Brain tumor Classification task

• 

  ModePresentation-CNN Model

Inspiration

The implementation of an AI model that accurately detects asymptomatic Covid-19 infections through recorded coughs ( https://www.embs.org/ojemb/articles/covid-19-artificial-intelligence-diagnosis-using-only-cough-recordings/). More especially, accurate diagnostics of brain tumors improve the life expectaction of patients. However, current diagnosis techniques are error-prone due to their complexities. Thus, building ML models that accurately diagnose the type of brain tumor from an MRI scan is of great interest. Since MRI scans only depict low-level spatial features, using transfer learning from performant CNN models points to a more viable solution.

What it does

This solution accurately predicts the presence and type of brain tumor(pituitary, meningio, glioma or no tumor) of an individual's MRI scan image from any anatomical plane.

How we built it

I compared the performance of two ML models: a simple multi-task SVM classifier and a pre-trained CNN model. The webapp was build with a flask backend and a react frontend.

Challenges we ran into

Despite the implementation of multiple regularization techniques, my model was overfitting. Moreover, I struggled to pre-process the data in the correct format as the pre-trained CNN model used (kera's EffNet B0).

Accomplishments that we're proud of

I'm well pleased that I achieved a 93% overall test accuracy with my pre-trained CNN model.

What we learned

Transfer Learning brings the stability and training of the most performant CNN models, from which we can build upon adaptable CNN layers according to the task at hand.

What's next for Brain tumor Classification task

In the nearby future, I would like to build and train my own CNN model based on other performant applications of brain tumor classification ML models. I would also like to train my models to detect other types of brain tumors.

Built With

• flask
• javascript
• material-ui
• python
• react