PET/CT-based deep learning model predicts distant metastasis after SBRT for early-stage NSCLC: A multicenter study

Area of Science:

• Oncology
• Radiology
• Medical Imaging

Background:

• Distant metastasis (DM) is the primary recurrence pattern after stereotactic body radiation therapy (SBRT) for early-stage non-small cell lung cancer (NSCLC).
• Accurate prediction of DM risk before treatment is crucial for effective patient management and therapeutic strategy selection.

Purpose of the Study:

• To develop and validate a deep learning fusion model utilizing 18F-FDG PET/CT imaging.
• To predict the risk of distant metastasis (DM) in early-stage NSCLC patients treated with SBRT.

Main Methods:

• A cohort of 566 patients was divided into training, internal test, and external test sets.
• Deep learning features were extracted from CT, PET, and fused PET/CT images using a variational autoencoder.
• Metastasis-free survival (MFS) prognostic models were constructed using fully connected networks.

Main Results:

• The deep learning fusion model achieved superior predictive performance (C-indices: 0.864 training, 0.819 internal, 0.782 external) compared to CT or PET models alone.
• The model effectively stratified patients into high- and low-risk groups with significantly different MFS.
• The fusion model was identified as an independent prognostic factor for MFS.

Conclusions:

• The 18F-FDG PET/CT deep learning-based fusion model offers a robust method for predicting DM risk and MFS in early-stage NSCLC patients receiving SBRT.
• This predictive tool can provide objective data to support individualized treatment planning and decision-making.