# Artificial Intelligence and Machine Learning in Bone Metastasis Management: A Narrative Review

## TL;DR

This review explores how AI and machine learning can improve the diagnosis and treatment of bone metastases by enhancing imaging analysis and decision-making for doctors.

## Contribution

The paper provides a comprehensive overview of AI/ML applications in bone metastasis management, highlighting gaps and future research directions.

## Key findings

- AI/ML tools show high internal performance in detecting and segmenting bone lesions but lack external validation.
- Emerging models for fracture risk and prognosis are at early stages and rarely integrated into clinical workflows.
- Current AI applications have limited evaluation of explainability, bias, and health-economic impacts.

## Abstract

Bone metastases are a major source of pain, pathological fracture, and loss of function in patients with advanced cancer, yet clinical decision-making still relies heavily on subjective image interpretation and empirical tools such as the Mirels scoring system. This narrative review synthesizes current work on how artificial intelligence and machine learning can support orthopedic surgeons and oncologists along the entire metastatic bone disease pathway. We describe applications that automatically detect and segment bone lesions on computed tomography, magnetic resonance imaging, and nuclear medicine studies; quantify bone quality and lesion burden; estimate fracture risk using imaging and biomechanical features; and predict survival and functional outcomes after radiotherapy or surgical reconstruction. We also highlight key limitations, including small, single-center cohorts, lack of external validation, poor integration into picture archiving and communication systems, limited interpretability, and absence of health economic evaluation. We conclude by outlining practical research priorities, such as developing transparent, prospectively validated fracture risk and prognostic models that can meaningfully improve patient selection, optimize implant choice, and reduce skeletal-related events in real-world practice.

Background: Artificial intelligence (AI) and machine learning (ML) are increasingly used in the diagnosis and management of bone metastases, spanning lesion detection, segmentation, prognostic modeling, fracture risk assessment, and surgical decision support. However, the literature is heterogeneous and rapidly evolving, making it difficult for clinicians to contextualize these developments. Methods: We performed a narrative review of the literature on AI/ML applications in bone metastasis management, focusing on studies that address clinically relevant problems such as detection and segmentation of metastatic lesions, prediction of skeletal-related events and survival, and support for reconstructive decision-making. We prioritized recent, peer-reviewed work that reports model performance and highlights opportunities for clinical translation. Results: Most published studies center on imaging-based diagnosis and lesion segmentation using radiomics and deep learning, with generally high internal performance but limited external validation. Emerging work explores prognostic models and biomechanically informed fracture risk estimation, yet these remain at an early proof-of-concept stage. Very few frameworks are integrated into routine workflows, and explainability, bias mitigation, and health-economic impacts are rarely evaluated. Conclusions: AI and ML tools have substantial potential to standardize imaging assessment, refine risk stratification, and ultimately support personalized management of bone metastases. Future research should focus on externally validated, multimodal models; development of AI-augmented alternatives to the Mirels score; federated multicenter collaboration; and routine incorporation of explainability and cost-effectiveness analyses.