Brain Tumor Classification using EfficientNet-B0 Model
Veeranki Goutham, Abdul Sameerunnisa, Babu Sallagundla, Tugu Bhanu Prakash
Abstract
The diagnosis of cancer cells is a problematic task. Early detection of cancer can save a person's life. Brain(cerebrum) tumors can be classified into two types they are benign (non-harmful) and malignant (harmful). Benign tumors don't propagate to nearby structures or distant body parts. Malignant tumors are propagated to other body parts. The proposed strategy predicts the outcome as no_tumor, pituitary_tumor, glioma_tumor, and meningioma_tumor. To predict this outcome, we have employed the EfficientNet-B0 model. EfficientNet-B0 is a pre-trained model trained over millions of images. EfficientNet takes the Magnetic Resonance Images (MRI) as input and keeps on adding the hidden layers to improve efficiency. The dataset images are fed to the model with the help of GUI, which then displays the predicted output. In Regular classification, the type of tumor by Radiologists takes about two hours but using this proposed system we can classify the tumor within minutes. This methodology achieved a validation accuracy is 96.94%.