ESTRO 2025 - Abstract Book

S3837

Physics - Radiomics, functional and biological imaging and outcome prediction

ESTRO 2025

Material/Methods:

We analyzed a retrospective cohort of 96 patients diagnosed with LSCC treated at our institution between June 2018 and November 2022, consisting of 77 males (80.2%) and 19 females (19.8%), with a median age of 65.7 years (range 48-86). Patients underwent either open partial horizontal laryngectomy (52% OPHL) or total laryngectomy (48% TL). Radiomic features were extracted from segmented regions of interest (VOIs), including the cricoid cartilage and laryngeal region, using an automated contouring process on preoperative contrast-enhanced CT scans to reduce inter-observer variability. The cohort was randomly divided into training (48 patients) and validation sets (48 patients). Radiomic features were correlated with T4 staging using univariate logistic regression with Bonferroni correction. Features with Pearson correlation > 0.80 were excluded to prevent multicollinearity. In addition to the radiomic model, a separate model based on conventional radiological data was constructed. Finally, integrated models combining radiological and radiomic features were developed to comprehensively assess predictive performance. The comparison between the different models was performed by assessing the respective AUCs of the ROC curves to determine which model provided the highest accuracy in distinguishing between patients with T4 tumors and those with T1-T3 tumors.