Probabilistic approach to longitudinal response prediction: application to radiomics from brain cancer imaging

2025-05-12

AI Summary

To address the challenge of longitudinal prediction of disease progression in brain cancer patients, this paper proposes a probabilistic modeling framework that integrates baseline radiomic features with follow-up temporal information. Methodologically, it introduces, for the first time, a full-probability Bayesian state-space model into radiomic longitudinal analysis, enabling uncertainty propagation to quantify predictive confidence naturally and supporting robust forecasting even under missing intermediate follow-up data. Key contributions include: (1) eliminating reliance on dense longitudinal sampling; (2) balancing high-dimensional feature interpretability with controllable model dimensionality; and (3) establishing a synthetic-data validation framework to ensure generalizability. Evaluated on real-world brain cancer imaging data, the method achieves predictive performance comparable to state-of-the-art models while significantly enhancing clinicians’ ability to assess prediction reliability—critical for informed decision-making.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Machine Learning: Probabilistic Circuits and Graphical Models
• Intelligent Robots: State Estimation

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Longitudinal imaging analysis tracks disease progression and treatment response over time, providing dynamic insights into treatment efficacy and disease evolution. Radiomic features extracted from medical imaging can support the study of disease progression and facilitate longitudinal prediction of clinical outcomes. This study presents a probabilistic model for longitudinal response prediction, integrating baseline features with intermediate follow-ups. The probabilistic nature of the model naturally allows to handle the instrinsic uncertainty of the longitudinal prediction of disease progression. We evaluate the proposed model against state-of-the-art disease progression models in both a synthetic scenario and using a brain cancer dataset. Results demonstrate that the approach is competitive against existing methods while uniquely accounting for uncertainty and controlling the growth of problem dimensionality, eliminating the need for data from intermediate follow-ups.

Problem

• Predicts brain cancer progression using probabilistic radiomics modeling
• Integrates baseline and follow-up data for dynamic outcome prediction
• Handles uncertainty and reduces dimensionality in longitudinal analysis

Innovation

• Probabilistic model for longitudinal response prediction
• Integrates baseline and follow-up radiomic features
• Handles uncertainty and controls dimensionality growth