Radiographic prediction of meningioma grade by semantic and radiomic features

• DOI: 10.1371/journal.pone.0187908
• Publication Date: 2017-11-16T20:56:46Z
• Authors: Thibaud P. Coroller; Wenya Linda Bi; Elizabeth Huynh; Malak Abedalthagafi; Ayal A. Aizer; Noah F. Greenwald; Chintan Parmar; Vivek Narayan; Winona W. Wu; Samuel Miranda de...; Saksham Gupta; Rameen Beroukhim; Patrick Y. Wen; Ossama Al-Mefty; Ian F. Dunn; Sandro Santagata; Brian M. Alexander; Raymond Y. Huang; Hugo J. W. L. Aerts
• Keywords: Adult; Male; Imaging Techniques; Science; Cancer Treatment; 610; Hemorrhage; Surgical and Invasive Medical Procedures; Pathology and Laboratory Medicine; Vascular Medicine; Diagnostic Radiology; Necrosis; Young Adult; 03 medical and health sciences; Signs and Symptoms; 0302 clinical medicine; Diagnostic Medicine; 616; Medicine and Health Sciences; Meningeal Neoplasms; Humans; Neurological Tumors; Musculoskeletal System; Skeleton; Aged; Aged, 80 and over; Radiology and Imaging; Skull; Q; R; Cancers and Neoplasms; Biology and Life Sciences; Middle Aged; Magnetic Resonance Imaging; Semantics; Signal Filtering; Oncology; Neurology; Signal Processing; Engineering and Technology; Medicine; Female; Anatomy; Meningioma; Research Article

Objectives The clinical management of meningioma is guided by tumor grade and biological behavior. Currently, the assessment follows surgical resection histopathologic review. Reliable techniques for pre-operative determination may enhance decision-making. Methods A total 175 patients (103 low-grade 72 high-grade) with contrast-enhanced T1-MRI were included. Fifteen radiomic (quantitative) 10 semantic (qualitative) features applied to quantify imaging phenotype. Area under curve (AUC) odd ratios (OR) computed multiple-hypothesis correction. Random-forest classifiers developed validated on an independent dataset (n = 44). Results Twelve radiographic (eight four semantic) significantly associated grade. High-grade tumors exhibited necrosis/hemorrhage (ORsem 6.6, AUCrad 0.62–0.68), intratumoral heterogeneity 7.9, 0.65), non-spherical shape (AUCrad 0.61), larger volumes 0.69) compared tumors. Radiomic sematic could predict (AUCsem 0.76 0.78). Furthermore, combining them increased classification power (AUCradio 0.86). Clinical variables alone did not effectively (AUCclin 0.65) or show complementary value data (AUCcomb 0.84). Conclusions We found a strong association between grade, ready application management. Combining qualitative quantitative improved power.