Systematic Review & Meta-Analysis Protocol

Diagnostic Accuracy of Artificial Intelligence Using MRI for Differentiating Benign and Malignant Musculoskeletal Tumors

A Systematic Review and Meta-Analysis

PROSPERO ID 342284 Registered 16 March 2026 Version 1.0 Guidelines PRISMA-DTA 2023

Background

Accurate differentiation between benign and malignant musculoskeletal tumors is critical for treatment planning, yet MRI interpretation in this domain remains challenging due to lesion rarity, morphological overlap, and significant inter-reader variability. Artificial intelligence — encompassing deep learning, convolutional neural networks, vision transformers, and radiomics — has emerged as a promising tool to standardize and augment radiological assessment.

Despite a rapidly expanding literature, no comprehensive synthesis has evaluated AI performance across all three computational tasks (detection, segmentation, and classification) within a single methodological framework. This protocol describes a systematic review and meta-analysis designed to address that gap.

AI MSK Tumor Classification MRI Analysis

Study at a Glance

🦴

Population

Patients with suspected or confirmed bone and soft tissue musculoskeletal tumors imaged with MRI

🤖

Index Test

Any AI / ML / DL model (CNN, transformer, radiomics, ensemble) applied to MRI for detection, segmentation, or classification

🔬

Reference Standard

Histopathology, radiologist consensus, or manual segmentation ground truth

📊

Primary Outcomes

AUC, sensitivity, specificity (classification); Dice coefficient (segmentation); detection rate (detection)

Objectives

  1. Systematically evaluate and meta-analyze the diagnostic accuracy of AI models applied to MRI for differentiating benign from malignant musculoskeletal tumors across detection, segmentation, and classification tasks.
  2. Compare performance across AI architectures: CNN, vision transformer, radiomics pipelines, and ensemble models.
  3. Quantify the impact of validation strategy (internal vs. external) on reported accuracy metrics.
  4. Assess methodological quality and risk of bias using the QUADAS-AI framework.
  5. Identify evidence gaps and provide design recommendations for future AI studies in musculoskeletal oncology imaging.

Methods

Random-Effects Meta-Analysis Bivariate Model (Reitsma) DerSimonian-Laird SROC Curves QUADAS-AI Risk of Bias GRADE-DTA Meta-Regression Deeks' Funnel Plot 6 Databases PubMed · Embase · Web of Science · IEEE · Cochrane · Scopus

PROSPERO Registration

CRD420261342284

Diagnostic Performance of Artificial Intelligence Using MRI for Detection, Segmentation, and Classification of Musculoskeletal Tumors: A Systematic Review and Meta-Analysis

Active AI · MRI · OncologyPRISMA-DTA 2023
PROSPERO