Brain Tumor Detection and Segmentation Using CNN-Based Classification and U-Net Based Segmentation
Dr.D.V.Lalita Parameswari ,Associate Professor
Department of Computer Science & Engineering,
G. Narayanamma Institute of Technology & Science
Hyderabad, INDIA
dvlalitha@gnits.ac.in
Dr.B.Sasidhar ,Assistant Professor
Department of Computer Science & Engineering(AI&ML),
G. Narayanamma Institute of Technology & Science
Hydeerbad ,INDIA
bolasasi@gnits.ac.in
Mekala Poorvaja
Department of Computer Science & Engineering,
G. Narayanamma Institute of Technology & Science
Hyderabad, INDIA
mekalapoorvaja@gmail.com
Abstract—The precise identification of brain tumors stands necessary for successful disease and situation preparation activities. The primary difficulty in this process involves handling incomplete MRI scans since these create problems for standard machine intelligence approaches because of unstable edema combined with missing information. A U-Net model together with a multimodal engine network forms an integrated design targeted at fixing incomplete MRI scan issues. The facial characteristics extraction process from multiple presentation methods is done by modality-distinguishing encoding systems which feed information to a multimodal generator that rebuilds missing modalities alongside their associated connections. The system brings together features from a joint-lawyers lingual system which relies on dimensional self-contemplation mechanisms with channel self-consideration mechanisms to maintain precise separation. The proposed method demonstrates better results than current separation techniques when operating with deficient MRI datasets according to BraTS dataset examinations. The multimodal transformer network excels at processing incomplete data better than traditional methods which is proven through a SegNet-based segmentation comparison that validates its enhanced accuracy and reliability during tumor separation operations. The research demonstrated that UNet obtained 98.73% segmentation accuracy which exceeded CNN score of 91.7% classification accuracy.
Keywords—Deep Learning, UNet, VGG16 UNet, Segmentation.
