(132 days)
Change Healthcare Anatomical AI is intended to analyze pixel data from CT or MR images to create comprehensive anatomic descriptors for export to integrated healthcare systems.
This supplements traditional methods used for the selection, presentation or analysis of image based medical data. The application is intended to enable physicians, or other healthcare providers as well as integrated healthcare systems to rapidly identify images, series, and/ or studies of interest.
Change Healthcare Anatomical AI is not indicated for patients under the age of 18 years old.
Change Healthcare Anatomical AI is a standalone image processing software application that analyzes CT and MR DICOM images to associate anatomic regions with images and exports the derived information for use in integrated healthcare systems. These anatomic descriptors can be applied by integrated applications to categorize anatomy from a patient's CT or MR image, series, or study.
The device communicates via Application Programmable Interfaces (APIs) which allow for receiving DICOM images and returning inference results. The algorithm produces a JSON file which contains results of the analysis for each image and study with the corresponding identified body regions.
Change Healthcare Anatomical AI works in parallel to and in conjunction with the standard of care workflow. The device does not alter the original medical image in any way. The anatomic descriptors are used as supplemental metadata for a patient's imaging study.
Change Healthcare Anatomical AI contains the following core components:
API endpoints
The device uses API endpoints which allow for receiving DICOM images and returning results.
Following receipt of an image, the device performs data validation to ensure appropriateness and compatibility for the algorithm. If the validation fails and the image cannot be processed, an error is returned with the corresponding code and description.
AI algorithm
After validation, the algorithm analyzes the CT or MR image pixel data and generates the anatomic descriptors.
Study results aggregator
The results of the analysis for each image in a study are aggregated and returned to the integrated system.
Data store
The results of the inference for each analyzed image are maintained in a persistent data store. The results are stored by the algorithm inference model and retrieved by the study results aggregation component.
This document describes the regulatory submission for Change Healthcare Anatomical AI (K210719), an image processing software that analyzes CT and MR images to identify anatomical regions.
1. Table of Acceptance Criteria and Reported Device Performance
The submission does not explicitly state numerical acceptance criteria for a clinical study. However, it mentions that "Software verification testing assessed the performance of the software's anatomical structure detection function, performance characteristics of the algorithm including image-level accuracy..." and that "Test Summary Reports have been created to evaluate the acceptability of test results and all applicable verification and validation activities and records have been completed to ensure safety and effectiveness of the device."
The overall "performance" is implicitly evaluated against the predicate device (AquariusAPS Server, K061214) by demonstrating substantial equivalence, meaning the device performs as intended and does not raise new safety or effectiveness concerns. The specific performance reported is the ability to identify a broader range of anatomical structures and support additional modalities compared to the predicate.
| Acceptance Criteria (Implicit) | Reported Device Performance (Implicit) |
|---|---|
| Accurate anatomical structure detection | Identifies anatomical structures: Abdomen, breast (MR only), calf, chest, elbow, foot, forearm, hand, head, arm, knee, neck, pelvis, shoulder, spine cervical, spine thoracic, spine lumbar, and thigh. |
| Performance characteristics of the algorithm (image-level accuracy) | Retrospective study designed to assess subject device accuracy, with results evaluated according to patient demographics, healthcare institution, and other confounding imaging factors. (No specific metrics provided in this document). |
| Safety and effectiveness | Assessed through comprehensive software V&V, risk management, and cybersecurity controls. Concluded to be substantially equivalent to predicate, implying no new safety/effectiveness issues. |
| Broad anatomical coverage | Identifies significantly more anatomical structures compared to predicate (Brain, Heart, Heart Vasculature, Liver, Lung). |
| Multi-modality support | Supports CT and MR modalities, while predicate only supports CT. |
2. Sample Size for Test Set and Data Provenance
- Sample Size for Test Set: Not explicitly stated as a single number. The document mentions that for each modality, three databases were built (training, validation, and testing). The test databases "originated from a different healthcare system."
- Data Provenance:
- Country of Origin: Not explicitly stated, but implies multiple centers/institutions. "27 institutions from primary care hospitals, community hospitals and imaging centers contributed to the test datasets."
- Retrospective or Prospective: Retrospective study.
3. Number of Experts and Qualifications for Ground Truth – Test Set
This information is not provided in the given document. The document states "A retrospective study was designed to assess the subject device accuracy," but does not detail how the ground truth for this test set was established, including the number or qualifications of experts.
4. Adjudication Method for Test Set
This information is not provided in the given document.
5. Multi-Reader Multi-Case (MRMC) Comparative Effectiveness Study
A multi-reader multi-case (MRMC) comparative effectiveness study was not done based on the provided information. The study described focuses on the standalone performance of the AI algorithm.
6. Standalone Performance Study
Yes, a standalone (algorithm only without human-in-the-loop performance) study was done. The retrospective study was designed "to assess the subject device accuracy" focusing on the algorithm's ability to generate "anatomic descriptors." The "AI algorithm" component of the device "analyzes the CT or MR image pixel data and generates the anatomic descriptors."
7. Type of Ground Truth Used (for Test Set)
The specific type of ground truth (e.g., expert consensus, pathology, outcomes data) used for the test set is not explicitly stated. It is implied that for anatomical structure detection, the ground truth would be based on expert anatomical labeling or established anatomical atlases, but the document does not confirm this.
8. Sample Size for Training Set
The exact sample size for the training set is not explicitly stated. The document mentions that "For each modality, three databases were built for the AI model training, validation, and testing, with a balanced distribution of studies per body region."
9. How Ground Truth for Training Set Was Established
This information is not explicitly stated in the provided document. It can be inferred that the "balanced distribution of studies per body region" implies these regions were labeled, but the methodology for establishing these labels (e.g., expert labeling, automated segmentation, etc.) is not described.
§ 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).