IntelliLung AI-DSS Trustworthiness Evaluation Framework

The Artificial Intelligence Act was adopted in the European Parliament in March 2024 to establish a uniform legal framework for the development and uptake of human-centric and trustworthy artificial intelligence (AI) in Europe. Considering that AI may generate risks and cause harm, the approach for evaluating the newly developed AI and decision support systems (DSS) will vary from domain to domain. This paper presents the approach the European Union (EU) funded project IntelliLung is currently implementing in the healthcare domain. The IEC 62559-2:2015 methodology was used to structure the IntelliLung AI-DSS requirements and define key performance indicators relevant for the functional parts of the system (data integration, pre-processing, AI modelling). "Ethics-by-design" and "Transparent AI design" methodologies have been used for the IntelliLung AI-DSS implementation. The compliance requirements analysis has shown that the trustworthiness of AI-DSS relies on a wide set of measures including information and communication security, explainable AI algorithms, data governance and privacy considerations, as well as risk-management throughout the entire lifecycle of AI-DSS as a high-risk AI system.