Tumor Segmentation in Brain MRI using Fully Convolutional Network

Le résumé fourni par la source

In biomedical image processing and cancer studies tumor segmentation is one of the most indispensable tasks. Early diagnosis of tumorous cells aids substantially in early treatment planning and enhances the chance of survival of the patient. Manual segmentation of tumor cells in Brain MRI is difficult and time-consuming and also requires expertise in this field. In this paper, we have presented a deep Fully Convolutional Network designed using the TensorFlow library, which can successfully perform segmentation tasks in medical images. Brain MRI images of 225 patients are used as a training data set and a separate set from 20 patients is used to test the performance of the network. Two different loss functions: dice loss and weighted cross-entropy are used in the learning algorithm. We obtained a 0.76 Dice Score coefficient (i.e., 76% of the overlapping between the predicted image and ground truth) for a training of 100 epochs.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.