Search Results

MEDIHUB PROSTATE is an image processing software package that performs outlining, processing, viewing, and editing of prostate MR images. The software can support study review and analysis of prostate MR data with computed modules. The analysis result can be presented in various formats, including images overlaid onto source MR images and a structured report.

MEDIHUB PROSTATE semi-automatically outlines the prostate based on MR images by contour, and it requires the user to edit with image manipulating tools and confirm the final result. The package provides additional functionalities including registered multiparametric-MRI viewing and combining MR sequences into a single image to support visualization. Edited PI-RADS report and semi annotated prostate region can be viewed in each single image in the final report.

MEDIHUB PROSTATE is intended to be used by trained radiologists and urologists. Patient management decisions should not be made solely based on the analysis performed by MEDIHUB PROSTATE.

Limitations:

• . MEDIHUB PROSTATE has been validated for use with Siemens 3T Vida and Skyra MRI machines.
• . MEDIHUB PROSTATE is also designed for use with the Siemens 3T T2 MRI series, supporting slice thicknesses ranging from 3.5 mm to 5 mm.
• . MEDIHUB PROSTATE has been tested on patients aged 55 years and above.

MEDIHUB PROSTATE is an image processing software package for multi-parametric prostate MR image analysis. The analysis may assist trained radiologists in clinical interpretation of prostate MR studies. It can be accessed through a web browser, and provides the following main features:

• . A semi-automatic processing module that outlines the prostate region and performs multiparametric MRI image registration.
• . A user-interaction module in which the user can edit and approve the computed prostate outline and determine PSA density using serum PSA level.
• . A user-interaction module in which the user can view multi-parametric MRI images, and outline and analyze ROIs. This extension will also apply a mathematical operation on the input images to combine information from another MRI sequences into a single combination image.
• A semi-automatic processing module that collects all results for exporting and transferring back to the user.
  All measurements are manual except for the prostate volume, which is semi-automatic and requires user review. The method for measuring prostate volume is straightforward and unaffected by patient demographics. Users have the option to outline the prostate either completely manually or with semi-automatic assistance. This device is not intended to be used for fully automatic prostate delineation. It does not involve any segmentation functions by itself. The Al functionality is limited to assessing the total prostate volume, without segmenting lesions.

In semi-automatic mode, our device employs an Al-based algorithm to initially outline the prostate volume, and then it requires the user to edit, review and approve. Additionally, the device calculates the total prostate volume. However, users are responsible for performing all other image annotations and measurements manually. This implies that the final decision should be confirmed by the user, and the user should not rely solely on the device's analysis.

Additional annotations and measurements: all calculated manually

• PI-RADS
• Location of seminal vesicles
• . Prostate zones
• . DWI and DCE graphs

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

I. Acceptance Criteria and Reported Device Performance

The primary performance metric for the prostate region segmentation algorithm is the Dice coefficient, and the secondary metric is the Hausdorff Distance.

II. Sample Sizes and Data Provenance

• Training Dataset:

  • Korea: 748 cases
  • US (University of Missouri Health Care): 709 cases
  • Total Training: 1457 cases
  • Data Provenance: Retrospective, collected from Korea and the US (University of Missouri Health Care).

• Validation Dataset:

  • Korea: 80 cases
  • US (University of Missouri Health Care): 136 cases
  • Total Validation: 216 cases
  • Data Provenance: Retrospective, collected from Korea and the US (University of Missouri Health Care).

• Clinical Test Dataset (Standalone Performance):

  • Sample Size: 114 T2 MR images
  • Data Provenance: Retrospective, collected from the US (University of Missouri Health Care).

• Clinical Test Dataset (Reader Performance):

  • Sample Size: 73 cases (a subset of the 114 cases used for standalone performance)
  • Data Provenance: Retrospective, collected from the US (University of Missouri Health Care).

III. Number and Qualifications of Experts for Test Set Ground Truth

• Number of Experts: Three expert radiologists.
• Qualifications of Experts: The document states they were "expert-level radiologists" without further specifying their years of experience or board certifications.

IV. Adjudication Method for Test Set Ground Truth

• Adjudication Method: Majority rule approach. In cases of ties, annotations were consolidated through discussion and mutual agreement among the three radiologists. This is a 2+1 (if two agree, that becomes the truth; if all three disagree, they discuss to reach consensus) or potentially a 3-way consensus if agreement is required.

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

• Was an MRMC study done? Yes, a "Clinical Testing (Reader Performance)" study was conducted, which is a type of MRMC study where human reader performance with and without AI assistance is compared.
• Effect Size of Improvement: The improvements in the Dice coefficient for prostate outlining performance for the three radiologists when using the prostate region segmentation algorithm of MEDIHUB PROSTATE were:
  • Radiologist 1: 0.156
  • Radiologist 2: 0.011
  • Radiologist 3: 0.008
    These values represent the change in their individual Dice coefficients when assisted by the AI.

VI. Standalone Performance (Algorithm Only)

• Was a standalone study done? Yes, a "Clinical Testing (Stand-alone Performance)" study was conducted.
• Performance Metrics: The mean Dice coefficient was 0.928 (95% CI: [0.925, 0.931]) and the Hausdorff distance was 2.171 (95% CI: [1.097, 3.245]).

VII. Type of Ground Truth Used

• Type of Ground Truth: Expert consensus. The ground truth for the test datasets was established by three expert radiologists through independent annotation followed by a majority rule approach with discussion for ties.

VIII. Sample Size for Training Set

• As detailed in Section II, the total training sample size was 1457 cases (748 from Korea, 709 from US).

IX. How the Ground Truth for the Training Set Was Established

The document states: "Study Population Dataset: The Study Population Dataset is the same as algorithm development dataset." It then describes how the "ground truth produced by three expert radiologists" was used for the "performance check test" on the segmentation algorithm, implying a similar ground truthing process for the development/training data as was used for the test set. Therefore, it is implied that the ground truth for the training set was also established by expert consensus of radiologists, similar to the test set, though the specifics of the number of radiologists and adjudication method for the training set itself are not separately detailed from the general "ground truth produced by three expert radiologists."

Ask a specific question about this device