(20 days)
Saige-Density is a software application intended for use with compatible full-field digital mammography (FFDM) and digital breast tomosynthesis (DBT) systems. Saige-Density provides an ACR BI-RADS Atlas 5th Edition breast density category to aid interpreting physicians in the assessment of breast tissue composition. Saige-Density produces adjunctive information. It is not a diagnostic aid.
Saige-Density is Software as a Medical Device that processes screening and diagnostic digital mammograms using deep learning techniques and generates outputs that serve as an aid for interpreting radiologists in assessing breast density. The software takes as input a single x-ray mammogram study and processes all acceptable 2D image DICOM files (FFDM and/or 2D synthetics) and generates a single study-level breast density category. Two DICOM files are outputted as a result: 1) a structured report (SR) DICOM object containing the case-level breast density category and 2) a secondary capture (SC) DICOM object containing a summary report with the study-level density category. Both output files contain the same breast density category ranging from "A" through "D" following Breast Imaging Reporting and System (BI-RADS) 5th Edition reporting guidelines. The SC report and/or the SR file may be viewed on a mammography viewing workstation.
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
The provided text doesn't explicitly list a table of acceptance criteria with specific numerical targets. Instead, it states that the device was validated through a retrospective study (as described in a prior submission, K222275) and that "Verification and Validation testing conducted to support this submission confirm that Saige-Density is safe and effective for its intended use."
The key performance described is the ability to produce an ACR BI-RADS Atlas 5th Edition breast density category to aid interpreting physicians. The device outputs a study-level breast density category ranging from "A" through "D."
To infer the de facto acceptance criterion for performance, we must assume it aligns with demonstrating substantial equivalence to the predicate device (Saige-Density v2.0.0, K222275). This implies that the current version (v2.5.0) performs at least as well as, or equivalently to, the predicate in its ability to classify breast density according to the BI-RADS standard. While no specific performance metrics (like accuracy, sensitivity, specificity, or agreement rates) are stated in this document for this specific submission's validation, the statement of substantial equivalence implies that these metrics were deemed acceptable in the original K222275 submission.
Study Details:
The provided text primarily refers back to the validation performed for the predicate device (K222275) for its clinical performance data. The current submission focuses on verifying that minor technological changes in v2.5.0 do not impact safety or effectiveness.
-
A table of acceptance criteria and the reported device performance:
As noted above, no explicit table of numerical acceptance criteria or performance metrics for this specific submission is provided. The acceptance hinges on demonstrating "safety and effectiveness for its intended use" and "substantial equivalence" to the predicate, which implies the previous validation (K222275) satisfied performance requirements. -
Sample size used for the test set and the data provenance (e.g., country of origin of the data, retrospective or prospective):
- Sample Size (Test Set): Not explicitly stated in this document. It refers to the validation study described in K222275.
- Data Provenance: Retrospective study. Data was obtained from "different clinical sites than those used to develop the Saige-Density algorithm." Geographic locations for the training data included "various geographic locations within the US, including racially diverse regions such as New York City and Los Angeles." It's reasonable to infer the test set likely drew from similar diverse US populations to ensure generalizability.
-
Number of experts used to establish the ground truth for the test set and the qualifications of those experts (e.g., radiologist with 10 years of experience):
Not explicitly stated in this document. This information would typically be found in the K222275 submission details. -
Adjudication method (e.g., 2+1, 3+1, none) for the test set:
Not explicitly stated in this document. This information would typically be found in the K222275 submission details. -
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:
Not explicitly stated in this document. The device "provides an ACR BI-RADS Atlas 5th Edition breast density category to aid interpreting physicians," suggesting it's an adjunctive tool, but this document does not describe an MRMC study comparing human performance with and without the AI. -
If a standalone (i.e. algorithm only without human-in-the-loop performance) was done:
Yes, the device outputs "a single study-level breast density category" and DICOM files containing this category. The validation study references in K222275 would have assessed the algorithm's performance in categorizing density. The use of "retrospective study" suggests an assessment of the algorithm's output against a ground truth. -
The type of ground truth used (expert consensus, pathology, outcomes data, etc.):
Not explicitly stated in this document. Given that the output is an "ACR BI-RADS Atlas 5th Edition breast density category," the ground truth was most likely established by expert radiologists (likely through consensus or a similar process using their interpretation of the mammograms). Pathology or outcomes data are less likely to directly establish BI-RADS density categories. -
The sample size for the training set:
Not explicitly stated in this document. It mentions the training data consisted of "four datasets across various geographic locations within the US." -
How the ground truth for the training set was established:
Not explicitly stated in this document. It is implied that the ground truth for training would also be established by similar expert interpretation of BI-RADS density categories. The text notes "DeepHealth ensured that there was no overlap between the data used to train and test the Saige-Density algorithm," indicating good practice in study design.
§ 892.2050 Medical image management and processing system.
(a)
Identification. A medical image management and processing system is a device that provides one or more capabilities relating to the review and digital processing of medical images for the purposes of interpretation by a trained practitioner of disease detection, diagnosis, or patient management. The software components may provide advanced or complex image processing functions for image manipulation, enhancement, or quantification that are intended for use in the interpretation and analysis of medical images. Advanced image manipulation functions may include image segmentation, multimodality image registration, or 3D visualization. Complex quantitative functions may include semi-automated measurements or time-series measurements.(b)
Classification. Class II (special controls; voluntary standards—Digital Imaging and Communications in Medicine (DICOM) Std., Joint Photographic Experts Group (JPEG) Std., Society of Motion Picture and Television Engineers (SMPTE) Test Pattern).