Dataset used: https://figshare.com/articles/brain_tumor_dataset/1512427
Model used: AlexNet
Trained on: Google Collab with Python 3.8
To build a deep neural network which can identify what kind of tumor exists in the brain. The model is supposed to classify brain tumors between: Meningloma, Glioma and Pitutary Tumors.
Form the dataset mentioned above, we are using 2440 images as training set and rest 660 images as validation set. The dataset consists of 708 images for Meningloma, 1146 images for Glioma and 915 images for Pitutary Tumor. Also, the images in the dataset can belong to any of the three different views: Sagittal, Axial and Coronal Views (These views are nothing but different angle of of brain MRI). We have to train a model that can detect tumors accurately despite the views.
NOTE: I have put all files from the dataset into one single folder in this project.
Our model is an implementation of AlexNet with 3 convolutoin layers and then 3 fully connected layers, as mentioned below:
!
Required Modules: Tensorflow 2.0, Numpy, h5py, Matplotlib.pyplot, os and cv2
- Fetch the entire dataset to your google drive, unzip all folders and store them in one folder. (you can do this easily using https://www.multcloud.com/)
- Open the notebook in Google Collab.
- Mount Google Drive.
- Do ya thing!
- Hit me up on namanbansalcodes@gmail.com, let's talk.
OR
- Add my pickle file to your drive using this link: https://drive.google.com/file/d/1IKvwqGRjZajEf78I41j2VOdv_ztsQl6R/view?usp=sharing
- Open the notebook in Google Collab.
- Mount google drive and run the code block that says "load pickle file".
- Do ya thing!
- Enlighten me please.
When trained on 20 epochs with a batch size of 32, the model gives a training accuracy of 0.93 based on Sparse_Categorical_CrossEntropy loss function and it achieves a validation accuracy of 0.89 and a loss of around 0.23.
Please help me improve the results. Hit me up at: namanbansalcodes@gmail.com