What is TOUCH-AI?
TOUCH-AI stands for Technically Oriented Use Cases for Healthcare-AI.
What is a use case?
Use case is a generic term for “a methodology used in system analysis to identify, clarify and organize system requirements.” Each ACR DSI Use Case provides narrative descriptions and standards which specify the goal of the algorithm, the required clinical input, how it should integrate into the clinical workflow, and how it could interface with users and tools.
In what way is the TOUCH-AI Directory unique?
These first-of-their-kind use cases are building a framework to facilitate the development and implementation of artificial intelligence (AI) applications that will help radiology professionals in disease detection, characterization and treatment. They enable data scientists to produce algorithms that:
- Address relevant clinical questions and can improve patient care
- Can be implemented across multiple electronic workflow systems
- Comply with requirements to submit data to relevant registries to enable ongoing assessment
- Comply with applicable legal, regulatory and ethical requirements
What are some examples of use cases?
Pneumothorax detection (collapsed lung) is a use case which may generate a lot of interest because it is a potentially life-threatening condition, in which time is of the essence. AI can be used to speed diagnosis and intervention from a variety of end points — prioritizing worklists, queuing up diagnostic tools for the radiologists, alerting referring providers of the condition.
LV functional assessment (for conditions like heart disease) is another important use case because algorithms will provide fast, accurate measurements of the most important factors in heart healthiness. This gives clinicians more information for making a diagnosis and facilitates better patient care without requiring more effort on the part of a physician. Access to trusted, large-scale data could provide great insight into research across patient histories and demographics.
Classifying disease severity based on micro-calcifications (for breast cancer) also has potential because automatically classifying micro-calcifications to prevent false positive and over diagnosis, will allow patients to avoid undergoing biopsies, without the associated risk of underdiagnoses.
Scoliosis detection is an interesting case because of the variance in reporting and managing scoliosis patients. In this scenario algorithms return a few common metrics that are typically time-intensive to do manually. Though metrics like the Cobb angle are commonly reported today, they have their limitations in accounting for vertebrae rotation. So the use case defines a series of secondary metrics that would best help radiologists call and treat scoliosis.
Who develops use cases for the ACR DSI?
Physicians, medical physicists, data scientists, software engineers, radiology business leaders and patient advocates selected to provide expert support for the activities of the Institute. They were chosen based on their clinical expertise and experience in the areas of data science, framework, terminology, methodology and clinical integration.
Can you give an example of the type of developer who would be most interested in use cases?
Any developer looking for a high impact problem to solve with AI would be interested in use cases. Despite the AI hype, at this stage most successful companies are focused on key narrow AI problems, and the smaller those problems the better. In radiology AI most developers will want to maintain a sharp focus on a specific medical problem, develop an algorithm that works alongside and augments humans (providing clinical decision support), gain access to good data for algorithm training and validation, and maintain awareness of the federal regulatory process throughout.
Will other use cases be released? When?
Yes, the ACR DSI will continue to develop use cases for the next few years. The potential exists to create thousands of use cases. The ACR DSI anticipates creating several hundred freely available use cases by 2020.