A Multimodal Ensemble Driven by Multiobjective Optimisation to Predict Overall Survival in Non-Small-Cell Lung Cancer

Area of Science:

• Oncology
• Medical Imaging
• Machine Learning

Background:

• Lung cancer is the leading cause of cancer death globally.
• Multimodal learning offers a promising approach to integrate diverse data for improved cancer prognosis.
• Effective treatment strategies require accurate prognostic and predictive information.

Purpose of the Study:

• To predict overall survival in non-small-cell lung cancer (NSCLC) patients.
• To investigate the efficacy of multimodal learning using CT images and clinical data.
• To develop an optimized ensemble method for improved predictive performance.

Main Methods:

• Utilized the CLARO dataset comprising CT images and clinical data from NSCLC patients.
• Developed unimodal models for each data modality (CT images, clinical data).
• Implemented a late fusion approach with a multiobjective optimization problem to select the best ensemble of classifiers, maximizing recognition performance and prediction diversity.

Main Results:

• The proposed multimodal ensemble model significantly outperformed unimodal models.
• Achieved state-of-the-art results for predicting overall survival in NSCLC.
• Demonstrated the effectiveness of combining diverse data modalities for cancer prognosis.

Conclusions:

• Multimodal learning is a powerful tool for enhancing lung cancer survival prediction.
• The developed ensemble method provides a robust approach for integrating complex patient data.
• This approach has the potential to optimize NSCLC treatment strategies and improve patient outcomes.