Bridging AI and digital twins for real-time precision surgery: translating the COFFEE histopathological classifier into clinical workflows

Dear Editor, We read with great interest the prospective study by Lin et al detailing COFFEE, an innovative Transformer-based AI model for classifying histological growth patterns (HGPs) in colorectal liver metastases[1]. The model’s exceptional performance (AUC up to 1.00 in prospective cohorts) and its ability to augment junior pathologists’ diagnostic accuracy by 8.8% underscore its transformative potential in precision oncology. Although this work advances computational pathology, we wish to extend the discussion on strategies to accelerate its clinical translation, particularly through integration with digital health ecosystems and digital twin frameworks (Fig. 1). Figure 1.: Bridging AI and digital twins for real-time precision surgery: translating the COFFEE histopathological classifier into clinical workflows. First, the integration of AI models like COFFEE into dynamic surgical decision-support systems could enable real-time, adaptive treatment planning[2]. By embedding validated algorithms into interoperable platforms (e.g., Fast Healthcare Interoperability Resources-enabled Electronic Health Record), AI outputs could continuously inform pre-, intra-, and postoperative interventions. For instance, desmoplastic HGP identification could trigger personalized surveillance protocols or neoadjuvant therapy adjustments based on predicted overall survival/progression-free survival thresholds. Such automation would operationalize COFFEE’s prognostic insights beyond pathology reports, aligning with the vision of AI as a “perioperative copilot” that reduces cognitive burden[3]. Second, digital twin technology offers a scaffold for validating and scaling AI tools like COFFEE[4]. Creating virtual patient replicas – synthesizing pathology, genomics, radiomics, and clinical trajectories – would allow simulations of HGP-driven therapeutic scenarios. A liver metastasis “twin” could model how desmoplastic versus replacement HGPs respond to locoregional therapies or systemic agents (e.g., bevacizumab), refining COFFEE’s predictions against multimodal data streams. This approach resonates with Uzma Saddia Asghar et al’s framework, where digital twins contextualize AI outputs within individual biology, mitigating overgeneralization[5]. However, seamless translation necessitates addressing algorithmic adaptability across diverse populations[6]. COFFEE’s training on Chinese cohorts risks implicit geographical bias. Future iterations should validate robustness in multi-ethnic datasets to ensure generalizability. Federated learning architectures could enable collaborative model refinement without data centralization, upholding privacy while enhancing performance. Finally, regulatory harmonization remains critical[7]. COFFEE’s accuracy in prospective validation merits expedited certification under FDA pathways for AI-as-a-medical-device[8]. The EMA’s recent guidance on continuous real-world performance monitoring could serve as a blueprint for post-deployment surveillance, ensuring sustained efficacy. In conclusion, COFFEE exemplifies how domain-specific AI can revolutionize prognostication in surgical oncology[1]. Embedding such models within digital twin ecosystems will catalyze the shift from retrospective analysis to proactive clinical intervention, ultimately fulfilling the promise of precision surgery. We ensure that this article is compliant with the TITAN Guidelines[9].