Understanding the Barriers to Translating Artificial Intelligence into the Clinical Laboratory

Artificial intelligence (AI) is having a transformational impact on society, yet its adoption in laboratory medicine has proceeded notably slower than in many other industries and even different specialities within medicine. This review sets out to examine why, despite such technical progress, meaningful clinical translation beyond rule-based autoverification has remained elusive. We argue that three principal barriers account for this gap. First, modelling approaches have been insufficiently robust for the inherent complexity of laboratory data. Second, the datasets available for model training and validation lack the scale, diversity, and operational representativeness required for genuine generalisation. Third, the regulatory environment constrains both the acquisition of data and the subsequent deployment of trained models. We trace the field's attempts at automation from autoverification systems through classical machine learning and single-modal deep learning approaches to the current generation of foundation and generative models, and we highlight the limitations and constraints of each approach. We then examine existing regulatory frameworks, available large-scale data initiatives and federated learning, a potential means to address these barriers.