18th AIAI 2022, 17 - 20 June 2022, Greece

An Automated 2D U-Net Segmentation Method for the Identification of Cancer Brain Metastases Using MRI Images

Vangelis Tzardis, Efthyvoulos Kyriacou, Christos P Loizou, Anastasia Constantinidou


  In this study, we propose an automated system for the segmentation of cancer brain metastases (CBM) using MRI images. The goal is the correlation with regards to the primary cancer site. The segmentation of CBM is a challenging task due to their wide range in terms of number, shape, size and location in the brain. We experimented with the training of a modified U-Net convolutional neural network (CNN) using N=3474 brain image slices for training, Nv=579 for validation and NT=579 for testing from the public dataset BrainMetShare. The proposed model was evaluated on the testing data (NT), on a lesion-cross section basis with areas from 2.8 to 1225.7 mm2 and yielded a mean Sensitivity (SE) 0.70±0.30, Specificity (SP) 0.77±0.26 and Dice similarity coefficient (DSC) of 0.73±0.29 across the entire dataset. The present results show the good agreement of the proposed method with the ground truth.  

*** Title, author list and abstract as seen in the Camera-Ready version of the paper that was provided to Conference Committee. Small changes that may have occurred during processing by Springer may not appear in this window.