Skip to content

Latest commit

 

History

History
107 lines (63 loc) · 4.16 KB

README.md

File metadata and controls

107 lines (63 loc) · 4.16 KB
Raw

Introduction

Problem

The goal of this project is to develop a deep learning model that can accurately predict the presence of a brain tumor in MRI images. The model will take as input an MRI scan of a patient's brain and output a probability of the presence of a tumor. The model will be trained on a dataset of MRI images of both healthy brains and brains with tumors, and will be evaluated on a separate dataset to determine its accuracy in predicting brain tumors.

Description

The project involved training a convolutional neural network (CNN) on a dataset of MRI images of both healthy brains and brains with tumors. The CNN trained to learn the features of a brain tumor, such as its shape and location, in order to make predictions on new MRI images. Also there is simple website made on Flask and bootstrap libraries that has a functionality to upload some MRI image of brain and get text result of model prediction.

By pressing “Predict!” button there will be an POST request on server where saved model is being loaded and using model.predict() method it gives a prediction on uploaded image:

  • 0 - if there is no brain tumor
  • 1 - otherwise. And depending on this output there in website appears corresponding text.

How to run

  1. Type in your terminal:

    python mainTrain.py

    After executing command above you will see that there is new file named BrainTumor20Epochs.h5

  2. Run your application by typing:

    python app.py

    It will run your server

  3. Go to http://127.0.0.1:5000 and enjoy :)

Data and Methods

Information about the data

Dataset is found on Kaggle.com and it contains 3060 images of brain MRI images which is divided into 3 folders:

  • 1500 images with regular clear and healthy brain
  • 1500 images with cancer brains
  • 60 images in separate folder named “pred”

data

no-cancer cancer

All images in .jpg format. After downloading data as zip file I am processing it in following way:

process

And after dataset and labels are corresponds to each other I am splitting data into train and test sets with 80/20 proportion correspondingly(Pareto rule did not do anything, lol) using scikit-learn utility:

split

Description of the ML model For this particular project I have used pretty average model that we have covered during lectures on CNN topic, also similar model been used in second assignment. I have used Sequential model with 3 Conv2D convolutional layers and 3 Pooling layers with (3,3) pool. Also there is Dropout layer to prevent overfitting. Last one is Dense layer with 1 neuron and sigmoid activation function. There is only two possible outputs of our model, so for binary outputs it is more likely to use sigmoid function. Therefore, i have user binary_crossentropy in model.compile function. To sum up everything that had been mentioned before, several convolutional layers extract features from the input image, pooling layers reduce the spatial dimensions of the output, a completely connected layer labels the features, and a dropout layer prevents overfitting.

model

Results

After 10 epochs model has an accuracy of 96-98% and data loss that is in range 7-8%:

acc

loss

And this is how model works step-by-step:

web-1

web-2

web-3

Just for fun I tried to upload my photo and that is pretty funny, ahahah.

Next steps

Next step is to master ML/DL skills, learn Data Sceince from scratch, because I am feeling lack of fundamental knowledge in this domain and make this project ready-to-use and integrate to hospitals on non-commercial base. I am planning to add some features like:

  • Cancer type classification
  • How much is it dangerous
  • What may be caused by this exact type of tumor
    And so on...

Useful links:

Dataset: https://www.kaggle.com/datasets/ahmedhamada0/brain-tumor-detection)

Telegram: https://t.me/boobieloverr

LinkedIn: https://linkedin.com/in/islam-adilkhan

Email: [email protected]