Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

"It depends": Configuring AI to Improve Clinical Usefulness Across Contexts

Hubert D. Zając, Jorge M. N. Ribeiro, Silvia Ingala, Simona Gentile, Ruth Wanjohi, Samuel N. Gitau, Jonathan F. Carlsen, Michael B. Nielsen, Tariq O. Andersen

TL;DR

The paper tackles the gap between lab-level AI performance and real-world clinical usefulness in radiology. It employs a design-through-research approach across Denmark and Kenya, iterating three AI prototype versions with 13 radiologists to identify four configurable technical dimensions (AI functionality, AI medical focus, AI decision threshold, AI Explainability) and ten sociotechnical dependencies (social dimensions) that condition usefulness. The main contributions are four concrete design recommendations and a dependency-informed framework for configuring AI before-use and in-use to align with clinic type, expertise level, patient context, and workload. The findings have practical significance for deploying AI in diverse clinical settings, suggesting that configurable AI can improve adoption, trust, and potentially patient outcomes by tailoring AI support to local practice. The work also highlights limitations and points to future exploration of richer clinical data integration and the potential role of large language models in radiology support.

Abstract

Artificial Intelligence (AI) repeatedly match or outperform radiologists in lab experiments. However, real-world implementations of radiological AI-based systems are found to provide little to no clinical value. This paper explores how to design AI for clinical usefulness in different contexts. We conducted 19 design sessions and design interventions with 13 radiologists from 7 clinical sites in Denmark and Kenya, based on three iterations of a functional AI-based prototype. Ten sociotechnical dependencies were identified as crucial for the design of AI in radiology. We conceptualised four technical dimensions that must be configured to the intended clinical context of use: AI functionality, AI medical focus, AI decision threshold, and AI Explainability. We present four design recommendations on how to address dependencies pertaining to the medical knowledge, clinic type, user expertise level, patient context, and user situation that condition the configuration of these technical dimensions.

"It depends": Configuring AI to Improve Clinical Usefulness Across Contexts

TL;DR

The paper tackles the gap between lab-level AI performance and real-world clinical usefulness in radiology. It employs a design-through-research approach across Denmark and Kenya, iterating three AI prototype versions with 13 radiologists to identify four configurable technical dimensions (AI functionality, AI medical focus, AI decision threshold, AI Explainability) and ten sociotechnical dependencies (social dimensions) that condition usefulness. The main contributions are four concrete design recommendations and a dependency-informed framework for configuring AI before-use and in-use to align with clinic type, expertise level, patient context, and workload. The findings have practical significance for deploying AI in diverse clinical settings, suggesting that configurable AI can improve adoption, trust, and potentially patient outcomes by tailoring AI support to local practice. The work also highlights limitations and points to future exploration of richer clinical data integration and the potential role of large language models in radiology support.

Abstract

Artificial Intelligence (AI) repeatedly match or outperform radiologists in lab experiments. However, real-world implementations of radiological AI-based systems are found to provide little to no clinical value. This paper explores how to design AI for clinical usefulness in different contexts. We conducted 19 design sessions and design interventions with 13 radiologists from 7 clinical sites in Denmark and Kenya, based on three iterations of a functional AI-based prototype. Ten sociotechnical dependencies were identified as crucial for the design of AI in radiology. We conceptualised four technical dimensions that must be configured to the intended clinical context of use: AI functionality, AI medical focus, AI decision threshold, and AI Explainability. We present four design recommendations on how to address dependencies pertaining to the medical knowledge, clinic type, user expertise level, patient context, and user situation that condition the configuration of these technical dimensions.
Paper Structure (37 sections, 8 figures, 2 tables)

This paper contains 37 sections, 8 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Clinical Usefulness of AI Systems in Healthcare
  4. System Configurability
  5. Methodology
  6. Research Through Design: Design Interventions with Working Prototypes
  7. Prototypes
  8. Analysis Positionality
  9. Data Analysis
  10. Configuring Four Technical Dimensions of Clinically Useful Radiological AI
  11. Social Dimensions of Clinical AI
  12. Medical knowledge.
  13. Clinic type.
  14. User expertise level.
  15. Patient context.
  16. ...and 22 more sections

Figures (8)

  • Figure 1: Online and co-located design sessions with user interface mock-ups and functional prototypes (from left: iteration I, II, and III)
  • Figure 2: The collage showcases three iterations of the AI prototype, including two screens: the X-ray viewer and the X-ray worklist.
  • Figure 3: Configuration matrix for achieving clinical usefulness of AI. Showing how the technical AI dimensions need to be configured against the social dimensions, which they depend on.
  • Figure 4: Two UI elements allowing the user to toggle prioritisation and decision support functionalities and to select findings on the X-ray work list and X-ray viewer.
  • Figure 5: A configuration panel allowing users to select radiological findings detected by the AI model.
  • ...and 3 more figures