Search Results

3D-RD-S is intended to estimate radiation absorbed dose (and related quantities) to tissues after administration of a radioactive product. For use with internally administrated radioactive products, 3D-RD-S should not be used to deviate from product dosing and administrations. Refer to the product's prescribing information for instructions.

3D-RD-S is a cloud-based software as a medical device (SaMD) that interacts with the user via web browsers (for example Google Chrome). Users are trained healthcare professionals with significant dosimetry knowledge and experience and also responsible for the input of the appropriate values and to make correct interpretation of the output data. 3D-RD-S takes numerical input data in the form of activity in source tissues as a function of time (TAC data) or the integral of the activity (TIA data) in source tissues over time. It then calculates the absorbed dose to a set of target tissues based on the organ sizes and anatomies of a set of standard phantoms. The software provides the user the ability to account for the differences in tissue masses between the phantoms and the subject and model uncertainties in the input data.

Calculation results can be viewed and updated by other users. The software provides the ability to calculate absorbed doses and related radiobiological quantities from input data. The calculations can be made for supported radionuclides based on data in the report 89 from the International Council on Radiation Protection (ICRP). Doses to target tissues are a function of the activity integrated over time (time-integrated activity. TIA) in a set of specified source organs. The software provides two modules for the integration of input time vs. activity curve (TAC) data. First, the user can use curve fitting methods to estimate a curve that passes through the TAC data from a set of supported fitting functions. Visual and numerical indicators of how well the fitting function works with the data are provided. Notifications are given if fitting parameters are non-physical. The TAC data can then be integrated using the fitting function, or by approximating the activity between measured time points with line and assuming activity after the last time-point decays with the radionuclide's physical half-life. If desired, the user can use a combination of the curve fit, linear interpolation between the lines, and exponentially decaying extrapolation based on the physical half-life, to integrate the time-activity curves.

The calculated radiobiological quantities purport to relate physical dose to biological response and are dependent on the specification of radiobiological constants. The guantities supported include the whole-body effective dose and the relative biological effectiveness (RBE) weighted dose. The effective dose is calculated based on ICRP tissue weighting factors. The RBE weighted dose is calculated using user specified RBEs for the different radiation types (standard values are provided as defaults).

3D-RD-S provides total and individual dose estimates for the various particle types, i.e., alpha particles, beta (+ and -) particles, discrete electrons (e.g., Auger electrons), and photons (gamma and x-rays). The resulting doses are plotted in a bar graph and can, along with input data, be exported in a spreadsheet.

The provided document, a 510(k) Summary for the 3D-RD-S device, details the acceptance criteria and the studies conducted to demonstrate its performance.

Here's an analysis of the provided information:

1. Table of Acceptance Criteria and Reported Device Performance:

The document describes several benchmark tests, each with an implicit or explicit acceptance criterion and the corresponding performance.

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

• Benchmark Test (1) (Comparison with Predicate):

  • Sample Size: Not explicitly stated as a number of cases or patients. It mentions using "Clinical data, obtained from Rapid's clinical trials dosimetry service business with clinically relevant administered activities."
  • Data Provenance: Implied to be retrospective clinical data from "Rapid's clinical trials dosimetry service business." The country of origin is not specified but given the FDA submission, it's likely US-based or from regions with compatible data standards.
  • Radionuclides tested: In-111, F-18, Ga-68 (photon emitters); I-131, Lu-177 (beta emitters); Pb-212, Ra-223, Ac-225 (alpha emitters' parents).

• Benchmark Test (2) (Comparison with Published Literature):

  • Sample Size: Not explicitly stated as a number of cases or patients. It refers to "several published studies" that met specific criteria.
  • Data Provenance: Retrospective, derived from "published studies" that used specific phantom models (ICRP 110) and SAF values (ICRP 133).
  • Radionuclides tested: F-18, Zr-89, Y-90, I-131, Lu-177, At-211.

• Benchmark Test (3) (Inter-operator Variability):

  • Sample Size: Data from "two patients (A and B)" from the SNMMI Lu-177 Dosimetry Challenge.
  • Data Provenance: Retrospective, from a "SNMMI Lu-177 Dosimetry Challenge (Uribe et al. J Nuc Med. 2021)." This implies a standardized or widely recognized dataset.

