Variogram Modeling of Spatially Variant Early Response to Therapy in Advanced Non-Small Cell Lung Cancer.

Area of Science:

• Oncology
• Radiomics
• Statistical Modeling

Background:

• Personalized oncologic management requires predicting treatment response heterogeneity in non-small cell lung cancer (NSCLC).
• Understanding spatial correlation of tumor response at the voxel level is crucial for accurate prediction.

Purpose of the Study:

• To evaluate and compare different variogram models for predicting voxel-level spatial correlation in tumor response within NSCLC patients.
• To identify the most accurate model for characterizing spatial heterogeneity in treatment response.

Main Methods:

• Analysis of tumor response data from two clinical trials involving locally advanced and metastatic NSCLC.
• Evaluation of various variogram models, including Stable, Matérn, and Exponential, to assess spatial correlation.
• Comparison of model performance using root mean squared error (RMSE).

Main Results:

• The Stable model demonstrated the lowest average RMSE (5.2-5.5%), indicating superior performance in predicting spatial response.
• The Matérn model also performed well (mean RMSE: 5.8-7.4%), outperforming most other models.
• The Exponential model exhibited the highest RMSE (9.4-15.6%), suggesting it is less suitable for this application.

Conclusions:

• The Stable variogram model is a robust choice for modeling spatial response heterogeneity in NSCLC.
• Consistent performance across different NSCLC cohorts suggests the Stable model's potential generalizability to other clinical settings.
• Further research into the Stable model could enhance personalized treatment strategies in oncology.