The prognostic value of Cone Beam CT (CBCT) in brain metastasis patients using Radiomics and Machine learning.
Description
this data includes radiomics features extracted from CBCT (original and resampled), merged MRI and CBCT, and MRI data. that was used to assess the performance of CBCT in brain metastasis and compare finding to MRI model.
Files
Steps to reproduce
The regions of interest (ROIs) were manually segmented by an experienced neuro-surgeon slice by slice on the contrast enhanced T1-weighted (CE-T1W) images using the gamma knife planning system. These CE-T1W images were then co-registered with the CBCT images in the treatment planning system, and the segmented ROIs were then superimposed onto the CBCT. The DICOM data and ROI segmentation were uploaded to LifeX radiomics software for feature extraction. Only tumors with a voxel count greater than 64 were included, resulting in the selection of 103 tumors out of the 145 tumors from 27 patients. Feature extraction was performed follows. 1. We used the default bin number in LifeX (64) with mean relative ROI ±3 SD intensity rescaling in CE-T1W MR images, without applying any filter. The image were resampled to 1 x 1 x 1 mm3 to harmonise the voxel size based on Image Biomarker Standardisation Initiative (IBSI) recommendations (20). 2. The original CBCT data, with a voxel size of 0.5 x 0.5 x 0.5 mm, was used with the default bin number (64) and mean relative ROI ±3 SD intensity rescaling. 3. The CBCT data was also resampled to the voxel size of 1 x 1 x 1 mm3 consistent with the MRI data, with the default bin number (64) and mean relative ROI ±3 SD intensity rescaling. 4. The CBCT was superimposed on the MRI images and manually merged. Radiomic features were extracted from the merged MRI/CBCT data, the image was resampled to a voxel size of 1 x 1 x 1 mm3, with the default bin number (64) and mean relative ROI ±3 SD intensity rescaling.