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

• Benchmark Test (1) & (2): No explicit mention of human experts defining ground truth for these tests. The ground truth for Test (1) is the output of the predicate device (OLINDA/EXM v2.0), and for Test (2), it's the results published in peer-reviewed literature. These are treated as reference standards for comparison.
• Benchmark Test (3): "Two (2) analysts" performed the processing. Their qualifications are stated as "background in medical physics and extensive experience in radiopharmaceutical therapy dosimetry." This test assessed inter-operator variability rather than establishing a new ground truth.

4. Adjudication Method for the Test Set:

• No explicit adjudication method (e.g., 2+1, 3+1) is described for establishing ground truth or resolving discrepancies for any of the tests.
• For Benchmark Test (1), the ground truth is the output of the predicate device.
• For Benchmark Test (2), the ground truth is data from published literature.
• For Benchmark Test (3), the test itself is designed to measure variability between two independent operators, so there isn't a need for adjudication to establish a "single" ground truth from their different outputs; rather, the agreement between them is the metric of interest.

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:

• No MRMC comparative effectiveness study involving human readers improving with AI vs. without AI assistance was reported.
• The device, 3D-RD-S, is a "cloud-based software as a medical device (SaMD)" for absorbed dose calculation. It is not an AI-assisted diagnostic tool for human readers in the traditional sense of image interpretation. It takes numerical input data (TAC or TIA) and calculates absorbed doses.
• Benchmark Test (3) involves multiple operators, but it assesses inter-operator variability in processing input data for the software, not how the software assists in human interpretation or decision-making.

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

• Yes, implicitly. Benchmark Test (1) and Benchmark Test (2) represent the "algorithm only" performance.
  • In Benchmark Test (1), the 3D-RD-S's calculated absorbed doses are compared directly against the predicate device's output, simulating a standalone performance against a benchmark.
  • In Benchmark Test (2), 3D-RD-S's outputs are compared against published literature, again a direct algorithmic comparison.
• Benchmark Test (3) involves human operators as input providers, but the focus is on the variability of the input provided by humans resulting in differences in the software's output, rather than the software changing human decision making. The software itself performs its calculations in a standalone manner based on the input it receives.

7. The Type of Ground Truth Used:

• Benchmark Test (1): The "ground truth" or reference was the output data from a legally marketed device, OLINDA/EXM v2.0 (a type of comparative reference standard).
• Benchmark Test (2): The "ground truth" or reference was absorbed dose values reported in published scientific literature, specifically those using ICRP 110 phantoms, ICRP 133 SAF values, and ICRP 107 nuclear decay data (literature-based reference standard/computational comparison).
• Benchmark Test (3): This test did not establish a "ground truth" in the traditional sense of a definitive diagnosis or outcome. Instead, it examined the consistency/variability of the device's output when processed by different medical physics experts, using SNMMI Dosimetry Challenge data. The "ground truth" for the input process was likely the raw image data from known patient cases, and the test measured the agreement of the derived dose outputs.

8. The Sample Size for the Training Set:

• The document does not provide information about the sample size used for the training set.
• Given that 3D-RD-S sounds like a deterministic calculation software based on established physical models (ICRP data, MIRD dose calculations, curve fitting for TAC), it's possible it's not a machine learning/AI model that undergoes "training" on a dataset in the conventional sense. Instead, it might be validated against known physical principles and benchmarked against existing validated tools and literature. The "training" in such a system often refers to the development of the underlying physical models and constants rather than statistical learning from a dataset.

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

• Since there's no mention of a training set or typical machine learning "training" process, there's no information on how its "ground truth" was established.
• The core functioning of 3D-RD-S relies on established scientific data and models, such as:
  • ICRP Report 89 for supported radionuclides.
  • ICRP 110 phantoms and ICRP 133 SAF values.
  • ICRP 107 for nuclear decay data.
  • Standard curve fitting methods and trapezoidal integration for TAC data.
    These scientific principles and reference values form the inherent "ground truth" upon which the software's calculations are built.

Ask a specific question about this device