Prediction of Pseudoprogression versus Progression using Machine Learning Algorithm in Glioblastoma.

Area of Science:

• Neuro-oncology
• Medical Imaging
• Machine Learning

Background:

• Differentiating pseudoprogression (PsPD) from true progression (PD) is critical for glioblastoma (GBM) patient management.
• Accurate assessment impacts timely treatment adjustments, improving patient outcomes.

Purpose of the Study:

• To evaluate the feasibility of a machine learning (ML) algorithm for distinguishing PsPD from PD in GBM patients.
• To develop and validate an ML model using MRI and clinical data for improved diagnostic accuracy.

Main Methods:

• A convolutional neural network combined with a long short-term memory (CNN-LSTM) ML structure was developed.
• The model incorporated 9 axial post-contrast T1-weighted MRI images and clinical features.
• Data from 78 primary GBM patients treated with gross total resection (GTR) and concurrent chemoradiotherapy (CCRT) were used for training (N=59) and testing (N=19).

Main Results:

• The developed ML algorithm (Model 1) achieved an Area Under the Curve (AUC) of 0.83, an Area Under the Precision-Recall Curve (AUPRC) of 0.87, and an F1-score of 0.74 in the testing set.
• Model 1 demonstrated superior performance compared to models using only MRI data (Model 2) or only clinical features (Model 3).

Conclusions:

• The developed ML algorithm integrating MRI and clinical data shows high feasibility and accuracy in differentiating PsPD from PD in GBM patients.
• This tool can assist clinicians in making crucial treatment decisions during the follow-up of GBM patients undergoing GTR and CCRT.