Table 3 Overview of challenges around AI implementation, change initiatives at Southern and learnings based on performed case study

Technology level

• Current AI applications have narrow functions, requiring the combined use of multiple applications [8]

• Organisations need to work with multiple vendors, each providing a narrow AI application, resulting in considerable overhead costs [17]

• There is a lack of standard user interfaces to integrate AI results into the clinical workflow [4]

Scalable and seamless AI implementation

• VNAI: automatically routes data, reads the metadata and triggers relevant AI applications to orchestrate the use of multiple applications

• VNAI: centralises legal and contractual procedures around AI implementation to reduce overhead costs

• Development of integrative frameworks to ensure integration into routine clinical software: enables seamless integration of AI results into the clinical workflow

• Initiate change efforts that comprehensively address not only technology but also structure, tasks and personnel involved in integrating AI into clinical practice

• Consider integrative platforms such as vendor-neutral AI platforms for integrating and orchestrating multiple AI initiatives

• Reduce the cost and time of implementation by defining a holistic framework for the testing, validation, integration, monitoring of AI tools

Workflow level

• Due to a lack of best practice, the use of how AI applications is highly varied in practice, making it hard to prove their value [18]

• AI applications are fitted to specific data, which may not correspond to the patient population of the organisation implementing them [13]

• Current AI applications generate results with limited ability to interact with radiologists and receive their feedback real-time [7]

Value-centric AI implementation

• IPG (radiographers): ensures a standardised and consistent use of technology and centralises expertise on technology use

• IPG (technical physicians): leads the local validation, fine-tuning, and continued monitoring of AI applications

• Making AI results modifiable: ensure that the end user remains in control and can accept/reject or adjust AI results before they are entered into the PACS system

• Select and prioritise AI implementation projects with a focus on creating value and addressing specific clinical problems. To this end, leverage the expertise and insights of clinicians to identify areas where AI can have the most significant contribution to clinical practice

• Ensure seamless integration of AI applications into clinical workflows to facilitate real-world adoption and maximise their impact

People and organisation level

• Radiologists have diverging expectations of the benefits and risks of AI [19]

• Radiologists do not yet have a basic proficiency in AI and machine learning [10]

• There is a gap between clinical and technical expertise, creating a barrier to the effective integration of AI technologies in healthcare

Organisation learning from AI implementation:

• CAI Group: helps radiologists and other clinicians by assessing the viability and necessity of proposed AI projects and providing a comprehensive checklist

• CAI Group: fosters learning among data scientists, radiologists and other clinicians as an umbrella organisation, encourages cross-departmental communication and collaboration and disseminates knowledge on AI

• Innovation steering committee: streamlines AI projects by centralising and evaluating bottom-up ideas on AI use cases

• Provide a clear organisational structure with dedicated people/roles for AI implementation (as opposed to being a part of the work of clinicians)

• Establish dedicated occasions for interaction between technical expertise and clinical knowledge to facilitate effective communication and collaboration between these two critical areas for AI implementation

• Make sure to manage local initiatives whilst also considering changes at a more global level and consider reorganising the current workflow to generate more value from AI applications

• Encourage cross-departmental (radiology vs other medical departments) knowledge sharing to foster a culture of learning and collaboration around AI implementation