Feed

Background: This study employed artificial intelligence (AI) to analyze quantitative nuclear morphological features obtained from digital pathology images to predict postoperative recurrence in patients with lung squamous cell carcinoma (LSQCC). We aimed to develop a prediction model that contributes to the realization of 'personalized postoperative management' tailored to individual tumor biology by integrating AI-extracted morphological features with clinical information. Methods: A total of 185 of the 253 surgically resected LSQCC cases were included; 136 were randomly assigned to the training set and 49 to the test set. Nuclear features from manually selected regions of interest were extracted and used to build AI-based prediction models. Three recurrence models were developed: recurrence within 2 years, within 5 years, and a three-category model (≤2 years, 3-5 years, >5 years or no recurrence). Support vector machine (SVM) and random forest (RF) algorithms were applied to each, yielding six predictive models. An ensemble approach was used to calculate AI-based risk scores, and a "total risk score" was developed by integrating these with the pathologic stage. Results: All six AI models demonstrated stable predictive performance, with AUC values ranging from 0.76 to 0.91. Kaplan-Meier analysis showed that the total risk score provided the most precise risk stratification (p < 0.005), with clearer separation between risk groups than the AI-based risk score alone. Conclusions: The integration of AI-based nuclear morphology analysis and clinical data provides an objective and practical tool for personalized postoperative management in LSQCC. This approach enables tailored clinical decision-making by identifying patients at high risk for early recurrence and customizing postoperative treatment plans to meet the specific needs of each individual.