This thesis examines the development and integration of an AI-driven radiology protocol selection software into the business processes of a small-sized radiology service provider. The research addressed the practical aspects of integrating AI into clinical workflows, particularly within the front desk and radiology acceptance processes.

Stakeholder interviews highlighted the importance of transparency in AI systems and the need for thorough employee training to ensure smooth adoption. A "to-be" process was developed to outline exactly where the AI solution would be implemented, including the necessary data, interfaces, and new processes. Key performance indicators (KPIs) were identified to evaluate the integration of AI. The study also developed and evaluated a machine learning framework which was used to automate the selection of radiology imaging procedures. This framework integrates traditional machine learning algorithms with advanced NLP techniques within a multi-tiered model architecture, culminating in an ensemble model. The Random Forest ensemble model achieved the highest evaluation accuracy at 74.1%, indicating its potential for effective automation. However, challenges such as overfitting, data limitations, and concerns about AI transparency were noted.

The findings suggest that further research is necessary to enhance model generalization, improve transparency, and ensure the successful incorporation of AI into business operations. This work provides a foundation for broader AI adoption in radiology, offering insights into both the technical and organizational aspects of AI integration.