Dottorato in BIOTECNOLOGIE INNOVATIVE NELLE PATOLOGIE DEGLI ORGANI
DI SENSO

Titolo della tesi: Preliminary Digital Twin Model for Predicting Big Bubble Formation in Deep Anterior Lamellar Keratoplasty

Background Personalized medicine is progressively evolving from population-based evidence toward individualized prediction and simulation. Digital twins (DTs), virtual counterparts of biological systems, offer a framework to translate patient-specific data into predictive and adaptive models. Ophthalmology, characterized by quantitative imaging and structured biomarkers, provides an ideal setting for their development. Purpose This work presents the development of a preliminary digital twin model of the cornea aimed at predicting Big Bubble (BB) formation during Deep Anterior Lamellar Keratoplasty (DALK), a key intraoperative event determining the success of pneumatic dissection. Methods A retrospective analysis was performed on a dataset of patients with keratoconus who underwent DALK. Preoperative tomographic, topographic, and AS-OCT parameters were analyzed using logistic regression and a Random Forest (RF) model. Predictive performance was evaluated through receiver operating characteristic (ROC) analysis, and simulation procedures were implemented to assess how virtual modifications in corneal parameters influenced the predicted probability of BB formation at both population and individual levels. Results BB formation occurred in 69.2% of cases. Logistic regression identified AS-OCT stage and anterior keratometry grade as independent predictors (AUC = 0.72), whereas the RF model achieved higher discrimination (AUC = 0.82) without overfitting. Simulation analysis revealed physiologically consistent trends in BB probability with variations in corneal curvature and thickness. Retraining the model with an additional surgical observation led to adaptive modifications in prediction patterns, demonstrating the model’s potential for iterative learning consistent with the digital twin concept. Conclusions This study establishes a methodological framework integrating imaging-derived parameters with machine learning to simulate individualized surgical outcomes. The proposed model represents the predictive layer of a future corneal digital twin, capable of evolving through data assimilation and supporting simulation-informed decision-making in ophthalmic surgery.