FDA 510(k) Search - By Innolitics (SaMD and AI Experts)

The intended use of Voxel Dosimetry™ is to provide estimates (deterministic) of absorbed radiation dose at voxel as a result of administering one of the supported radionuclides and to provide a dose map. This is dependent on input data regarding bio distribution being supplied to the application.

Voxel Dosimetry™ only allows voxel-based dose calculations for patients who have been administered with radioisotopes.

Warning! The Voxel Dosimetry™ is only intended for calculating dose for FDA approved radiopharmaceuticals for any clinical purpose, and calculation of unapproved drugs can only be used for research purpose.

Device Description

Voxel Dosimetry™ is a tool for voxel level absorbed dose calculation resulting from radiotracer injection. Voxel Dosimetry™ workflow consists of the following steps:

SPECT/CT or PET/CT DICOM data loading from the data manager GOLD or PACS
Registration of different time-points to a common reference study
Generation and integration of voxel-level time-activity curves
Voxel-level absorbed dose calculation using a Monte Carlo-method
Saving of the absorbed dose-map back to GOLD or PACS in DICOM format

AI/ML Overview

Here's a breakdown of the acceptance criteria and the study proving the device meets them, based on the provided text:

Acceptance Criteria and Device Performance

Acceptance Criteria (Implicit)	Reported Device Performance
Cumulated Activity Accuracy
Difference in cumulated activity between Voxel Dosimetry™ and true cumulated activity (XCAT phantom with mono-exponential model).

For Ga68 (Kidney, Tumor, Spleen)
For I123 (Kidney, Tumor, Spleen)
For I131 (Kidney, Tumor, Spleen)
For In111 (Kidney, Tumor, Spleen)
For Lu177 (Kidney, Tumor, Spleen)
For Tc99m (Kidney, Tumor, Spleen)
For Y90 (Kidney, Tumor, Spleen) | - Ga68: 6%, 6%, 7%
I123: 3%, 1%, 2%
I131: 7%, 2%, 3%
In111: 11%, 7%, 7%
Lu177: 7%, 3%, 3%
Tc99m: 8%, 7%, 6%
Y90: 12%, 8%, 8% |
| Dose Calculation Accuracy | |
| Difference in Voxel Dosimetry™ dose compared to PENELOPE dose.
For I123 (Kidney, Tumor, Spleen)
For I131 (Kidney, Tumor, Spleen)
For Ga68 (Kidney, Tumor, Spleen)
For In111 (Kidney, Tumor, Spleen)
For Lu177 (Kidney, Tumor, Spleen)
For Tc99m (Kidney, Tumor, Spleen)
For Y90 (Kidney, Tumor, Spleen) | - I123: 2%, 3%, 3%
I131: 3%, 3%, 3%
Ga68: 12%, 12%, 12%
In111: 2%, 2%, 3%
Lu177: 1%, 1%, 1%
Tc99m: 2%, 3%, 3%
Y90: 5%, 6%, 4% |
| Correlation with Predicate Device (OLINDA/EXM® v2.0)
Pearson's r for left kidney doses
Pearson's r for right kidney doses | - r_left = 0.97
r_right = 0.98 |
| Relative Difference from Predicate Device (OLINDA/EXM® v2.0)
Average relative difference in kidney doses | - -2% |
| Safety and Effectiveness | The stated differences in cumulated activities and doses in phantom studies are considered small, with the exception of Ga68, which has high positron energy. The differences between SMC and OLINDA/EXM® v2.0 in kidney dosimetry (2%) are less than the known uncertainty in Lu177 kidney dosimetry, indicating no impact on safety or effectiveness. |
| Compliance with Software Specifications | "The testing results support that all the software specifications have met the acceptance criteria." |

Study Details

Sample size used for the test set and the data provenance:
- Phantom Testing:
  - The exact "sample size" in terms of number of different XCAT phantoms generated is not explicitly stated. However, it involved generating an XCAT phantom for each isotope tested (Ga68, I123, I131, In111, Lu177, Tc99m, Y90), with four time points for each isotope.
  - Provenance: Synthetic/simulated data (XCAT phantom).
- Clinical Data Comparison:
  - Patient Sample Size: Six patients, twelve treatment cycles.
  - Provenance: This appears to be retrospective clinical data, as patients underwent Lu177-DOTATE treatments and were scanned at specific time points. The publication (Hippeläinen et al., 2017) suggests it was a real-world study. The specific country of origin is not mentioned.
Number of experts used to establish the ground truth for the test set and the qualifications of those experts:
- Phantom Testing: Ground truth was established by analytical or established Monte Carlo methods (PENELOPE, mono-exponential model). No human experts were directly involved in establishing this ground truth.
- Clinical Data Comparison: The comparison was against the predicate device OLINDA/EXM® v2.0, which itself is a calculation tool. While presumably experts would have performed the initial OLINDA/EXM® calculations, the text doesn't specify experts for this comparison's ground truth beyond the output of the predicate.
Adjudication method for the test set:
- No adjudication method (like 2+1 or 3+1) is mentioned, as the ground truth for both test sets (phantom and clinical comparison) was established via computational models or comparison with another software, not human consensus.
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 or human-in-the-loop study with human readers/AI assistance was conducted or reported. This device is a dose calculation software, not an AI-assisted diagnostic tool for human interpretation.
If a standalone (i.e. algorithm only without human-in-the-loop performance) was done:
- Yes, the performance evaluation was entirely a standalone assessment of the algorithm. Its calculations were compared against analytical results (phantom studies) or another standalone algorithm (OLINDA/EXM® v2.0).
The type of ground truth used (expert consensus, pathology, outcomes data, etc.):
- Phantom Testing:
  - For cumulated activity: Analytical integration of a mono-exponential model.
  - For dose calculations: Monte Carlo simulation results from PENELOPE code.
- Clinical Data Comparison: Reference standard was the output of the legally marketed predicate device, OLINDA/EXM® v2.0.
The sample size for the training set:
- The document does not explicitly mention a "training set" in the context of machine learning model development. This device appears to be based on a Semi-Monte Carlo (SMC) method for dose calculation, which is a physics-based model rather than a data-driven machine learning model requiring a specific training set. Therefore, this question is not directly applicable in the typical sense. The underlying physics models and S-values might be "trained" or derived from theoretical physics and extensive pre-computed data, but not in the sense of a deep learning model.
How the ground truth for the training set was established:
- As explained above, there's no mention of a traditional machine learning "training set" or its ground truth establishment in the provided text. The SMC method is a computational technique based on physical principles, not a model learned from labeled data in the usual machine learning context.

Summary

Regulation Number and Section

§ 892.2050 Medical image management and processing system.

(a)
Identification. A medical image management and processing system is a device that provides one or more capabilities relating to the review and digital processing of medical images for the purposes of interpretation by a trained practitioner of disease detection, diagnosis, or patient management. The software components may provide advanced or complex image processing functions for image manipulation, enhancement, or quantification that are intended for use in the interpretation and analysis of medical images. Advanced image manipulation functions may include image segmentation, multimodality image registration, or 3D visualization. Complex quantitative functions may include semi-automated measurements or time-series measurements.(b)
Classification. Class II (special controls; voluntary standards—Digital Imaging and Communications in Medicine (DICOM) Std., Joint Photographic Experts Group (JPEG) Std., Society of Motion Picture and Television Engineers (SMPTE) Test Pattern).