Search Results

Deep Recon is a data driven image reconstruction method based on deep learning technology. It is intended to produce cross-sectional images by computer reconstruction of X-ray transmission data taken at different angles planes, including Axial, Helical, and Cardiac acquisition.

Deep Recon is designed to generate CT images with lower image noise, and improved low contrast detectability, and it can reduce the dose required for diagnostic CT imaging.

Deep Recon can be used for head, chest, abdomen, cardiac and vascular CT applications for adults. Deep Recon is intended to be used with uCT 760 and uCT 780 only.

The Deep Recon is a data driven image reconstruction method based on deep learning technology. Dedicated deep neural networks are designed and trained for different body parts. As a part of reconstruction chain, the Deep Recon generates CT images with an appearance similar to traditional FBP, but with a decreased image noise, and an improved low contrast detectability. The Deep Recon was specifically trained on uCT 760 and uCT 780 (K172135). The function is integrated on the mentioned CT systems as a part of reconstruction chain.

The initial document provides a 510(k) summary for the Deep Recon device, a data-driven image reconstruction method based on deep learning technology for CT systems. The device is intended to produce cross-sectional images with lower image noise, improved low contrast detectability, and the ability to reduce the required dose for diagnostic CT imaging.

Here's an analysis of the acceptance criteria and study information provided:

1. Table of Acceptance Criteria and Reported Device Performance

The document does not explicitly state numerical acceptance criteria in a dedicated table. Instead, it describes performance goals as "equivalent or better performance" compared to Filtered Back Projection (FBP) for various image quality metrics, and "equivalent or better" diagnostic quality in clinical evaluations.

2. Sample Size and Data Provenance for Test Set

• Clinical Image Evaluation (Reader Study 1):
  • Sample Size: 80 retrospectively collected clinical cases.
  • Data Provenance: Retrospective, country of origin not specified, but the device manufacturer is based in Shanghai, China.
• Clinical Image Evaluation (Reader Study 2):
  • Sample Size: 40 retrospectively collected clinical cases (20 low dose, 20 standard dose).
  • Data Provenance: Retrospective, country of origin not specified, but the device manufacturer is based in Shanghai, China.

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

• Clinical Image Evaluation (Reader Study 1):
  • Number of Experts: 2 board-certified radiologists.
  • Qualifications: "board-certified radiologists." No further details on years of experience are provided.
• Clinical Image Evaluation (Reader Study 2):
  • Number of Experts: "a board-certified radiologist." This implies only one radiologist was used.
  • Qualifications: "board-certified radiologist." No further details on years of experience are provided.

4. Adjudication Method for the Test Set

• The document describes that for Reader Study 1, "Each image was read by 2 board-certified radiologists." It does not specify an adjudication method like 2+1 or 3+1 for discrepancies.
• For Reader Study 2, "Each image was read by a board-certified radiologist," indicating no inter-reader adjudication was performed.

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

• No explicit MRMC comparative effectiveness study is mentioned that quantifies the "effect size of how much human readers improve with AI vs without AI assistance." The studies performed are comparative evaluations of the image quality and diagnostic usefulness of Deep Recon images versus FBP images, as assessed by human readers. They do not describe a scenario where AI assists human readers and measures the improvement.

6. Standalone Performance

• No explicit standalone performance study (algorithm only without human-in-the-loop performance) is described in terms of diagnostic accuracy. The document focuses on the output of the algorithm (the reconstructed images) and how those images are perceived by human readers. The phantom studies (bench testing) could be considered standalone in the sense that they evaluate the algorithm's direct image quality metrics, but not diagnostic performance.

7. Type of Ground Truth Used

• Clinical Image Evaluation Studies: The "ground truth" for the clinical evaluations (Reader Studies 1 & 2) was the expert consensus of the radiologists regarding image noise, structure fidelity, image quality, and clinical features based on a 4-point or 5-point scale. It does not refer to pathology, patient outcomes data, or an independent gold standard for diagnosis.
• Bench Testing: The ground truth for bench testing (LCD, image noise, CT number, uniformity, spatial resolution, section thickness) involved standard phantom measurements and model observers, which represent established physical metrics rather than clinical ground truth derived from patients.

8. Sample Size for the Training Set

• The document states that the Deep Recon's "Dedicated deep neural networks are designed and trained for different body parts." It also mentions that the "Deep Recon was specifically trained on uCT 760 and uCT 780 (K172135)."
• However, the specific sample size (number of images or cases) used for the training set is not provided in this document.

9. How Ground Truth for the Training Set Was Established

• The document mentions that the DNN is "trained on low dose FBP images to get normal dose (high quality) FBP images." This implies that the training likely used pairs of low-dose FBP images (as input to the network) and corresponding "normal dose (high quality) FBP images" (as the target or ground truth for the network to learn from).
• The methodology for establishing the "normal dose (high quality) FBP images" as ground truth for training is not explicitly detailed. It's implicitly assumed that these are standard-of-care FBP reconstructions from standard-dose acquisitions.

Ask a specific question about this device