Inspiration
Alzheimer's disease is one of the most common diseases in the world and detecting the onset of the disease is very important to get the proper treatment and diagnostics for the disease.
What it does
The CNN model predicts patterns into the images and classifies the different forms of dementia. class 1 = Mildly demented, class 2 = moderately demented, class 3 = non-demented, class 4 = very mildly demented
How we built it
We used ResNet18 as the architecture model and Flask as the web app to display the results online.
Challenges we ran into
Accurately predicting the probabilities since overfitting and underfitting can lead to inaccuracy in training and testing data.
Accomplishments that we're proud of
With our fine-tuned ResNet18 model, we have achieved a validation set accuracy of 90% , showing promising results about the project's future.
What we learned
- How to build a CNN architecture from zero.
- How to batch dataset
- How to train the model
- How to validate/test the model
- How to fine-tune pretrained models, freezing/unfreezing parameters, saving them etc.
What's next for Alzheimer Detector
We want to use longitudinal research data to predict whether you will have Alzheimer's in the future - 5 years from now, 10 years from now, or 20 years from now. And we want to expend our model to other diseases as well.
Log in or sign up for Devpost to join the conversation.