CAUSAL FOUNDATION MODELS FOR ONCOLOGY: A TEMPORAL MULTI-MODAL FRAMEWORK FOR COUNTERFACTUAL PROGNOSIS AND TREATMENT RESPONSE IN GLIOBLASTOMA

Read full text Summarise with VIRAT
Authors: RAMAKRISHNA KOLIKIPOGU, Dr. K.V.VIVEKANANDA, CHOPPA.ANANDA KUMAR REDDY, THANU KURIAN, AMIT VERMA, Dr.R.SENTHAMIL SELVAN
Cited by: 0

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Modern clinical AI systems are dominated by correlative models that predict outcomes from observed patterns but typically fail to provide reliable answers to interventional “what-if” questions required for treatment selection and policy evaluation. Such limitations reduce robustness when clinical practice, treatments, or patient subgroups shift. The study develops a causal multi-modal foundation model to estimate prognosis and treatment response for glioblastoma multiforme (GBM), enabling counterfactual reasoning over longitudinal treatment timelines. The present study compares three model classes: (i) Baseline - a standard multi-modal transformer using pre-trained encoders (UMME); (ii) Advanced - a causal multi-modal foundation model that injects structural causal constraints and causal regularization during training; and (iii) Proposed - a Temporal Causal Multi-Modal Transformer (TCMMT) that models temporal treatment patterns, integrates dynamic multi-modal diagnostics, and contains a counterfactual head for intervention simulation. Training data were drawn from TCGA-GBM and longitudinal institutional cohorts with curated treatment timelines and censoring-aware labels. The proposed TCMMT achieved a concordance index (C-index) of 0.87 for survival prediction and improved counterfactual treatment response prediction accuracy by 23% relative to non-causal baselines (relative improvement). By combining causal graph structure, temporal encoding of treatment histories, and multi-modal foundation encoders, TCMMT supports in-silico clinical trials and personalized treatment optimization, marking a shift from purely predictive to prescriptive AI in oncology.

Fields of Study & Subjects

Machine Learning in Healthcare Glioma Diagnosis and Treatment Radiomics and Machine Learning in Medical Imaging

Publication details

Journal / source
Journal of Theoretical and Applied Information Technology
Publication date
15 March 2026
Language
EN