Multimodal Omics-Integrated Prognosis Prediction for MRI-Guided Adaptive Radiotherapy In Gliomas
Abstract
Purpose
To develop a multi-omics data integration method based on ensemble learning to construct a prognostic decision model for MRI-guided radiotherapy in gliomas, providing reliable reference for clinical decision-making.
Methods
A retrospective collection of multi-omics data from 64 patients with high-grade gliomas treated using Unity was conducted. Senior radiation oncologists re-contoured the gross tumor volume (GTV) and regions of interest (ROIs) on each fraction MRI image for all patients. A subsequent statistical analysis examined GTV changes across fractions throughout treatment. The manual radiomics feature extraction process involved the calculation of delta radiomics features for both inter-fraction MRI images and dose distribution. Multiple machine learning methodologies were implemented to develop recurrence site prediction models, and the accuracy of these models was subsequently evaluated. Based on ensemble learning principles, a decision model was constructed by combining the predictions of optimal machine learning models for each modal feature through weighted voting or analogous methodologies. Model performance was evaluated on an independent validation set.
Results
As treatment progressed in fractions, the average gross tumor volume (GTV) in glioma patients underwent a gradual decrease, while the average GTV displacement exhibited a progressive increase. The implementation of synchronous electric field therapy resulted in amplified alterations in both the GTV volume and its spatial location. Regarding recurrence site prediction, the multi-omics integrated predictive model demonstrated higher accuracy (accuracy: 0.83), significantly outperforming both the single-dose radiomics model (0.75) and the single-MRI radiomics model (AUC: 0.79). These findings were statistically significant (p<0.005).
Conclusion
This study developed a multi-omics integrated prognostic prediction model for glioma based on ensemble learning, thereby significantly enhancing the model's predictive performance. This finding suggests that the proposed method has the potential to enhance the efficacy of glioma treatment strategies and facilitate the implementation of precision therapy that is tailored to the individual patient's needs.