Search Results

CephX3D is a software program for the analysis of dental and craniomaxillofacial image information and can be used to provide design input for dental solutions. It processes and displays digital images from various sources to support the diagnostic process and treatment planning. It is intended for use in patients aged 12 and older.

CEPHX3D is a cloud-based software-as-a-service (SaaS) solution designed for the automated processing and visualization of dental and maxillofacial Cone Beam Computed Tomography (CBCT) data. The software utilizes deep learning algorithms to perform three-dimensional segmentation of key anatomical structures, including skeletal bone (mandible and maxilla), the full dentition, and the inferior alveolar nerve (IAN).

Here's a detailed breakdown of the acceptance criteria and study information for the CEPHX3D device, extracted from the provided FDA 510(k) clearance letter:

Acceptance Criteria and Device Performance

Study Details

This information is extracted from "Section 9. Performance Data" of the 510(k) summary.

1. Sample sizes used for the test set and the data provenance:

   • Quantitative Non-Clinical Performance Testing: 30 CBCT datasets.
   • Qualitative Clinical Performance Testing: 53 CBCT cases.
   • Data Provenance: The document does not explicitly state the country of origin or if the data was retrospective or prospective. It only mentions "CBCT datasets" and "CBCT cases."

2. Number of experts used to establish the ground truth for the test set and the qualifications of those experts:

   • Number of Experts: Two independent experts were used for validating the ground truth (Inter-Rater Reliability study).
   • Qualifications of Experts (for Qualitative Clinical Performance Testing): A panel of independent, board-certified specialists. Their specific specialties (e.g., orthodontists, oral surgeons) are not listed, nor is their years of experience.
   • Qualifications of Experts (for Ground Truth Validation): Referred to as "two independent experts," but specific qualifications are not detailed beyond "expert-refined."

3. Adjudication method for the test set:

   • Quantitative Non-Clinical Performance Testing: The ground truth was "expert-refined" via the "Refined Automated Segmentation (R-AS) methodology."
   • Ground Truth Validation: "Per the study protocol, no formal adjudication was required as all expert-to-expert spatial discrepancies remained significantly below the 0.5 mm escalation trigger." This implies a form of consensus or agreement without needing a third adjudicator.
   • Qualitative Clinical Performance Testing: The document mentions "Inter-rater reliability analysis," indicating multiple specialists reviewed cases, but it does not specify a formal adjudication method (e.g., 2+1, 3+1) if initial agreement was not reached. The high Fleiss' Kappa scores suggest strong agreement, potentially rendering extensive adjudication unnecessary.

4. 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 was explicitly described in the provided text. The performance data focuses on the standalone algorithm's accuracy against ground truth and the clinical acceptability of its output by human readers, not on how human readers improve with AI assistance compared to without it.

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

   • Yes, a standalone performance evaluation was done. The "Quantitative Non-Clinical Performance Testing" directly assesses the algorithm's performance (Dice Similarity Coefficient, RMS error, Hausdorff Distance) against an expert-refined ground truth, demonstrating the algorithm's capability independent of human intervention in the segmentation process itself. The "Automated Segmentation" functionality described in Section 5.2 confirms this.

6. The type of ground truth used (expert consensus, pathology, outcomes data, etc.):

   • Expert Consensus/Refinement: For the quantitative non-clinical testing, the ground truth was "expert-refined 'Ground Truth' established via the Refined Automated Segmentation (R-AS) methodology." This indicates an initial automated segmentation refined or validated by experts to serve as the gold standard.
   • Expert Judgment (for Clinical Acceptability): For the qualitative clinical performance testing, the primary endpoint was "Clinically Acceptable" ratings by a "panel of independent, board-certified specialists." This is a form of ground truth established through expert clinical judgment.

7. The sample size for the training set:

   • The document does not provide the sample size for the training set used to develop the deep learning algorithms. It only describes the test sets.

8. How the ground truth for the training set was established:

   • The document does not provide information on how the ground truth for the training set was established. It describes the ground truth for the test set as "expert-refined 'Ground Truth' established via the Refined Automated Segmentation (R-AS) methodology." While it's implied similar methods might be used for training data, this is not explicitly stated.

Ask a specific question about this device