functionality without human-in-the-loop performance being a primary measure. Human interaction is for adjusting the segmentation threshold ("manual adjust the segmentation threshold" or "Auto" option), but the core performance evaluation metrics (dH, DC) are for the algorithm's output relative to the ground truth.
    6. Type of Ground Truth Used:

      • The ground truth for the segmentation performance evaluation was based on a "reference standard" from "anatomically relevant phantom models derived from clinical scans." This implies a highly accurate or 'gold standard' segmentation for these phantom models, likely established through precise measurements or expert consensus on the phantom data, rather than direct pathology or outcomes data from live patients.
    7. Training Set Sample Size:

      • The document states that the AngioCloud Service "does not use any artificial intelligence or machine learning functionality." Therefore, there is no "training set" in the context of machine learning model development. The segmentation algorithm is based on "level-set methods" and leverages "The Visualization Toolkit (VTK) and Vascular Model Toolkit (VMTK)."
    8. Ground Truth Establishment for Training Set:

      • As there is no AI/ML component, there is no "training set" with ground truth established in the machine learning sense. The underlying algorithms (level-set methods, VTK, VMTK) are established scientific and engineering tools, not trained models requiring annotated ground truth data for learning.
    Ask a Question

    Ask a specific question about this device

    Page 1 of 1