Search Filters

Search Results

Found 1 results

510(k) Data Aggregation

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

    HERA Z20 Diagnostic Ultrasound System; HERA Z20e Diagnostic Ultrasound System; HERA Z20s Diagnostic Ultrasound System; R20 Diagnostic Ultrasound System; HERA Z30 Diagnostic Ultrasound System; R30 Diagnostic Ultrasound System and probes are designed to obtain ultrasound images and analyze body fluids.

    The clinical applications include: Fetal/Obstetrics, Abdominal, Gynecology, Intra-operative, Pediatric, Small Organ, Neonatal Cephalic, Adult Cephalic, Trans-rectal, Trans-vaginal, Muscular-Skeletal (Conventional, Superficial), Urology, Cardiac Adult, Cardiac Pediatric, Thoracic, Trans-esophageal (Cardiac), Peripheral vessel and Ophthalmic.

    It is intended for use by, or by the order of, and under the supervision of, an appropriately trained healthcare professional who is qualified for direct use of medical devices. It can be used in hospitals, private practices, clinics and similar care environment for clinical diagnosis of patients.

    Modes of Operation: 2D mode, M mode, Color Doppler mode, Pulsed Wave (PW) Doppler mode, Continuous Wave (CW) Doppler mode, Tissue Doppler Imaging (TDI) mode, Tissue Doppler Wave (TDW) mode, Power Doppler (PD) mode, ElastoScan Mode, MV-Flow Mode, Multi Image mode(Dual, Quad), Combined modes, 3D/4D mode.

    Device Description

    HERA Z20 Diagnostic Ultrasound System; HERA Z20e Diagnostic Ultrasound System; HERA Z20s Diagnostic Ultrasound System; R20 Diagnostic Ultrasound System; HERA Z30 Diagnostic Ultrasound System; R30 Diagnostic Ultrasound System are a general purpose, mobile, software controlled, diagnostic ultrasound system. Their function is to acquire ultrasound data and to display the data as 2D mode, Color Doppler mode, Power Doppler (PD) mode, M mode, Pulsed Wave (PW) Doppler mode, Continuous Wave (CW) Doppler mode, Tissue Doppler Imaging (TDI) mode, Tissue Doppler Wave (TDW) mode, ElastoScan Mode, Combined modes, MV-Flow mode, Multi-Image mode(Dual, Quad), 3D/4D mode.

    HERA Z20 Diagnostic Ultrasound System; HERA Z20e Diagnostic Ultrasound System; HERA Z20s Diagnostic Ultrasound System; R20 Diagnostic Ultrasound System; HERA Z30 Diagnostic Ultrasound System; R30 Diagnostic Ultrasound System also give the operator the ability to measure anatomical structures and offer analysis packages that provide information that is used to make a diagnosis by competent health care professionals. HERA Z20 Diagnostic Ultrasound System; HERA Z20e Diagnostic Ultrasound System; HERA Z20s Diagnostic Ultrasound System; R20 Diagnostic Ultrasound System; HERA Z30 Diagnostic Ultrasound System; R30 Diagnostic Ultrasound System have a real time acoustic output display with two basic indices, a mechanical index and a thermal index, which are both automatically displayed.

    AI/ML Overview

    Here's a breakdown of the acceptance criteria and the study details for each AI-ML based software feature described in the FDA 510(k) summary:

    Overview of Acceptance Criteria and Device Performance (AI-ML Features)

    The provided document details the testing and performance for several new and updated AI-ML based software features. The information given for each feature constitutes the acceptance criteria and the device's reported performance against those criteria.

