Multimodal imaging and artificial intelligence for bone tumor diagnosis: Comparative performance and emerging technologies
- 1Department of Radiology, Endocrine and Metabolic Diseases Hospital of Shandong First Medical University, Jinan, Shandong 250000, China
- 2Department of Radiology, Shandong Cancer Hospital and Institute, Shandong First Medical University and Shandong Academy of Medical Sciences, Jinan, Shandong 250117, China
Abstract
Multimodal imaging is central to the diagnosis, staging, and management of bone tumors. Radiography and computed tomography remain essential for initial assessment and cortical evaluation, magnetic resonance imaging provides superior soft tissue contrast and local staging accuracy, and positron emission tomography enhances detection of metastatic disease and treatment response. No single technique suffices across clinical scenarios. Advances in artificial intelligence and radiomics offer avenues to augment this framework, with demonstrated potential in lesion classification, segmentation, prognostication, and response prediction. This review advances a specific argument rather than surveying available technologies: that the principal constraint on artificial intelligence in musculoskeletal oncology is the stability of the reference standard rather than algorithmic capability. Ground truth in bone tumors is unstable on four grounds. Core needle biopsy samples a limited volume and may miss the highest-grade component: agreement between biopsy and resection grade in central chondrosarcoma falls from a weighted kappa of 0.63 in long bones to 0.12 in the pelvis. Expert interpretation is itself inconsistent, with interobserver kappa for grading cartilaginous tumors of 0.44 among musculoskeletal pathologists and 0.35 among radiologists. Molecular criteria have redefined entities within the past decade, so older datasets contain classes that no longer exist as defined. And histological classification does not reliably predict clinical behavior. Consequently, measurable accuracy has a ceiling set by the reference standard rather than the algorithm, and apparent failure on external validation may reflect label divergence between institutions rather than model deficiency. Evidence is graded throughout, and clinical readiness distinguished from technical performance. No system for primary bone tumor characterization has regulatory clearance in the United States or Europe, and formal appraisal of this literature confirms low methodological compliance and infrequent external validation. Progress will depend on multicenter validation and protocol harmonization, and, distinctively for this field, on composite reference standards, outcome-based endpoints, and explicit uncertainty modelling, since ground-truth instability can be mitigated but not eliminated.
Related
- Multimodal artificial intelligence assessment of simulated cardiac arrest performance: a pilot study
- Vulnerable Coronary Plaques on CT Angiography: Imaging Biomarkers, Artificial Intelligence, and Emerging Technologies for Risk Stratification
- Artificial Intelligence and AI-Adjacent Digital Technologies in Nursing Practice Within Gastroenterology: A Scoping Review of Current Applications, Challenges, and Future Directions.
- 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
BibTeX
@article{liu2026multimodal,
title = {{Multimodal imaging and artificial intelligence for bone tumor diagnosis: Comparative performance and emerging technologies}},
author = {Xiaowen Liu and Lingfei Liu},
journal = {Journal of Bone Oncology},
year = {2026},
month = oct,
doi = {10.1016/j.jbo.2026.100808},
url = {https://doi.org/10.1016/j.jbo.2026.100808},
note = {CC BY-NC-ND 4.0},
}
RIS
TY - JOUR
TI - Multimodal imaging and artificial intelligence for bone tumor diagnosis: Comparative performance and emerging technologies
AU - Xiaowen Liu
AU - Lingfei Liu
PY - 2026
DA - 2026/10/23
JO - Journal of Bone Oncology
DO - 10.1016/j.jbo.2026.100808
UR - https://doi.org/10.1016/j.jbo.2026.100808
AB - Multimodal imaging is central to the diagnosis, staging, and management of bone tumors. Radiography and computed tomography remain essential for initial assessment and cortical evaluation, magnetic resonance imaging provides superior soft tissue contrast and local staging accuracy, and positron emission tomography enhances detection of metastatic disease and treatment response. No single technique suffices across clinical scenarios. Advances in artificial intelligence and radiomics offer avenues to augment this framework, with demonstrated potential in lesion classification, segmentation, prognostication, and response prediction. This review advances a specific argument rather than surveying available technologies: that the principal constraint on artificial intelligence in musculoskeletal oncology is the stability of the reference standard rather than algorithmic capability. Ground truth in bone tumors is unstable on four grounds. Core needle biopsy samples a limited volume and may miss the highest-grade component: agreement between biopsy and resection grade in central chondrosarcoma falls from a weighted kappa of 0.63 in long bones to 0.12 in the pelvis. Expert interpretation is itself inconsistent, with interobserver kappa for grading cartilaginous tumors of 0.44 among musculoskeletal pathologists and 0.35 among radiologists. Molecular criteria have redefined entities within the past decade, so older datasets contain classes that no longer exist as defined. And histological classification does not reliably predict clinical behavior. Consequently, measurable accuracy has a ceiling set by the reference standard rather than the algorithm, and apparent failure on external validation may reflect label divergence between institutions rather than model deficiency. Evidence is graded throughout, and clinical readiness distinguished from technical performance. No system for primary bone tumor characterization has regulatory clearance in the United States or Europe, and formal appraisal of this literature confirms low methodological compliance and infrequent external validation. Progress will depend on multicenter validation and protocol harmonization, and, distinctively for this field, on composite reference standards, outcome-based endpoints, and explicit uncertainty modelling, since ground-truth instability can be mitigated but not eliminated.
SN - 2212-1366
SN - 2212-1374
ER -