Search Results

QP-Prostate® CAD is a Computed Aided Detection and Diagnosis (CADe/CADx) image processing software that automatically detects and identifies suspected lesions in the prostate gland based on bi-parametric prostate MRI. The software is intended to be used as a concurrent read by physicians with proper training in a clinical setting as an aid for interpreting prostate MRI studies. The results can be displayed in a variety of DICOM outputs, including identified suspected lesions marked as an overlay onto source MR images. The output can be displayed on third-party DICOM workstations and Picture Archive and Communication Systems (PACS). Patient management decisions should not be based solely on the results of QP-Prostate® CAD.

QP-Prostate® CAD is an artificial intelligence-based Computed Aided Detection and Diagnosis (CADe/CADx) image processing software. QP-Prostate® CAD uses Al-based algorithms trained with pathology data to detect suspicious lesions for clinically significant prostate cancer. The device automatically detects and identifies suspected lesions in the prostate gland based on bi-parametric prostate MRI and provides marks over regions of the images that may contain suspected lesions. There are two possible markers that are provided in different colors suggesting different levels of suspicion of clinically significant prostate cancer (moderate or high suspicion level).

The software is intended to be used as a concurrent read by physicians with proper training in a clinical setting as an aid for interpreting prostate MRI studies. The results can be displayed in a variety of DICOM outputs, including identified suspected lesions marked as an overlay onto source MR images. The output can be displayed on third-party DICOM workstations and Picture Archive and Communication Systems (PACS). Based on biparametric input consisting of T2W and DWI series, the analysis is run automatically, and the output in standard DICOM formats is returned to PACS.

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

Acceptance Criteria and Reported Device Performance

Table 1: Acceptance Criteria and Reported Device Performance (Standalone)

Note: The document does not explicitly state numerical acceptance criteria thresholds for the standalone performance metrics (AUC-ROC, Sensitivity, FPR). Instead, it presents the results and implies that these values "demonstrate the safety and effectiveness" in comparison to the predicate device. The general implicit acceptance criterion for these metrics would be that they exhibit performance levels indicative of a useful diagnostic aid.

Table 2: Acceptance Criteria and Reported Device Performance (Multi-Reader Multi-Case Study)

The document stated directly that "The test results demonstrate that QP-Prostate® CAD functioned as intended and met its primary endpoint, is acceptable for clinical use, and is as safe and effective as its predicate device, without introducing new questions of safety and efficacy."

Study Details for QP-Prostate® CAD performance:

1. Sample Size and Data Provenance for Test Set:

   • Sample Size: 228 cases for the clinical reader performance assessment (MRMC study) and 247 for the standalone performance assessment (lesion-level).
   • Data Provenance: Retrospectively collected from multiple centers across the US. This dataset was "completely independent from the training dataset that was acquired from different institutions."

2. Number of Experts and Qualifications for Ground Truth (Test Set):

   • The document states that the ground truth for the standalone performance evaluation was derived from "associated pathology reports and radiologist interpretations." It does not specify the number or qualifications of radiologists involved in these interpretations for the ground truth establishment of the test set.
   • For the MRMC study, 9 readers (presumably radiologists, though specific qualifications for each weren't detailed beyond "clinical readers") participated in the study itself, but this is about their performance, not their role in establishing the ground truth.

3. Adjudication Method for Test Set:

   • The document doesn't explicitly describe an adjudication method (e.g., 2+1, 3+1) for establishing the ground truth of the test set cases based on radiologist interpretations. It mentions "biopsy outcomes" and "biopsy confirmed (Gleason score ≥ 7)" for positive cases, and "biopsy-confirmed (Gleason score < 7) or non-biopsied with a clinical followup of at least one year" for negative cases. This suggests pathology and long-term clinical follow-up played a primary role in ground truth for patient status.

4. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study:

   • Was it done? Yes, a fully crossed multi-reader multi-case study was performed.
   • Effect Size of Human Reader Improvement: The primary endpoint was an improvement in AUC.
     • AUC_unaided: 0.849 (95% CI: 0.814-0.884)
     • AUC_aided: 0.868 (95% CI: 0.834-0.902)
     • ΔAUC (AUC_aided - AUC_unaided): 0.019 (95% CI: 0.001-0.038), p-value: 0.039.
     • This indicates a small but statistically significant improvement in reader performance (AUC) when assisted by the AI.

5. Standalone Performance Study (Algorithm Only):

   • Was it done? Yes, a standalone performance assessment was conducted.
   • The results are summarized in "Table 2: Summary of the standalone performance testing for QP-Prostate® CAD" (refer to the table above).

6. Type of Ground Truth Used:

   • For the Test Set:
     • Pathology: Biopsy confirmed for both positive (Gleason score ≥ 7) and negative (Gleason score < 7) cases of clinically significant prostate cancer (csPCa).
     • Outcomes Data: For non-biopsied negative cases, ground truth was established by "a clinical followup of at least one year."
     • Expert Consensus/Radiologist Interpretation: Used in conjunction with pathology for the standalone lesion-level evaluation ("QP-Prostate® CAD outputs were compared to ground truth diagnoses derived from associated pathology reports and radiologist interpretations, at lesion-level").

7. Sample Size for Training Set:

   • The exact sample size for the training set is not explicitly stated in numerical form. It is described as "cases acquired in the US from multiple centers."

8. How Ground Truth for Training Set Was Established:

   • The AI-based algorithms were "trained with biopsy outcomes as ground truth to detect suspicious lesions for csPCa (Gleason score ≥7)." This indicates that pathology reports (biopsy outcomes) were the primary source of ground truth for the training data.

Ask a specific question about this device