    1. Table of Acceptance Criteria and Reported Device Performance

    AI/ML FeatureAcceptance CriterionReported Device Performance
    AbdomenAssist - Kidney Length MeasurementSuccess Rate: Within a pre-specified clinical error margin compared to reference standard.100% [94.17%, 100.00%] (Success Rate)
    Bland-Altman mean difference (Bias) / 95% Limits of Agreement (LoA).1.5% (Bias); [-4.53%, 7.64%] (95% LoA)
    AbdomenAssist - Spleen Length MeasurementSuccess Rate: Within a pre-specified clinical error margin compared to reference standard.96.77% [88.98%, 99.11%] (Success Rate)
    Bland-Altman mean difference (Bias) / 95% Limits of Agreement (LoA).2.8% (Bias); [-7.02%, 12.61%] (95% LoA)
    BladderAssist - Bladder Width Measurement (First Instance)Success Rate: Within a pre-specified clinical error margin compared to reference standard.95.24% [86.91%, 98.37%] (Success Rate)
    Bland-Altman mean difference (Bias) / 95% Limits of Agreement (LoA).-4.81% (Bias); [-12.94%, 3.31%] (95% LoA)
    BladderAssist - Bladder Width Measurement (Second Instance)Success Rate: Within a pre-specified clinical error margin compared to reference standard.96.83% [89.14%, 99.13%] (Success Rate)
    Bland-Altman mean difference (Bias) / 95% Limits of Agreement (LoA).-3.2% (Bias); [-11.86%, 5.47%] (95% LoA)
    QualityCheck - View Classification (Manual Images)Sensitivity: > 0.80 (implied, by meeting 0.85-1.00 range)Ranged from 0.85 to 1.00
    Specificity: > 0.80 (implied, by meeting 0.85-1.00 range)Ranged from 0.85 to 1.00
    Positive Predictive Value (PPV): > 0.80 (implied, by meeting 0.80-1.00 range)Ranged from 0.80 to 1.00
    Negative Predictive Value (NPV): > 0.90 (implied, by meeting 0.90-1.00 range)Ranged from 0.90 to 1.00
    QualityCheck - Structure Detection (Manual Images)Sensitivity: > 0.80 (implied, by meeting 0.81-1.00 range)Ranged from 0.81 to 1.00
    Specificity: > 0.99 (implied, by meeting 0.99-1.00 range)Ranged from 0.99 to 1.00
    Positive Predictive Value (PPV): > 0.89 (implied, by meeting 0.89-1.00 range)Ranged from 0.89 to 1.00
    Negative Predictive Value (NPV): > 0.99 (implied, by meeting 0.99-1.00 range)Ranged from 0.99 to 1.00
    QualityCheck - View Classification (LVA Images)Sensitivity: > 0.80 (implied, by meeting 0.86-1.00 range)Ranged from 0.86 to 1.00
    Specificity: > 0.80 (implied, by meeting 0.85-1.00 range)Ranged from 0.85 to 1.00
    Positive Predictive Value (PPV): > 0.80 (implied, by meeting 0.80-1.00 range)Ranged from 0.80 to 1.00
    Negative Predictive Value (NPV): > 0.91 (implied, by meeting 0.91-1.00 range)Ranged from 0.91 to 1.00
    QualityCheck - Structure Detection (LVA Images)Sensitivity: > 0.80 (implied, by meeting 0.82-1.00 range)Ranged from 0.82 to 1.00
    Specificity: = 1.00 (implied, by meeting 1.00-1.00 range)Ranged from 1.00 to 1.00
    Positive Predictive Value (PPV): > 0.89 (implied, by meeting 0.89-1.00 range)Ranged from 0.89 to 1.00
    Negative Predictive Value (NPV): > 0.99 (implied, by meeting 0.99-1.00 range)Ranged from 0.99 to 1.00
    PelvicAssist - Volume AlignmentAcceptance rate96.67%
    PelvicAssist - LH Measurement (ICC)Intraclass Correlation Coefficient (ICC) for six measurements.LH Area (0.9802), LH Circ. (0.9837), LH AP (0.9910), LH Lat. (0.9536), Right LUG (0.9423), Left LUG (0.9596)
    PelvicAssist - LH Measurement (Bland-Altman)Mean difference near zero; majority of data points within 95% LoA.Mean difference near zero (up to 1.1cm² and 0.53cm); 95.8% of data points fell within 95% LoA.
    EzVolume - Measurement TestBias (Mean Difference) for each label: Does not exceed ±2%.Did not exceed ±2%
    95% confidence interval for mean error includes zero.Included zero
    95% Limits of Agreement (LoA) for all labels: fell within ±15%.Fell within ±15%
    UterineAssist - Segmentation (Sagittal)Average Dice-score of uterus.96.7%
    UterineAssist - Segmentation (Transverse)Average Dice-score of uterus.95.8%
    UterineAssist - Segmentation (Endometrium)Average Dice-score of endometrium.86.8%
    UterineAssist - Feature Points Extraction (Uterus)Average error range of uterus feature points.1.5 – 2.6 mm
    UterineAssist - Feature Points Extraction (Endometrium)Average error range of endometrium feature points.0.9 - 1.7 mm
    UterineAssist - Size Measurement (Uterus)Average error range of Measurements.0.87 – 1.79 mm
    Widest 95% LoA range for uterus measurements; Largest Mean difference.[-2.96, 4.04] (widest 95% LoA); 1.23 mm (largest Mean difference)
    UterineAssist - Size Measurement (Endometrium)95% LoA range for endometrium measurements; Mean difference.[-1.59, 2.09] (95% LoA); 0.25 mm (mean difference)
    NerveTrack - DetectionLocalization accuracy success rate (95% CI); Processing speed.92.19% (95% CI: [90.03%, 94.34%]) (Success Rate); ~3.98 FPS

