About Use Cases

Use cases as corner stones

ACR DSI use cases describe specific clinical scenarios in which an AI application adds value and suggest features to garner trust and endorsement from the community. Use cases define basic requirements to complete certain automation and nest within broader IT environments. In addition to guidance and standards for developers, this list indicates the applications the DSI supports with curated datasets codified according to our use case specifications. 

Building wide-scale health care AI requires access to robust structured data. The DSI is part of a global initiative to open access to data for AI development and is engaging institutions across the country on data access projects. These projects rest upon the foundation of the use case, and each use case spells out the conditions algorithms are expected to reliably execute. For a given use case, the DSI develops validation models, some of which are part of ongoing demonstration projects with the FDA to support expedited approval processes for AI algorithms. Once in the field, the DSI leverages the ACR registry infrastructure connected to facilities across the country to provide analytics on algorithm performance, which can be used as real-world evidence or a basis for updates. All of these initiatives begin with scrupulously describing relevant algorithmic functions and clinical insights which ensure the application is effectively utilized.

Use case interpretation

ACR DSI use cases are organized in five sections: overview, clinical implementation, dataset considerations, technical specifications and future developments. Parts of the pneumothorax use case are explained below as an example.


Contains metadata on the use case which provides users with a quick understanding of use case content.


Detection and quantification of pneumothorax on chest radiograph








Eric J. Stern

Panel Chair

Eric J. Stern

Panel Reviewers

Thoracic Panel


Creative Commons 4.0


Related use cases may be bundled under a similar project represented in a tagging system.

Clinical implementation

This section describes the ideal algorithm that could improve the radiology care stream. Content is subdivided into the Value Proposition, Narrative and Workflow Description. The sections explain how certain AI automation is valuable to radiology workflows and the precise clinical scenario. Suggestions on how an application would integrate with common radiology IT infrastructure are included.

Considerations for dataset development

This section pinpoints the variants that may affect the presentation of the image and are often evaluated when determining a finding or recommending clinical management. This content serves to recommend the types of data that algorithms are expected to handle to earn trust in clinical environments.

Procedures(s): {X-ray, Chest; CT, Chest}

View(s): {AP, PA/Lat, inclination (eg, upright, semi-upright, supine)}
Clinical Note: Subtle cases of PTX on supine CXRs manifested by “deep sulcus sign” can be overlooked by less experienced radiologists

Sex at Birth: {Male, Female}

Age: [0,90]

Chest Trauma (or Intervention): {Blunt injury, blast injury, chest wall injury, rib fracture, lung contusions, lacerations, pleural fluid}

Comorbidities: {Pleural fluid (including air/fluid levels), lung disease (eg, Pneumonia/lung abscess, bullous emphysema, bronchiectasis), pneumomediastinum, other extrapleural air collections}

Lung Tissue Involvement: {Complete Collapse, Partial collapse}

Chest Tube: {With one or more chest tubes/drainage catheters, interval placement, or former tubes}

Tension Pneumothorax: {Tension pneumothorax (flattening of ipsilateral hemidiaphragm or contralateral shift of the mediastinum)}

Other: {Skin fold artifacts, Pleural fluid}
Clinical Note: Skin fold artifacts are the most common cause of a false-positive diagnosis of PTX


The variants listed here are not meant to be comprehensive, instead are variants that may not be obvious to those without clinical backgrounds. We recommend ensuring that algorithms can handle these imaging scenarios.

Technical specifications

This section specifies the standard terminology and markup for executing an algorithm. The source of these standards is RadElements, a collaborative work between the ACR, RSNA and major standards-bodies. Vendors pursuing validation and clinical implementation should expect to handle and transmit these data elements when performing the use case.




XRAY, Chest


AP, PA/Lat, inclination e.g., upright, semi-upright, supine

Data Type




Body Region


Anatomic Focus




Pneumothorax Detection

RadElement ID



Detection of pneumothorax

Data Type


Value Set

0-Pneumothorax absent

1-Pneumothorax present





Secondary Outputs

Pleural Separation

RadElement ID



Measure pleural separation in mm

Data Type


Value Set





Outputs are organized by primary and secondary elements. This is intended to provide developers ideas on how to unfold additional features on their product to improve clinical decision making.

Future development ideas

This section highlights additional opportunities for developers to enhance their product. It may contain user interface suggestions or more abstract ideas on the evolution of the application.

Future Development Ideas

The algorithm compares prior imaging and returns differences in output elements (pneumothorax detection, pleural separation and pneumothorax size). By further extension, the algorithm may return percent of total volume of hemithorax or volume of pleural fluid.

Use cases in the directory are a short list of applications the DSI plans to support. Based on community feedback, use cases will improve to capture more information pertinent to algorithm development. If you would like to contribute to use case develoment, please visit the Participate page for more information.