Predicting Survival and Recurrence of Lung Ablation Patients Using Deep Learning-Based Automatic Segmentation and...

Area of Science:

• Oncology
• Radiology
• Artificial Intelligence

Background:

• Image-guided thermal ablation (IGTA) is a minimally invasive treatment for lung tumors.
• Accurate prediction of survival and recurrence post-ablation is crucial for patient management.
• Deep learning offers potential for automated tumor segmentation and feature extraction.

Purpose of the Study:

• To predict survival and tumor recurrence after IGTA for lung tumors.
• To evaluate a deep learning approach for tumor segmentation and radiomic feature extraction.
• To assess the performance of a radiomics-based model for outcome prediction.

Main Methods:

• Retrospective analysis of 113 patients undergoing IGTA for lung tumors.
• U-Net and UNETR (U-shaped encoder-decoder transformer) models for automated lung and tumor segmentation on CT scans.
• Radiomic features extracted from segmented tumors.
• Support vector machine (SVM) model trained for survival and recurrence prediction.
• Model performance evaluated using c-index and time-dependent AUC.

Main Results:

• UNETR achieved a Dice score of 0.75 for tumor segmentation.
• The radiomics model using post-ablation scans predicted survival with a c-index of 0.71 and AUC of 0.75.
• Pre-ablation scans yielded lower prediction metrics (c-index 0.56, AUC 0.56) for survival.
• For recurrence prediction, post-ablation scans achieved a c-index of 0.65 and AUC of 0.72.
• Pre-ablation scans showed reduced performance for recurrence prediction (c-index 0.54, AUC 0.54).

Conclusions:

• Automatic segmentation using transformer-based U-NET enables radiomic feature analysis.
• Radiomic features from post-ablation CT scans can predict survival and recurrence following IGTA.
• This AI-driven approach shows promise for improving prognostic accuracy in lung cancer treatment.