    2. Sample Sizes and Data Provenance

    AI/ML FeatureSample Size (Test Set)Data Provenance
    AbdomenAssist62 individual patients; 124 ultrasound images (62 kidney, 62 spleen)United States and Germany; Mix of retrospective and prospective
    BladderAssist63 individual patients; 63 ultrasound bladder transverse imagesUnited States and Germany; Mix of retrospective and prospective
    QualityCheck283 individual patients; 43,737 static 2D B-mode images (25,786 manual, 17,951 Live ViewAssist)United States; Mix of retrospective and prospective
    PelvicAssist40 individual patients; 120 volumes (40 rest, 40 contraction, 40 Valsalva)United States and Italy; Mix of retrospective and prospective
    EzVolume200 individual patients/3D volumes (100 1st trimester, 100 2nd/3rd trimesters)South Korea and United States; Mix of retrospective and prospective
    UterineAssist60 individual patients; 120 static images (60 sagittal, 60 transverse)South Korea and United States; Mix of retrospective and prospective
    NerveTrack46 individual patients; At least two nerve views per patient, with 2D sequences of at least 10 images. At least 24 and up to 42 ultrasound images for each of the 10 nerves.South Korea and United States; Mix of retrospective and prospective

    3. Number and Qualifications of Experts for Ground Truth (Test Set)

    AI/ML FeatureNumber of ExpertsQualifications of Experts
    AbdomenAssist3Two sonographers (one with >20 years exp., one with >10 years exp.); One senior expert radiologist (>20 years exp.)
    BladderAssist3Two sonographers (one with >20 years exp., one with >10 years exp.); One senior expert radiologist (>20 years exp.)
    QualityCheck3Two sonographers (each with >20 years exp.); One Obstetrician-Gynecologist (>10 years exp.)
    PelvicAssist3Three clinical experts (each with >20 years exp.)
    EzVolume3Three clinical experts (each with >10 years exp.)
    UterineAssist(At least) 2One sonographer (>10 years exp.) for view classification; Two sonographers (>10 years exp.) for manual drawing of anatomy areas/ground truth for validation images.
    NerveTrack3Two clinical experts (extensive experience in musculoskeletal ultrasound); One senior clinical expert (extensive experience in the field)

