Enhanced performance of Dark-Nets for brain tumor classification and segmentation using colormap-based superpixel techniques
Sakshi Ahuja, Bijaya Ketan Panigrahi, Tapan Kumar Gandhi
Machine Learning with Applications · 2022
doi:10.1016/j.mlwa.2021.100212 (opens in a new tab)
We email you when a new notice or a question about this paper appears.
Abstract
The brain tumor is the deadliest disease in adults as it arises due to an abnormal mass of cells that grows rapidly and it alters the proper functioning of the organs. In clinical practice, radiographic images of different modalities are used to diagnose types of brain tumors, their size, and location. The proposed work aims to automatically classify, localize, and segment brain tumors from T1W-CE Magnetic Resonance Image (MRI) datasets. The T1W-CE MRI dataset is divided into 8:1:1, i.e., 80% training set, 10% of each validation, and testing set. To address the overfitting issues, the training data set is augmented using 2-levels wavelet decomposition and geometrical operations (scaling, rotation, translation). Performance of pre-trained DarkNet model (DarkNet-19 and DarkNet-53) is evaluated for the multi-class classification and localization of brain tumors. The best performing pre-trained DarkNet model achieved 99.60% of training accuracy and 98.81% of validation accuracy. The performance evaluation parameters confirm the superiority of the proposed methodology in comparison to the state-of-the-art on the T1W-CE MRI dataset. On 1070 T1W-CE testing images, the best-performing pre-trained DarkNet-53 model obtained a testing accuracy of 98.54% and Area Under Curve (AUC) of 0.99. The tumor is segmented using a 2-D superpixel segmentation technique with an average dice index of 0.94 ± 2.6% on the 793 brain tumor testing data. To prove the superiority of the proposed technique, it is implemented on MRI images from the BraTS2018 dataset. The comparative analysis of performance evaluation parameters of the proposed methodology with the state-of-the-art technique proves its robustness and clinical significance.
Abstract licence: CC BY-NC
Reads
Read by people on this page
None yet.
Checked or read by AI through naulon
None yet.
A person counts once a day per paper. We keep no cookie and no address: a reader is a hash that changes every day. Any month under five reads shows no number.
Are you an author?
Is this your paper?
Claim it to add your note and be paid when agents cite it. Free.