Search Results

Precision DL is a deep learning-based image processing method intended to enhance image quality of non-ToF PET images for clinical oncology purpose, using F-18 FDG. Precision DL may be used for patients of all ages.

Precision DL is a deep learning-based image processing method intended for PET oncology 18F-FDG images obtained using the predicate device Omni Legend PET/CT system. Precision DL enhances the non-ToF Q.Clear images to have image quality performance similar to PET images obtained using ToF capable PET systems, including enhancement in image Contrast Recovery (CR), Contrast to Noise Ratio (CNR), and quantitation accuracy. Precision DL's training used clinical data from diverse clinical sites, accounting for relevant variations in patients and sites' protocols.

Precision DL brings three deep learning models to provide users the choice between different strengths of contrast enhancement and noise reduction. The three models, Low, Medium, and High Precision DL, are trained such that the High Precision DL brings the highest contrast enhancement and lowest noise reduction, while the Low Precision DL brings the lowest contrast enhancement and highest noise reduction. Medium Precision DL brings contrast-noise tradeoff in between High and Low Precision DL.

Precision DL is deployed within the acquisition and processing software of Omni Legend, for processing images produced using non-ToF Q.Clear image reconstruction.

Here's an analysis of the acceptance criteria and study for Precision DL, based on the provided FDA 510(k) summary:

Device: Precision DL (Deep Learning-based image processing for non-ToF PET images)

Intended Use: Enhance image quality of non-ToF PET images for clinical oncology using F-18 FDG.

1. Table of Acceptance Criteria and Reported Device Performance

The 510(k) summary describes performance improvements over non-ToF Q.Clear reconstruction. It does not explicitly state discrete acceptance criteria values but rather demonstrates general improvements in imaging metrics and equivalence to ToF images.

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

• Sample Size (Clinical Data for Bench Testing): 80 PET-CT exams.
  • 40 exams from an Omni Legend system.
  • 40 exams from Discovery MI systems (with hardware-based ToF).
• Sample Size (Clinical Reader Study): Not explicitly stated precisely for the number of cases and images. It mentions "clinical cases of the same patients obtained on Discovery MI and Omni Legend with Precision DL."
• Data Provenance: Multiple clinical sites in North America, Europe, and Israel. The data was "segregated, completely independent, and not used in any stage of the algorithm development, including training." This indicates prospective or retrospectively collected data used for testing only.

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

• Number of Experts (for Ground Truth related to phantom data): Not applicable for phantom data, as ground truth is known from inserted lesions.
• Number of Experts (for Clinical Reader Study): "Board certified radiologists." The exact number is not explicitly stated in the summary, nor are their specific years of experience. However, the study involved reviews and preference questions by these experts.

4. Adjudication Method

• The summary mentions a "clinical reader study" where "board certified radiologists... answered blinded preference questions comparing clinical cases." This suggests individual reader assessments were aggregated, but it does not explicitly state an adjudication method like 2+1 or 3+1 for resolving discrepancies in diagnostic findings. The focus appears to be on overall image quality and preference rather than a specific diagnostic consensus for each case.

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

• Yes, a clinical reader study was performed, which included "board certified radiologists" reviewing "clinical cases." They answered "blinded preference questions comparing clinical cases of the same patients obtained on Discovery MI and Omni Legend with Precision DL."
• Effect Size of Human Readers with AI vs. Without AI Assistance: The summary states, "The results of the reader study and preference questions support the determination of substantial equivalence. All readers attested that their assessments of Precision DL demonstrated acceptable diagnostic results." While it indicates positive results and physician acceptance, it does not quantify an effect size of how much human readers improved their performance (e.g., in diagnostic accuracy, confidence, or reduced read time) with AI assistance compared to reading without AI assistance (i.e., using only non-ToF Q.Clear images). The study primarily focused on image quality acceptability and preference.

6. Standalone (Algorithm Only Without Human-in-the-Loop Performance) Study

• Yes, a standalone performance assessment was conducted as part of the "additional engineering bench testing." This included quantitative assessments using both clinical and phantom data for metrics such as Quantitation Accuracy, Contrast Recovery, Contrast-to-Noise Ratio, Lesion Detectability, and Dose/Time Reduction. This part of the testing directly evaluated the algorithm's output (processed images) against established ground truths/references.

7. Type of Ground Truth Used

• For Bench Testing (Quantitative Metrics):
  • Phantom Data: Known quantitation from inserted lesions of known size, location, and contrast.
  • Clinical Data: Discovery MI's ToF PET images served as a reference for comparison, implying they are considered the gold standard for high-quality images that Precision DL aims to emulate.
• For Clinical Reader Study: The "ground truth" here is implied to be the expert consensus on acceptable diagnostic quality and preference rather than a definitive diagnosis based on pathology or long-term outcomes for each case. The goal was to confirm that the enhanced images retained or improved diagnostic acceptability.

8. Sample Size for the Training Set

• The summary states, "Precision DL's training used clinical data from diverse clinical sites, accounting for relevant variations in patients and sites' protocols."
• However, the specific sample size for the training set is not provided in the given text.

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

• The summary indicates that Precision DL "is trained to enhance non-ToF images to have IQ performance similar to ToF images." This implies that the ground truth for training would likely be high-quality ToF PET images (potentially from a system like Discovery MI) that the algorithm was designed to mimic or achieve certain quality metrics aligned with ToF performance.
• The text doesn't detail the process of establishing ground truth for individual images within the training set, but it's reasonable to infer a reference standard from ToF images was used.

Ask a specific question about this device