Artificial intelligence-driven radiological biomarkers: A narrative review of artificial intelligence in meningioma diagnosis

Antonio Navarro-Ballester

Radiology Department, Neuroradiology Section, Hospital General Universitari de Castelló, Castellón de la Plana, Castellón, Spain

Received 30 October 2024, Revised 18 November 2024, Accepted 24 November 2024, Available online 10 December 2024, Version of Record 10 January 2025.

https://doi.org/10.1016/j.neumar.2024.100033

Abstract

Meningiomas, the most common primary intracranial tumors, present unique challenges in diagnosis and management due to their diverse behaviors and potential for recurrence. Recent advances in artificial intelligence, particularly machine learning and deep learning, are transforming the way radiologists detect, classify, and predict outcomes for these tumors. Artificial intelligence-driven techniques enable precise extraction of imaging biomarkers from magnetic resonance imaging and computed tomography scans, enhancing diagnostic accuracy by distinguishing subtle tumor characteristics not readily visible to the human eye. Hybrid models that integrate multiple imaging modalities, such as T1-weighted and contrast-enhanced magnetic resonance imaging, further improve assessment accuracy by providing a comprehensive view of tumor heterogeneity and invasiveness. Moreover, radiomics, an artificial intelligence application that analyzes quantitative imaging features, has shown promise in predicting meningioma grade and recurrence risk, supporting personalized treatment strategies. Another area of progress involves multi-habitat analysis, which examines tumor heterogeneity within various “habitats” in a single tumor. This approach, combined with unsupervised machine learning algorithms, enhances the ability to differentiate low-grade from high-grade meningiomas by capturing differences in cellular structure. Artificial intelligence models incorporating clinical and molecular data, such as the Ki-67 proliferation index, are advancing recurrence prediction, which is critical for optimizing postoperative follow-up. Predictive models based on artificial intelligence are also being developed to assess quality of life outcomes, guiding supportive care in ways that optimize resources while addressing specific patient needs. Despite these advancements, several challenges remain, including variability in imaging protocols and the need for large, annotated datasets to ensure model generalizability. Future research should focus on developing robust algorithms capable of handling diverse data sources, improving the integration of functional and anatomical imaging modalities, and enhancing tumor heterogeneity analysis. These steps will bring artificial intelligence closer to routine clinical implementation, ultimately contributing to better patient outcomes through tailored, data-driven meningioma management. In summary, artificial intelligence-driven radiological biomarkers, machine learning models for tumor grading, multi-modality imaging, and the ability to predict recurrence and quality of life represent transformative developments in the field of meningioma diagnostics.

人工智能驱动的放射生物标记：人工智能在脑膜瘤诊断中的应用综述

摘要

脑膜瘤是最常见的原发性颅内肿瘤，由于其行为多样且可能复发，给诊断和管理带来了独特的挑战。人工智能领域的最新进展，尤其是机器学习和深度学习，正在改变放射科医生检测、分类和预测这些肿瘤预后的方式。人工智能驱动的技术能够从磁共振成像和计算机断层扫描中精确提取成像生物标志物，通过区分人眼不易察觉的微妙肿瘤特征来提高诊断准确性。整合了多种成像模式（如 T1 加权和对比增强磁共振成像）的混合模型可全面了解肿瘤的异质性和侵袭性，从而进一步提高评估的准确性。此外，放射组学是一种分析定量成像特征的人工智能应用，在预测脑膜瘤分级和复发风险、支持个性化治疗策略方面大有可为。另一个取得进展的领域涉及多生境分析，即研究单个肿瘤中不同 “生境 ”内的肿瘤异质性。这种方法与无监督机器学习算法相结合，通过捕捉细胞结构的差异，提高了区分低级别和高级别脑膜瘤的能力。结合临床和分子数据（如 Ki-67 增殖指数）的人工智能模型正在推进复发预测，这对优化术后随访至关重要。目前还在开发基于人工智能的预测模型，用于评估生活质量结果，以优化资源的方式指导支持性护理，同时满足患者的特定需求。尽管取得了这些进展，但仍存在一些挑战，包括成像方案的多变性，以及需要大型、有注释的数据集来确保模型的通用性。未来的研究应侧重于开发能够处理不同数据源的强大算法，改进功能和解剖成像模式的整合，以及加强肿瘤异质性分析。这些步骤将使人工智能更接近常规临床应用，最终通过量身定制、数据驱动的脑膜瘤管理，为改善患者预后做出贡献。总之，人工智能驱动的放射学生物标志物、用于肿瘤分级的机器学习模型、多模态成像以及预测复发和生活质量的能力代表了脑膜瘤诊断领域的变革性发展。此综述旨在全面概述当前人工智能在脑膜瘤分类中的应用进展，特别关注放射生物标志物的识别和利用，并强调人工智能在提高脑膜瘤诊断准确性和治疗策略方面的潜力。