    4. Adjudication Method for the Test Set

    AI/ML FeatureAdjudication Method
    AbdomenAssist2+1 (2 independent measurements, 1 senior expert adjudication). The third senior expert reviewed and adjudicated the two measurements to determine the final value.
    BladderAssist2+1 (2 independent measurements, 1 senior expert adjudication). The third senior expert reviewed and adjudicated the two measurements to determine the final value.
    QualityCheckThe expert panel for the validation ground truth consisted of two sonographers, who performed the annotation, and an Obstetrician-Gynecologist, who provided the review and final confirmation. (Implies a 2+1 model, where the third expert reviews and confirms).
    PelvicAssistGTs were permuted and sent to the experts for peer review. Rejected data were re-labeled by the initial assigned expert, and the process is repeated. (This suggests an iterative consensus approach rather than a strict 2+1 or 3+1 structure initially, but aims for consensus among the 3 experts.)
    EzVolumeConsensus process of 3 experts. Initial annotation by one expert, then reviewed independently and blindly by the other two. If both accept, it's final. If any propose modifications, all three convene for unanimous agreement.
    UterineAssistFor images, two sonographers manually drew anatomy areas. (Implies agreement or an internal process, but not explicitly stated as 2+1 or 3+1. Ground truth was "made by sonographer" and then "manually drawn for each of the image by two sonographers" - implies dual annotation to establish GT.)
    NerveTrack2+1 (2 independent manual segmentations, 1 senior clinical expert adjudication). The senior expert resolved any discrepancies to establish the definitive ground truth.

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

    There is no mention of a Multi-Reader Multi-Case (MRMC) comparative effectiveness study comparing human readers with AI vs. without AI assistance for any of the features described. The studies focused on the standalone performance of the AI algorithms.

    6. Standalone (Algorithm Only) Performance

    Yes, standalone (algorithm only without human-in-the-loop performance) was done for all the AI/ML features described.

    • AbdomenAssist: Evaluated success rate and Bland-Altman agreement of the algorithm's measurements.
    • BladderAssist: Evaluated success rate and Bland-Altman agreement of the algorithm's measurements.
    • QualityCheck: Evaluated sensitivity, specificity, PPV, and NPV of the algorithm's view classification and structure detection.
    • PelvicAssist: Evaluated acceptance rate for volume alignment and ICC/Bland-Altman for LH measurement of the algorithm.
    • EzVolume: Evaluated error rate, bias, and LoA for the algorithm's measurements based on its segmentation results.
    • UterineAssist: Evaluated Dice-score for segmentation, average error range for feature point extraction, and error range/Bland-Altman for size measurements of the algorithm.
    • NerveTrack: Evaluated localization accuracy success rate and processing speed of the algorithm's detection.

    7. Type of Ground Truth Used

    AI/ML FeatureType of Ground Truth
    AbdomenAssistExpert consensus / manual measurement by clinical experts
    BladderAssistExpert consensus / manual measurement by clinical experts
    QualityCheckExpert consensus / classifications and annotations by clinical experts
    PelvicAssistExpert consensus / annotations by clinical experts
    EzVolumeExpert consensus / 3D segmentation annotation by clinical experts
    UterineAssistExpert consensus / manual segmentation and feature point annotation by sonographers
    NerveTrackExpert consensus / manual segmentation (ROI drawing) by clinical experts

    8. Sample Size for the Training Set

    The document explicitly states for each feature that "Data used for test and training/tuning purpose are completely separated from the ones during training process and there is no overlap between the two." or "Data used for training, tuning and validation purpose are completely separated from the ones during training process and there is no overlap between the three."

    However, the specific sample sizes for the training sets are not provided in this FDA 510(k) summary document.

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

    Similar to the training set size, the document does not explicitly describe how the ground truth for the training set was established. It only details the process for establishing the ground truth for the test/validation sets. The inference is that a similar expert-driven annotation process would have been used for training data, but the specifics are absent from this document.

    Ask a Question

    Ask a specific question about this device

    Page 1 of 1