NSCLC tumor shrinkage prediction using quantitative image features.

Area of Science:

• Radiology
• Medical Imaging
• Oncology

Background:

• Non-small cell lung cancer (NSCLC) treatment response varies significantly among patients.
• Accurate prediction of tumor shrinkage is crucial for personalized treatment strategies and prognosis.
• Current methods for assessing tumor response often rely on delayed measurements.

Purpose of the Study:

• To develop and validate a quantitative image feature model for predicting NSCLC volume shrinkage.
• To utilize pre-treatment computed tomography (CT) images for early prediction of treatment response.
• To assess the model's performance against traditional population-based shrinkage predictions.

Main Methods:

• Extracted quantitative image features (geometric, histogram, gradient, co-occurrence, run-length) from pre-treatment CT scans of 64 NSCLC patients.
• Developed prediction models using principal component regression and simulated annealing subset selection.
• Quantified tumor shrinkage by comparing planning gross tumor volume (GTV) to week 6 treatment GTV, validated using permutation tests.

Main Results:

• The optimal prediction model achieved a strong correlation (r=0.81) between observed and predicted tumor shrinkage.
• The model demonstrated a mean squared error (MSE) of 8.60×10(-3).
• The developed model reduced MSE by 2.92-fold compared to predictions based on mean population shrinkage.

Conclusions:

• Quantitative image features from pre-treatment CT scans can effectively predict NSCLC tumor shrinkage.
• This predictive model offers valuable insights for clinical decision-making, including risk stratification and prognosis.
• The findings support the integration of quantitative imaging biomarkers into routine NSCLC patient management.