Search Results

Saige-Dx analyzes digital breast tomosynthesis (DBT) mammograms to identify the presence of soft tissue lesions and calcifications that may be indicative of cancer. For a given DBT mammogram, Saige-Dx analyzes the DBT image stacks and the accompanying 2D images, including full field digital mammography and/or synthetic images. The system assigns a Suspicion Level, indicating the strength of suspicion that cancer may be present, for each detected finding and for the entire case. The outputs of Saige-Dx are intended to be used as a concurrent reading aid for interpreting physicians on screening mammograms with compatible DBT hardware.

Saige-Dx is a software device that processes screening mammograms using artificial intelligence to aid interpreting radiologists. By automatically detecting the presence or absence of soft tissue lesions and calcifications in mammography images, Saige-Dx can help improve reader performance, while also reducing time. The software takes as input a set of x-ray mammogram DICOM files from a single digital breast tomosynthesis (DBT) study and generates finding-level outputs for each image analyzed, as well as an aggregate case-level assessment. Saige-Dx processes both the DBT image stacks and the associated 2D images (full-field digital mammography (FFDM) and/or synthetic 2D images) in a DBT study. For each image, Saige-Dx outputs bounding boxes circumscribing any detected findings and assigns a Finding Suspicion Level to each finding, indicating the degree of suspicion that the finding is malignant. Saige-Dx uses the results of the finding-level analysis to generate a Case Suspicion Level, indicating the degree of suspicion for malignancy across the case. Saige-Dx encapsulates the finding and caselevel results into a DICOM Structured Report (SR) object containing markings that can be overlaid on the original mammogram images using a viewing workstation and a DICOM Secondary Capture (SC) object containing a summary report of the Saige-Dx results.

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

Acceptance Criteria and Reported Device Performance

Study Details

Here's a breakdown of the studies conducted:

1. Multi-Reader Multi-Case (MRMC) Reader Study (Performance Testing: Reader Study)

• Sample Size for Test Set: 240 cases (100 pathology-proven cancer cases, 140 confirmed non-cancer cases).
  • Data Provenance: Retrospectively collected DBT mammogram exams from unique female patients 35 years of age or older, acquired from Hologic equipment. Data was from different clinical sites than those used for AI algorithm training. Patients represented a racially and ethnically diverse population in the US.
• Number of Experts for Ground Truth: Two MQSA qualified, highly experienced (>10 years in practice) breast imaging specialists, plus a third as an adjudicator.
• Qualifications of Experts for Ground Truth: MQSA qualified, highly experienced (>10 years in practice) breast imaging specialists.
• Adjudication Method: For exams with discrepancies between the two truthers' assessment of density, lesion type, and/or lesion location, a third truther served as the adjudicator.
• MRMC Comparative Effectiveness Study: Yes.
  • Effect Size (Human Reader Improvement with AI vs. without AI):
    • Average AUC increased by 0.06 (from 0.865 unaided to 0.925 aided).
    • Average reader sensitivity increased by 8.8%.
    • Average reader specificity increased by 0.9%.
• Standalone Performance: No, this specific study was for human reader performance with and without AI.
• Type of Ground Truth: Expert consensus with pathology confirmation for cancer cases. Each mammogram had a ground truth status of "cancer" or "non-cancer." For cancer exams, malignant lesions were annotated based on the biopsied location that led to malignant pathology.
• Sample Size for Training Set: Not explicitly stated, but the text mentions "six datasets across various geographic locations in the US and the UK," indicating a large, diverse dataset.
• How Ground Truth for Training Set was Established: Not explicitly detailed for the training set, but it is stated that "DeepHealth ensured that there was no overlap between the data used to train and test the Saige-Dx Al algorithm." It can be inferred that similar robust methods (likely expert review and pathology confirmation) were used, given the thoroughness described for the test set.

2. Standalone Study (Performance Testing: Standalone Study)

• Sample Size for Test Set: 1304 cases (136 cancer, 1168 non-cancer).
  • Data Provenance: Retrospective, blinded, multi-center study. Collected from 9 clinical sites in the United States. All data came from clinical sites that had never been used previously for training or testing of the Saige-Dx AI algorithm.
• Number of Experts for Ground Truth: "Truthed using similar procedures to those used for the reader study," which implies two highly experienced breast imaging specialists and a third adjudicator.
• Qualifications of Experts for Ground Truth: Implied to be MQSA qualified, highly experienced (>10 years in practice) breast imaging specialists, consistent with the reader study.
• Adjudication Method: Implied to be consistent with the reader study (third truther for discrepancies).
• MRMC Comparative Effectiveness Study: No, this was a standalone performance study of the algorithm only.
• Standalone Performance: Yes. Saige-Dx exhibited an AUC of 0.930 (95% CI: 0.902, 0.958).
• Type of Ground Truth: Implied to be expert consensus with pathology confirmation, consistent with the reader study, as data was "collected and truthed using similar procedures."
• Sample Size for Training Set: Not explicitly stated, but the data used was specifically excluded from the test set for this study, confirming separation.
• How Ground Truth for Training Set was Established: Implied to be through expert review and pathology confirmation, given the "similar procedures" used for test set truthing and the isolation of training data.

Ask a specific question about this device