| Autor: |
Aby Abahai, T., Arjun, Thulaseedhar, Sabarinath, S., Jerin, Babu |
| Předmět: |
|
| Zdroj: |
AIP Conference Proceedings; 2024, Vol. 3134 Issue 1, p1-8, 8p |
| Abstrakt: |
This study compares different CNN architectures for identifying and categorizing brain tumors. Brain tumors are mainly of three types which are Glioma, Meningioma and pituitary adenoma. Tumors can appear in a variety of places, and the location of a tumor may reveal information about the type of cells that are responsible for it, aiding in a more accurate diagnosis. Here we will use MRI to identify and classify tumors. The main issues in MRI are illumination which can provide wrong result. To remove this the these images should be processed before training. To enhance the quality of the MRI, many image pre-processing techniques are used, such as histogram equalization, blurring and thresholding. This will help in faster classification. Six different architectures are compared in this study. These are ResNet50, VGG19, InceptionV3, MobileNet, DenseNet and NASNetMobile. By comparing this architectures we can find which one is more faster and accurate. Also maybe some architecture will be better than others for identifying certain type of tumor. That can also be identified from this. The model will be trained at different epoch to validate the result. About 20 percent of images will be used for testing and remaining are used for training. Also a part of this paper deals with calculation of brain tumor volume from MRI. Finding volume of brain tumor at two different period can help us to identify progression rate. The direction in which tumor is progressing is also identified. This will help with treatment of tumor. Volume is obtained from various slices of MRI of same brain at different locations. From each slices area of tumor is identified by segmentation. Segmentation allows us to obtain tumor part from MRI. Area will depend on number of pixels in tumor part. Two slices will be separated by a distance. This distance is multiplied by area of slice to give its volume. Summing up volume of each slice will give the approximate volume of the tumour. The main problem is each MRI should be of same resolution. This is a major design issue which we are yet to tackle as it depends on the MRI technology involved. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
