3120 - Dosiomics and Radiomics for Predicting Local Recurrence In Local Advanced NSCLC Following IMRT

12:30pm - 01:45pm ET

Poster Hall - Exhibit Hall A

Screen: 19

POSTER

Presenter(s)

Yan Shao, PhD - Shanghai Chest Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, Shanghai

Y. Shao¹, and Z. Xu²; ¹Department of Radiation Oncology, Shanghai Chest Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, Shanghai, China, ²Department of Radiation Oncology, Shanghai Chest Hospital, Shanghai Jiao Tong University School of Medicine, Shanghai, Shanghai, China

Purpose/Objective(s): This study aims to investigate the predictive value of dosiomic and radiomic features for local recurrence in patients with locally advanced non-small cell lung cancer (LA-NSCLC).

Materials/Methods:

A retrospective study was conducted on 449 patients who received definitive radiotherapy at our institution between 2015 and 2021. Following the application of inclusion and exclusion criteria, 176 patients were ultimately enrolled and divided into a training set (n=140) and a test set (n=36). Clinical features, dose-volume histogram (DVH) parameters, dosiomic features, and radiomic features were extracted for each patient. Logistic regression models were then constructed using four different feature combinations: (1) clinical; (2) clinical + DVH; (3) clinical + DVH + radiomic; and (4) clinical + DVH + radiomic + dosiomic (hybrid). Model performance was evaluated using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, and specificity.

Results:

The models based on clinical, clinical + DVH, clinical + DVH + radiomic, and the hybrid model incorporated 2, 3, 9, and 20 features, respectively. In the first three models, treatment modality was the most important predictor. Dosiomic and radiomic features were more influential in the hybrid model. The AUCs for the four models were 0.66, 0.64, 0.68, and 0.88, with corresponding accuracies of 0.75, 0.72, 0.75, and 0.89.

Conclusion:

The proposed prognostic prediction model, which integrates both dosiomic and radiomic features, demonstrates strong predictive performance for local recurrence in patients with LA-NSCLC undergoing definitive radiotherapy.

Model	Accuracy	AUC (95% CI)	Sensitive	Specificity	Recall	F-score
Clinical
Training	0.69	0.74(0.65-0.83)	0.79	0.67	0.74	0.61
Testing	0.75	0.66(0.42-0.90)	0.50	0.88	0.50	0.57
Clinical+DVH
Training	0.73	0.76(0.68-0.85)	0.65	0.77	0.65	0.61
Testing	0.72	0.64(0.41-0.88)	0.67	0.75	0.67	0.62
Clinical+DVH+radiomics
Training	0.81	0.84(0.76-0.91)	0.65	0.89	0.65	0.70
Testing	0.75	0.68(0.48-0.89)	0.50	0.88	0.50	0.57
Hybrid
Training	0.86	0.89(0.83-0.95)	0.78	0.89	0.78	0.78
Testing	0.89	0.88(0.72-1.00)	0.83	0.92	0.83	0.83