From discordance to clinical decision support: integrating longitudinal imaging and ctDNA through multimodal artificial intelligence in precision oncology
- 1Department of Oncology, Zhejiang Provincial Hospital of Chinese Medicine, The First Affiliated Hospital of Zhejiang Chinese Medical University
Abstract
Accurate treatment-response assessment increasingly requires information that is spatial, molecular, and longitudinal. Imaging depicts lesion burden, anatomical distribution, and morphological evolution, whereas circulating tumor DNA (ctDNA) provides a systemic molecular signal shaped by tumor shedding, clonal composition, assay characteristics, and treatment pressure. Because these modalities measure non-identical dimensions, disagreement between them is clinically plausible and can reflect lead-lag kinetics, site-specific progression, clonal evolution, or technical uncertainty. In this focused narrative review, we synthesize evidence for imaging-ctDNA integration and propose a multidimensional, multilabel taxonomy of discordance spanning directional, temporal, spatial, molecular, and technical dimensions. We distinguish isolated from persistent discordance and emphasize exact temporal alignment, analytical confidence, and orthogonal confirmation. Current evidence supports prognostic complementarity and demonstrates measurable discordance across tumor types, but clinical actionability is context dependent. Randomized studies such as PADA-1 show that molecular-first treatment adaptation can be beneficial when a defined biomarker is linked to a validated intervention, whereas other trials illustrate that molecular detectability alone does not guarantee benefit from early treatment. We therefore frame multimodal artificial intelligence (AI) as an uncertainty-aware decision-support approach rather than an automated treatment selector. Future systems should preserve modality-specific information, handle irregular longitudinal sampling, quantify uncertainty, identify information gaps, and recommend evidence-matched reassessment. Prospective validation should prioritize clinical utility, harms, transportability, governance, and net benefit rather than discrimination alone.
Related
- Toward Precision Colorectal Cancer Screening: Integrating Clinical Risk Stratification, Molecular Biomarkers, and Artificial Intelligence
- Toward integrated multimodal artificial intelligence in bone marrow examination: current state and open challenges
- Artificial Intelligence–Driven Multimodal Imaging for Cancer During Pregnancy: Advances in Maternal–Fetal Diagnostics and Precision Oncology
- Artificial Intelligence, Multi-Omics, and Multimodal Integration in Precision Radiotherapy for Head and Neck Cancer: From Biological Insights to Treatment Decisions
- A conceptual framework integrating artificial intelligence with failure mode and effects analysis for diagnostic error reduction in anatomical pathology
BibTeX
@article{zheng2026from,
title = {{From discordance to clinical decision support: integrating longitudinal imaging and ctDNA through multimodal artificial intelligence in precision oncology}},
author = {Jian Zheng and Luyang Chi and Yong Guo},
journal = {Frontiers in Oncology},
year = {2026},
month = oct,
doi = {10.3389/fonc.2026.1963645},
url = {https://www.frontiersin.org/articles/10.3389/fonc.2026.1963645/full},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - From discordance to clinical decision support: integrating longitudinal imaging and ctDNA through multimodal artificial intelligence in precision oncology
AU - Jian Zheng
AU - Luyang Chi
AU - Yong Guo
PY - 2026
DA - 2026/10/27
JO - Frontiers in Oncology
DO - 10.3389/fonc.2026.1963645
UR - https://www.frontiersin.org/articles/10.3389/fonc.2026.1963645/full
AB - Accurate treatment-response assessment increasingly requires information that is spatial, molecular, and longitudinal. Imaging depicts lesion burden, anatomical distribution, and morphological evolution, whereas circulating tumor DNA (ctDNA) provides a systemic molecular signal shaped by tumor shedding, clonal composition, assay characteristics, and treatment pressure. Because these modalities measure non-identical dimensions, disagreement between them is clinically plausible and can reflect lead-lag kinetics, site-specific progression, clonal evolution, or technical uncertainty. In this focused narrative review, we synthesize evidence for imaging-ctDNA integration and propose a multidimensional, multilabel taxonomy of discordance spanning directional, temporal, spatial, molecular, and technical dimensions. We distinguish isolated from persistent discordance and emphasize exact temporal alignment, analytical confidence, and orthogonal confirmation. Current evidence supports prognostic complementarity and demonstrates measurable discordance across tumor types, but clinical actionability is context dependent. Randomized studies such as PADA-1 show that molecular-first treatment adaptation can be beneficial when a defined biomarker is linked to a validated intervention, whereas other trials illustrate that molecular detectability alone does not guarantee benefit from early treatment. We therefore frame multimodal artificial intelligence (AI) as an uncertainty-aware decision-support approach rather than an automated treatment selector. Future systems should preserve modality-specific information, handle irregular longitudinal sampling, quantify uncertainty, identify information gaps, and recommend evidence-matched reassessment. Prospective validation should prioritize clinical utility, harms, transportability, governance, and net benefit rather than discrimination alone.
SN - 2234-943X
ER -