# Development of a prognostic prediction model and visualization system for autologous costal cartilage rhinoplasty: an automated machine learning approach

## TL;DR

This paper introduces an automated machine learning system to predict outcomes and complications in costal cartilage rhinoplasty surgeries, improving patient satisfaction and surgical decision-making.

## Contribution

The novel contribution is an improved metaheuristic algorithm (INPDOA) integrated into an AutoML framework for ACCR prognosis and a MATLAB-based visualization system.

## Key findings

- The INPDOA-enhanced AutoML model achieved an AUC of 0.867 for predicting 1-month complications and an R2 of 0.862 for 1-year ROE scores.
- Key predictors of outcomes included nasal collision, smoking, and preoperative ROE scores, identified via SHAP values and bidirectional feature engineering.
- The developed CDSS reduced prediction latency and demonstrated net benefit improvement over conventional methods via decision curve analysis.

## Abstract

To develop an automated machine learning (AutoML)-based prognostic prediction model and visualization system for autologous costal cartilage rhinoplasty (ACCR), addressing the clinical challenges of postoperative complications and satisfaction disparity.

A retrospective cohort of 447 ACCR patients (2019–2024) was analyzed, integrating 20+ parameters spanning biological, surgical, and behavioral domains. We proposed an improved metaheuristic algorithm (INPDOA) for AutoML optimization, validated against 12 CEC2022 benchmark functions. Bidirectional feature engineering identified critical predictors, and SHAP values quantified variable contributions. A MATLAB-based clinical decision support system (CDSS) was developed for real-time prognosis visualization.

The INPDOA-enhanced AutoML model outperformed traditional algorithms, achieving a test-set AUC of 0.867 for 1-month complications and R2 = 0.862 for 1-year Rhinoplasty Outcome Evaluation (ROE) scores. Key predictors included nasal collision within 1 month, smoking, and preoperative ROE scores. Decision curve analysis demonstrated a net benefit improvement over conventional methods. The CDSS reduced prediction latency.

This study establishes the first AutoML-driven prognostic framework for ACCR, effectively bridging the gap between surgical precision and patient-reported outcomes. Its integration of dynamic risk prediction and explainable AI offers a paradigm for aesthetic surgical decision-making.