Difference between revisions of "AI Results"
|Line 30:||Line 30:|
AI Result Primitives:
AI Result Primitives:
Latest revision as of 17:20, 9 December 2020
AI Results specifies how medical imaging analysis results are reliably stored, retrieved, and displayed. The central use case involves results generated by artificial intelligence (AI Model) algorithms.
This profile standardizes storage & display of AI Results for image interpretation. It establishes baseline data handling and presentation capabilities for an image display product to be “AI-Ready”. Result generation products can leverage these data formats to be compatible with a variety of displays and site workflows.
- a set of result “primitives”
- Encoding requirements for each primitive (DICOM-based storage)
- transactions for moving that content around
- Baseline display requirements for each primitive
- Results need to be presented to radiologists in their reading environment. This profile provides interoperability between result generation products and radiology reading workstation systems/software.
Study Integrated Results:
- Radiologists expect AI-generated results to be presented in the context of the study to which they apply and expect them to supplement (rather than replace) traditional image analysis results and thus a given study will be composed of acquired images, AI results, and traditional clinical data. The result objects defined in this profile are defined for the existing imaging data storage infrastructure.
- Effective use of results hinges on presenting them in conjunction with the associated images during the busy process of reading the study.
Convergence of Result Encoding:
- Many AI results are results that a human could otherwise have produced, and those human results may be used as training data for the AI. Analysis results might also be used by other AIs, e.g., in a GAN (generative adversarial network). AI and non-AI results need to be handled together. Convergent encoding of results facilitates this, as well as data pooling and sharing between sites.
- It is unrealistic to expect radiology displays to implement specific display capabilities for each of the myriad of algorithms being developed. To minimize implementation complexity for displays, and avoid needing different software for each new analysis result, compose analysis results from a reasonable set of primitives.
AI Result Primitives:
- Qualitative Findings: E.g. Breast composition is Extremely Dense, Pneumothorax is Present
- Measurements: E.g. Adrenal Gland Nodule Length = 52mm
- Locations: E.g. Blood clot at (123.9, 57.2, 752.3)
- Segmentations: E.g. Lung Nodule surface
- Parametric Maps: E.g. Saliency Map for pneumonia
- Tracking IDs: E.g. Primary tumor for patient 32457 (longitudinal)
- Image References: E.g. Best view of brain hemorrhage
- Evidence Creator - encodes results as specified
- May an analysis or AI software package itself, or it may be a proxy, gateway, or partner system that encodes and transmits results on behalf of such software. The profile does not distinguish between analysis executed locally, in the cloud, hosted in processing servers, or running in standalone workstations.
- Image Manager (PACS) - stores and manages AI result objects as part of study
- Image Display - retrieves results and displays as specified
- Imaging Document Consumer - uses results in ways other than displaying them
- May include decision support systems, clinical databases, and report creators.
Actors & Transactions:
Profile Status: Trial Implementation
Documents: AI Results Profile - TI
- For an overview of the scope and goals, see "Introduction to this Supplement", and Volume 3 Section 6.5.1.
- For an overview of the participating actors and the transactions between them, see Volume 1 Section 49.1 and Section 49.4.2.
- For key concepts that affect implementation and deployment and may be helpful in understanding the rest of the profile, see Volume 1 Section 49.4.1.
- For the key technical requirements on encoding and displaying analysis results, see Volume 3 Section 6.5.3 and Volume 2 Section 220.127.116.11.3 respectively.
- AI Workflow for Imaging [AIW-I] can be used to coordinate and manage the execution of AI algorithms whose results can be encoded and displayed using AIR.
This page is based on the Profile Overview Template