Brain MRI tumor classification

• 

  Using the Model to Predict

• 

  Using the Model to Predict

• 

Inspiration

Different types of tumors are misclassified at varying rates. The glioma tumor is misclassified approximately 21%-35% of the time. Early accurate detection and classification of tumors can be extremally beneficial for patients to the according treatment , which could possibly save their lives.

How I built it

This convolutional neural network model was built using Keras, and TensorFlow. Data was analyzed then split in to training, testing, and validation sets. Training images were augmented, so the model would perform better on external images. The model was build using convolution layers, max pooling layers, dropout, dense and flatten layers. The model was trained over 18 epochs.

Challenges I ran into

The challenging part of the model was to increase the accuracy, especially when the accuracy would get stuck at what could have been a local minima. Machine Learning is about taking potential measures which could possibly make the model better, and by analyzing the training history the model was able to perform accurately. Getting used to the new environment of Flask was definitely new. Not having coded a lot in html and css, using Flask was a bit difficult to navigate when it came to styling.

Accomplishments that I'm proud of

I'm proud to say that the model is 91.5% generalized, and can somewhat accurately predict the type of tumor in a brain MRI scan. Throughout the entire process of building the model, there was a lot of hyperparameter tuning, to make it as effective as it can be.

What I learned

Through this process I was able to learn how to use Flask, which is what the model has been deployed on. Learning how to use and deploy models on Flask, was the part I think I had the largest growth in. The more I hyperparameter tuned, I understood better measures to get CNN models to perform better.

What's next for Tumor classification

For the future, the plan is to continue building the website, make it look a bit more user friendly, and as always trying to improve the machine learning model to perform better. Eventually, getting this website to people who can use it, like healthcare professionals, or continue to develop it, so it can actually help people would also be wonderful.

Built With

• flask
• keras
• python
• tensorflow