Differentiating Gliosarcoma from Glioblastoma: A Novel Approach Using PEACE and XGBoost to Deal with Datasets with Ultra-High Dimensional Confounders

Amir Saki ¹, Usef Faghihi ¹, Ismaila Baldé ²

⁰Département de Mathématiques et d'Informatique, Université du Québec à Trois-Rivières, Trois-Rivières, QC G8Z 4M3, Canada.

Life (basel, Switzerland) +

|

July 27, 2024

View abstract on PubMed

Summary

This study introduces a new causal method combining Probabilistic Easy Variational Causal Effect (PEACE) with XGBoost to differentiate gliosarcoma (GSM) from glioblastoma (GBM). The approach achieved 83% accuracy, offering a novel way to analyze tumor types using radiomic features.

Area Of Science

Neuro-oncology
Medical imaging analysis
Machine learning in medicine

Background

Distinguishing gliosarcoma (GSM) from glioblastoma (GBM) is clinically significant.
Traditional methods often involve data reduction before causal analysis, potentially losing information.
Radiomic features hold potential for non-invasive tumor characterization.

Purpose Of The Study

To develop and validate a novel causal methodology for differentiating GSM from GBM.
To leverage the Probabilistic Easy Variational Causal Effect (PEACE) metric combined with XGBoost.
To analyze the direct causal effects of radiomic features on tumor type classification.

Main Methods

Utilized a recently developed causal methodology: Probabilistic Easy Variational Causal Effect (PEACE).
Integrated PEACE with the eXtreme Gradient Boosting (XGBoost) algorithm for causal analysis.
Applied the method to the complete dataset without prior dimension reduction, analyzing direct causal effects of radiomic features on tumor type (GSM/GBM).
Incorporated a degree 'd' in PEACE variations (0-1) to weigh event rarity and frequency.

Main Results

Achieved a mean accuracy of 83% in differentiating GSM from GBM using the PEACE-XGBoost model.
Obtained an average Mean Squared Error (MSE) of 0.130.
Demonstrated the model's effectiveness despite a high number of features (columns) and a low number of samples (rows).

Conclusions

The PEACE-XGBoost approach provides a nuanced understanding of causal relationships in radiomic data for tumor differentiation.
This causal methodology offers an effective alternative to traditional statistical models for high-dimensional medical data.
The method facilitates accurate classification between GSM and GBM, with potential for broader applications in medical diagnostics.

Keywords:

Probabilistic Easy Variational Causal Effect (PEACE) XGBoost causal inference glioblastoma gliosarcoma