Research Area:  Machine Learning

Abstract:

The brain is a complex organ of the body. Any abnormality in brain cells can affect the function of the human body. Brain space-occupying lesions include tumors, abscesses, and cysts. Brain MRI images are noisy that degrades the detection accuracy. Therefore, 32-layers-denoise neural network is proposed on the selected hyper-parameters to improve the image quality. To classify the healthy/abnormal MRI slices, a novel seven layers Javeria Quanvolutional Neural Network model is proposed named as J. Qnet, that consists of the four dense, two drop-out, and one flattened layers. To localize the classified images, the open exchange neural network (ONNX)-YOLOv2tiny model is proposed based on the selected layers that is trained on the optimal hyper-parameters. To segment the localized images more accurately, 34 layers of U-net model are proposed, which is trained from the scratch using selected hyperparameters. The proposed model is evaluated on locally acquired images and BRATS-2020 dataset providing an accuracy of 0.96 and 0.98, respectively. Overall, the proposed method performed better as compared to the existing research works that authenticate the novelty of this work.

Keywords:  

Author(s) Name:  Javaria Amin, Muhammad Almas Anjum, Nadia Gul & Muhammad Sharif

Publisher name:  Springer

DOI:  10.1007/s00521-023-08717-4

Volume Information:  Volume 35, pages 19279-19295, (2023)

Paper Link:   https://link.springer.com/article/10.1007/s00521-023-08717-4#:~:text=Several%20techniques%20have%20been%20conducted,a%20challenge%20for%20tumor%20classification.