Applications and Performance of Artificial Intelligence in Spinal Metastasis Imaging: A Systematic Review

Area of Science:

• Oncology
• Radiology
• Medical Imaging
• Artificial Intelligence

Background:

• Spinal metastasis is a frequent complication of cancer, posing diagnostic challenges.
• Current manual detection methods are costly and inefficient.
• Artificial intelligence (AI) and computer-aided detection (CAD) offer potential improvements in oncological imaging.

Purpose of the Study:

• To review and evaluate the current applications of AI in diagnosing spinal metastasis across various imaging techniques.
• To synthesize findings from existing literature on AI's role in spinal metastasis detection and characterization.

Main Methods:

• A systematic review and meta-analysis of studies published between 2007 and 2024.
• Searched major databases (PubMed, Scopus, Web of Science, Cochrane, Embase) using relevant keywords.
• Included 39 studies (6267 patients), analyzing AI model performance metrics (AUC, accuracy, sensitivity, specificity) and risk of bias.

Main Results:

• AI models achieved high weighted average AUCs: 0.971 (training), 0.947 (internal validation), and 0.819 (external validation).
• Lung, breast, and prostate cancers were the most common primary tumors studied.
• A significant risk of bias (56%) was identified, mainly due to insufficient external validation and overfitting.

Conclusions:

• AI demonstrates significant potential for improving spinal metastatic lesion detection, segmentation, and characterization.
• Future research must prioritize developing generalizable AI models using larger, diverse datasets and prospective validation.
• Integrating clinical and imaging data is crucial for demonstrating AI's clinical utility in spinal metastasis management.