Jan Lukas Robertus, consultant cardiothoracic pathologist at the Royal Brompton and senior lecturer at Imperial College, delivered a comprehensive overview of how digital pathology and AI are revolutionising cancer diagnostics. He began by highlighting the shift from traditional glass slides to digital pathology, which transforms static samples into rich data assets accessible globally. This digital revolution allows remote diagnoses and facilitates extensive research using archived pathological material.
He underscored the critical role of pathologists in diagnosis, particularly in oncology, while noting the limitations of human subjectivity. AI algorithms, trained on digital slides, have now reached and in some cases exceeded the diagnostic accuracy of human experts. For example, AI tools can predict EGFR mutations in lung cancer directly from scanned slides, potentially bypassing lengthy genetic testing procedures.
Robertus turned attention to cytology, particularly fluid-based samples, which offer a minimally invasive method for early cancer detection. He detailed AI applications in cervical cytology and pleural effusion analysis, where machine learning models can not only flag atypical cells but now also provide direct diagnostic classifications, potentially replacing the pathologist in some workflows.
He presented a case study on mesothelioma, showcasing an unsupervised AI model that accurately differentiates between epithelioid and sarcomatoid subtypes, improving prognostic precision. However, the vast data size of whole slide imaging poses serious storage and computational challenges, especially when implementing high-resolution Z-stacking and deep learning models.
Looking ahead, Robertus introduced the potential of quantum machine learning to optimise diagnostic speed and efficiency. By using only partial datasets, these methods promise to maintain accuracy while reducing computational load – essential for scaling AI across the NHS.
He concluded with a call for multi-modal, hospital-integrated AI strategies, pointing to projects like real-time AI-guided diagnostics and federated data platforms. The presentation closed on a humorous yet thoughtful note, reflecting on the evolution of pathology and the coming era where AI not only supports but potentially surpasses human expertise in diagnostic medicine.
