Change Healthcare Anatomical AI

Change Healthcare Anatomical AI is an automated image processing software that utilizes a non-adaptive machine learning algorithm to analyze pixel data from CT and MR DICOM images, generating anatomic descriptors. Its purpose is to supplement medical image management by enabling rapid identification of relevant studies and anatomical regions, thereby aiding physicians and integrated healthcare systems in organizing and accessing medical imaging data for patients 18 years and older.

Change Healthcare Anatomical AI is a standalone image processing software designed to assist healthcare professionals and integrated systems by analyzing pixel data from CT and MR images. Its primary function is to generate comprehensive anatomic descriptors, which are exported to healthcare systems to supplement traditional methods of image selection, presentation, and analysis. This capability allows for the rapid identification of relevant images, series, or studies, thereby enhancing workflow efficiency. The system operates by receiving DICOM images through APIs, processing them to associate specific anatomic regions with the images, and then outputting these findings as supplemental metadata without altering the original medical images. It is intended for use in adult patients (18 years and older). The core of the system is a non-adaptive machine learning algorithm that analyzes CT or MR pixel data to identify a wide range of anatomical structures, including regions such as the abdomen, chest, head, spine, and extremities. The device validates incoming image data for compatibility and, upon successful analysis, produces a JSON file detailing the identified body regions for each image and study. Results are aggregated and can be stored persistently. The software's performance, including its anatomical structure detection accuracy, was validated through software verification and a retrospective study. This study utilized diverse, de-identified datasets from 27 different institutions, encompassing various demographics, scanner types, and imaging protocols, to ensure clinical relevance and reduce bias in assessing the algorithm's accuracy across different conditions.

AI Regulatory Concerns

Powered by

Jev reads the full FDA letter and returns calibrated concern probabilities per regulatory category. These are AI estimates, not regulatory advice.

• Q1. How does the 'non-adaptive' nature of the machine learning algorithm affect its performance consistency across diverse patient populations and novel pathologies not extensively represented in the training data?; What were the specific quantitative accuracy metrics (e.g., precision, recall, F1-score) for each of the identified anatomical structures, particularly for less common views or in the presence of significant image artifacts?; What is the typical processing time per study for the Change Healthcare Anatomical AI, and how does this scale with increasing image series complexity or volume within a study?; How does the system handle ambiguous anatomical regions or images with poor diagnostic quality, and what mechanisms are in place for error reporting or flagging such cases to the user or integrated system?; What is the demonstrated impact on radiologist workflow efficiency or diagnostic read times when this system's anatomic descriptors are integrated into a clinical Picture Archiving and Communication System (PACS) for image retrieval and organization?