IdenTumor

Color image of brain
Image of brain
Example of MRI scans
Logo

From the rapid growth of brain tumors and undetected cases, we had gotten inspired to create a piece of software that could potentially help with this.

Our app allows a user (preferably a doctor) to enter in an MRI scan of the brain. With this MRI scan, our deep learning model takes in this image, and outputs a prediction. This prediction could tell a doctor if the patient has no tumor, or if they may have a pituitary or glioma/meningioma tumor.

We had built the model using TensorFlow with the Keras API, and 5,000 training images of MRI scans. With TensorFlow, we are able to create state of the art models or applications in a fraction of the time. We used OpenCV and NumPy to preprocess the images. For example, greyscaling and resizing images are a key part of preprocessing and getting the image data in the right format. For the web app, we used HTML and bootstrap CSS, with a turquoise UI, to make the web app even more clean.

One challenge we ran into with building the app was configuring the model and Flask with the frontend. Resolving this issue took a bit of time but we were able to get it done in the end. Another challenge we had was with finding good datasets for the model. There were many datasets, but not all fit the requirements we needed to have, such as having multiple types of tumors in the dataset. Getting the image data in correct shape was another challenge, as we had to try and tinker with many preprocessing techniques to try to get the best possible result.

Even with all of these challenges and roadblocks, we were able to achieve an 89-90% accuracy on the model and create a great UI and frontend for the project.

For the project, we learned more about configuring frontends and backends, and how to deal with CNN's to achieve the best output for image data. These skills are essential, as they can carry us a long way in the future.

The next step for IdenTumor is to train our model on more data, to achieve the best results possible for anyone using this app. We could also go to more diseases/tumors and try to help in those fields, by providing early detection. For IdenTumor, we still have a long way to go.