Inter-equipment validation of PET-based radiomics for predicting EGFR mutation statuses in patients with non-small...

Area of Science:

• Medical Imaging
• Oncology
• Radiomics

Background:

• Non-small cell lung cancer (NSCLC) treatment is often guided by EGFR mutation status.
• Radiomics analysis of PET images offers a non-invasive method for predicting EGFR mutations.

Purpose of the Study:

• To assess the inter-equipment generalizability of radiomics models for predicting EGFR mutation status in NSCLC patients.
• To evaluate the performance of radiomics models trained and tested on data from different PET scanner vendors (Siemens and GE).

Main Methods:

• Retrospective collection of PET images from two hospital branches using Siemens and GE equipment.
• Development of five logistic regression models: trained/tested on GE, trained/tested on Siemens, trained/tested on mixed data, trained on GE/tested on Siemens, and trained on Siemens/tested on GE.
• Area Under the Curve (AUC) was used to evaluate model performance.

Main Results:

• Models trained and tested on data from the same equipment (GE or Siemens) achieved higher AUCs (e.g., 0.78/0.73 for GE, 0.74/0.72 for Siemens) compared to models tested on data from different equipment.
• Models trained on one equipment vendor and tested on the other showed significantly lower performance (AUCs of 0.74/0.65 and 0.68/0.63).

Conclusions:

• The inter-equipment generalizability of radiomics models for predicting EGFR mutation status in NSCLC is currently insufficient for widespread clinical adoption.
• Further standardization or harmonization techniques are needed to improve the robustness of radiomics models across different PET imaging devices.