Progression-Free Survival Prediction Model Based on AI-Enhanced Dynamic Radiomics for Personalized EGFR-TKI Treatment...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Epidermal growth factor receptor tyrosine kinase inhibitors (EGFR-TKIs) are standard first-line treatment for advanced lung adenocarcinoma (LUAD) with EGFR mutations.
• Treatment effectiveness varies, and predicting response remains challenging.
• Lack of effective models hinders personalized treatment strategies.

Purpose of the Study:

• To establish a progression-free survival (PFS) prediction model for advanced LUAD patients.
• To utilize dynamic changes in pre- and post-treatment CT scans combined with clinical features.
• To improve prediction of treatment response in EGFR-mutant LUAD.

Main Methods:

• A 3D-UNet model was fine-tuned for advanced lesion segmentation.
• Clinical and radiomic features were extracted from CT scans of 80 EGFR-mutant LUAD patients.
• A deep-learning binary classification model was developed and validated for PFS prediction.

Main Results:

• The prediction models demonstrated strong performance in the EGFR-mutant test set (N=53).
• Area Under the Curve (AUC) values for 9- and 12-month progression prediction were 0.858 and 0.873, respectively.
• High accuracy, specificity, sensitivity, and F1 scores were achieved, indicating robust predictive capabilities.

Conclusions:

• Developed treatment response prediction models for EGFR-mutant LUAD patients.
• Models show significant predictive value for PFS in patients receiving EGFR-TKIs.
• Potential to enable more efficient and personalized CT scan scheduling for LUAD patients.