Image Analysis

Using our training dataset (n = 91 patients), image analysis was performed by a commercially available FDA-approved software (Olea Sphere software, Olea Medical SAS). Automatic preprocessing was standardized for each case involving intensity normalization, resampling, and discretization. Since MR images were obtained using different MRI scanners from 2 vendors and with different magnetic fields, a normalization step was implemented to normalize images by centering at the mean with standard deviation using all gray values in the image. The resampling grid was aligned to the input origin enabling in-plane resampling. Size and number of bins was set to 25 and 64, respectively, for every case standardizing the process of making histogram and discretion of the image gray level. T1c+, FLAIR, and diffusion images (ADC/b1000) were coregistered on each examination using a 6-df transformation and a mutual information cost function.

Tumor segmentation was performed manually on every slice that the tumor was visualized using FLAIR images. This was performed by a trained radiologist and under supervision of a board certified neuroradiologist. Subsequently, a VOI was generated encompassing the entire region of FLAIR hyperintensity and overlaid onto coregistered T1c+ and diffusion datasets for radiomic texture analysis (Supplementary Figure 2).

A total of 92 radiomic features were assessed. These included 19 first-order metrics, such as the mean, standard deviation, skewness, and kurtosis, and second-order metrics including 23 gray level run length matrix (GLCM), 16 gray level run length matrix (GLRLM), 15 gray level size zone matrix (GLSZM), 5 neighboring gray tone difference matrix (NGTDM), and 14 gray level dependence matrix (GLDM). Details of the definitions and calculations of these features have previously been reported.^19–23 Texture feature extraction through Olea sphere software was in compliance with the image biomarker standardization initiative with the above 92 features categorized into (1) histogram features, which included grey intensity or brightness information of the lesion, (2) form factor features, which describe the shape and compactness of the lesions, and (3) texture features, which includes the remainder of the second-order metrics as above, and has been cited in prior studies.^24,25