Machine Learning Model to Predict Pseudoprogression Versus Progression in Glioblastoma Using MRI: A...

Area of Science:

• Neuro-oncology
• Artificial Intelligence in Medicine
• Medical Imaging Analysis

Background:

• Glioblastoma patients can exhibit imaging changes post-chemoradiation, mimicking disease progression.
• Differentiating pseudoprogression (PsPD) from true progressive disease (PD) is crucial for optimal glioblastoma management.
• Previous machine learning models showed promise but required validation on larger, multi-institutional datasets.

Purpose of the Study:

• To develop and validate a robust deep learning model for predicting pseudoprogression versus progressive disease in glioblastoma.
• To improve diagnostic accuracy in challenging post-treatment imaging scenarios.
• To provide a clinically applicable tool for real-world patient care.

Main Methods:

• A multi-institutional dataset (Korean Radiation Oncology Group, KROG, N=104) was combined with a prior dataset (Seoul National University Hospital, SNUH, N=78) for a total of N=182.
• A deep learning model was developed and optimized using hyperparameter tuning on the combined dataset.
• Performance was evaluated using 10-fold cross-validation, focusing on the micro-average area under the precision-recall curve (AUPRC).
• A calibration model was implemented to provide interpretable clinical probabilities.

Main Results:

• The initial model tested on the KROG dataset showed limited performance.
• The optimized deep learning model achieved a micro-average AUPRC of 0.86 on the combined dataset (N=182).
• The integrated calibration model provides direct, interpretable probability outputs.

Conclusions:

• A refined deep learning approach effectively differentiates pseudoprogression from progressive disease in glioblastoma.
• The developed model, enhanced with calibration, offers a reliable tool for clinical decision support.
• This technology has the potential to improve treatment strategies and patient outcomes in neuro-oncology.