Application of Radiomics in Meningioma Imaging: Global Trends and Needs

Introduction: Quantitative analyses of imaging features through radiomics is a powerful new technology to noninvasively assess biological and molecular correlates, provide prognostic information, and guide clinical decision-making. There has been growing interest in applying image-based phenotyping to meningiomas, a tumor with nuances in diagnosis, risk stratification, and operative management. Although studies on radiomics for brain tumors are rapidly increasing, their clinical applicability and rigor remain varied.

Objective: To characterize the diverse applications for radiomics to meningiomas, analyze radiomic model development and validation, quantify performance of radiomic approaches, and assess study quality in radiomics.

Methods: We systematically reviewed all radiomics analyses of meningiomas published until 12/20/2021 and extracted qualitative and quantitative data for each. We assessed study aims, radiomic applications, methodologies, strengths, limitations, and quality of publications, using the Radiomics Quality Score (RQS) as a metric for quality.

Results: A total of 170 eligible publications were divided into five categories of radiomics application to meningiomas: tumor detection and segmentation (21%), classification across tumor subtypes (54%), grading (14%), feature correlation (3%), and prognostication (8%). There has been exponential growth in meningioma radiomics studies globally, with the highest number of publications coming from Asia (128), followed by the United States (12). A majority of these studies had a technical focus on computational model development (73%) versus those focused more on clinical applications (27%). Studies dichotomized between private institutional (50%) and public (49%) meningioma image datasets, but rarely used both sources (1%). Radiomic models largely used MRIs (99%), most frequently drawing on T1-weighted sequences (64%). Only 10% of studies used 3–4 MRI sequence types concurrently to develop models. Deep learning models have surpassed supervised and unsupervised machine learning approaches for processing radiomic data, driven by deep learning development on public datasets in technically oriented studies. For detection and segmentation of meningiomas, radiomic models had a mean accuracy of 93.1 ± 8.1% with a dice coefficient of 88.8 ± 7.9%. Classification of meningiomas had a mean accuracy of 95.2 ± 4.0% with significant better performance in technical (95.7 ± 3.7%) versus clinical (89.8 ± 3.2%) studies. Grading of meningiomas had a mean AUC of 0.85 ± 0.08. Feature correlation including meningioma firmness, tumor fibrous quality, and K_i-67 proliferative index had a mean AUC of 0.89 ± 0.07. Prognostication of meningioma clinical course had a mean AUC of 0.83 ± 0.08. Despite the rapid rise in studies, quality remains low, with a mean RQS of 6.7 ± 5.9 (out of range −8 to 36) across studies, equivalent to 33% of the total possible RQS score. Clinical studies had better mean RQS (8.3 ± 6.6) compared with technical studies (6.1 ± 5.4). The greatest limitations in study quality were in the domains of model development, prospective validation, and clinical utility assessment.

Conclusion: Radiomics for meningioma is growing globally, with an exponential rise in the last few years driven by expanding computational bandwidth, evolution of deep learning methodology, publicly accessible databases, and improving model performance. While meningioma detection, segmentation, and classification have been broadly studied, translatability toward more complex tasks such as prognostication requires studies that improve quality, develop comprehensive patient datasets, and engage in prospective multi-institutional assessment.

Publication History

Article published online:
01 February 2023