Development and Validation of Survival Prediction Models for Patients With Pineoblastomas Using Deep Learning: A...

Area of Science:

• Neuro-oncology
• Computational biology
• Medical data science

Background:

• Pineoblastomas (PBs) are rare pediatric central nervous system tumors with limited prognostic data.
• Existing prediction models for PB survival outcomes are lacking.
• Accurate prognosis is crucial for guiding treatment and managing patient expectations.

Purpose of the Study:

• To develop and validate deep learning (DL) models for predicting 3-year overall survival (OS) and disease-specific survival (DSS) in patients with pineoblastoma.
• To compare the predictive performance of DL models against traditional Cox proportional hazard (CPH) models.

Main Methods:

• Patients diagnosed with pineoblastoma were identified from the Surveillance, Epidemiology, and End Results (SEER) database (1975-2019).
• Deep neural networks (DNN) were trained and tested using 5-fold cross-validation.
• Multivariate CPH models were constructed for comparative analysis. Model performance was assessed using ROC curve analysis and calibration plots.

Main Results:

• The study included 145 patients with pineoblastoma.
• DNN models achieved high predictive accuracy for 3-year OS (AUC=0.92) and DSS (AUC=0.76).
• The developed DNN models demonstrated good calibration for both OS and DSS predictions.

Conclusions:

• Deep learning models significantly enhance the prediction accuracy of survival outcomes in pineoblastoma patients compared to CPH models.
• These findings highlight the potential of DL algorithms to improve prognostic capabilities for rare tumor types.
• The developed DL models can aid in better outcome prediction and clinical decision-making for pineoblastoma.