3D Brain Tumor Volume Reconstruction and Quantification using MRI Multi-modalities Brain Images

Brain tumor volume quantification is not possible with magnetic resonance imaging (MRI) non-invasive imaging systems. Usually, the brain MR imaging modality is based on four modalities T1, T1ce (contrast-enhanced), T2, and FLAIR (Fluid-Attenuated Inversion Recovery) highlighting various complex brain structure. The aim of this study is to develop an automated brain tumor segmentation method to allow an effective 3D tumor reconstruction and volume quantification. The Local Graph Cut (LGC) and Flood fill (FF) segmentation tools are used. A 3D reconstruction is performed using the open-source ImageJ image processing package. To assess the brain tumor segmentation accuracy a 3D Dice score is used. This extensive study is conducted on the MICCAI BraTS2020 dataset and shows that the proposed model obtains good results. To evaluate the final full segmentation of 3D volume accuracy, we use three brain modalities, namely T1 and T2 weighted, and T1ce scans. Also, to measure accuracy of segmentation we take the advantage of the existing ground truth image dataset. We find an average Dice score of 0.501 ± 0.128 for T2-weighted images and LGC algorithm. Also, there are not important differences between the results provided by investigated algorithms for all images types. The 3D reconstruction and volumetric analysis via 2D segmented slices produce realistic shape samples. The computed volumes are in the limit of 71% from the tumor volume computed using the ground truth images.