Artificial intelligence (AI) and machine learning (ML) are increasingly applied across the transplantation care continuum to support clinical decision-making, improve diagnostic accuracy, and personalise patient management. The reviewed evidence spans preoperative planning, perioperative risk prediction and management, and postoperative monitoring and rehabilitation. While these technologies show potential to improve outcomes and workflow efficiency, successful clinical translation depends on rigorous validation, dataset diversity, and appropriate ethical and regulatory frameworks.
Preoperative applications of AI focus on anatomical assessment, volumetry, graft weight estimation, and functional evaluation to enhance donor and recipient selection. Deep learning algorithms have been used to improve segmentation and three-dimensional assessment of donor anatomy, enabling more accurate volumetric calculations and graft weight estimation from CT and magnetic resonance imaging. These imaging-based tools aid surgical planning and can refine risk assessment by providing precise anatomical metrics.
ML approaches are also utilised for evaluating recipient functional status and urgency scoring. Models that integrate metabolic and physiological data can contribute to surgical eligibility assessments and support targeted prehabilitation strategies. The review highlights that predictive models incorporating diverse clinical inputs may improve candidate selection, although specifics of model performance and datasets were not detailed in the abstract.
During the perioperative period, ML models have been developed to predict critical intraoperative and early postoperative events. Examples cited include prediction of massive transfusion, intraoperative hemorrhage, and acute kidney injury. Explainable outputs from ML models are emphasised as important for clinical interpretability and trust, allowing clinicians to understand contributing features and rationale behind risk estimates.
Robotic and AI-assisted surgical platforms are described as demonstrating functional equivalence or superiority compared with conventional techniques in some settings. The review notes potential benefits such as reduced intraoperative complications and accelerated recovery, particularly among high-risk cohorts. Details on specific platforms, comparative metrics, or trial results are not provided in the abstract and would be found in the full text and referenced studies.
Postoperative applications of ML include early prediction of infectious complications (for example, sepsis and pneumonia) and graft dysfunction. Longitudinal markers—such as the recipient-to-donor estimated glomerular filtration rate (eGFR) ratio—are reported as informative in long-term graft monitoring. The review also references novel imaging and biomarker-based approaches driven by ML to detect early signs of graft injury or rejection.
Optimised perioperative strategies informed by AI, including analgesic regimens and fluid management, are mentioned as contributing to improved donor recovery and rehabilitation outcomes. However, the abstract does not provide granular outcome data or quantitative effect sizes.
Cross-cutting imaging innovations include hyperspectral imaging for real-time assessment of graft viability and the application of deep learning to automate histopathological evaluation. These methods may improve diagnostic speed, reproducibility, and accuracy compared with manual interpretation. Automated histology using deep learning can standardise assessments and potentially reduce interobserver variability, while imaging-based assessments can provide immediate intraoperative feedback on tissue perfusion and viability.
The review highlights multimodal ML models that integrate electronic health records, intraoperative signals, ultrasound, and histology to deliver dynamic, system-level insights. Such integrated models aim to bridge diagnostic, prognostic, and therapeutic decision-making by synthesising heterogeneous data streams to estimate graft function and rejection risk over time. Multimodal approaches are positioned as a means to provide personalised, data-driven care across the transplant journey.
Despite promising applications, the translation of AI and ML into routine transplant care requires rigorous external validation and attention to dataset diversity to ensure generalisability across populations. The authors emphasise the need for strong ethical and regulatory governance when deploying these tools clinically. Specific regulatory pathways, performance thresholds, or governance frameworks are not detailed in the abstract and would require consultation of the full review and relevant guidance documents.
AI and ML hold substantial potential to personalise transplant care by improving preoperative planning, enhancing perioperative safety, and enabling earlier detection of postoperative complications and graft dysfunction. Imaging-based AI and automated histopathology offer opportunities to accelerate and standardise diagnostic workflows, while multimodal models could integrate diverse data sources for dynamic clinical decision support. The review concludes that real-world impact will depend on robust validation, representative datasets, and ethical/regulatory oversight. The authors report no conflicts of interest for this article.