New DeepCPath Paper Outlines the Path to Truly Translational Clinical AI

With an emphasis on applied evidence, we provide an in-depth evaluation of AI in cancer histopathology through the lens of United States Food and Drug Administration-approved and European Conformity-marked whole-slide image In Vitro Diagnostic Medical Devices. Having identified only four existing FDA-approved whole-slide image cancer solutions for a narrow range of applications, we conclude that AI in digital histopathology is still emerging.

Best practices were identified by examining development and validation evidence across market-approved solutions. Findings were contrasted with state-of-the-art research-only AI histopathology pipelines. Insights were drawn regarding applications, learning modalities, processing strategies, statistical methods, and validation approaches. Regulatory guidelines were evaluated from FDA and UK Government documentation and academic literature, where patient safety was a central concern. We observed that approved products integrated efficiently into existing clinical decision-making frameworks, with future potential to enhance the use of pathologist consensus for AI applications. We also observed that biomarker assays may be coupled specifically to emerging therapies, but with challenges for direct clinical adoption outside of the research-only sphere.

Although challenges remain in validating agentic and generative AI for medicine, further adoption of state-of-the-art algorithmic frameworks, including Transformer architectures and multimodal approaches, is anticipated. As pan-cancer systems emerge, computational modules and engineering principles may also be retargeted to underrepresented use cases. Consequently, this review provides a forward-looking framework for translational, market-relevant histopathology AI.

Last modified: Fri, 19 Jun 2026 20:43:19 BST