Early Stage Brain Tumor Detection With CNN

Inspiration

The International Association of Cancer Registries (IARC) reported that there are over 28,000 cases of brain tumors reported in India each year and more than 24,000 people reportedly die due to brain tumors annually. So Our Team, Decided We Should Build Something That Have impact on Peoples Lives. Most People Got to Know About The Brain Tumor in the Middle Stages. And Early Detection of Brain Tumor Can be Live Savior for Patients- And That Our Inspiration

What it does

... Medical check-ups at times can be very time consuming and costly. For example, if a patient needs to get checked up for a possible Brain Tumor condition then the patient has to go through multiple layers of processing to get the final result. First, the patient needs to visit a doctor who might further refer the patient to get an MRI scan and then an MRI image will be processed and then sent to the doctor who will conclude whether the person is suffering from a brain tumor or not. As we can feel the above process can take days and sometimes even a whole week to complete the process and also requires a lot of labor, operators, equipment and additional workers for each process to be carried out. But not anymore. With our revolutionary and carefully designed platform, we can detect the Brain Tumor’s possible presence in a patient within mere minutes. Even early-stage presence of Brain Tumor can be detected using our platform which might get unnoticed even by neurosurgery specialist. Our platform works using machine learning, which is first trained using specially designed data sets and using algorithms and statistical models to analyze and conclude results. Now the platform is ready and can give the result of a patient’s condition after just getting an MRI scan using an MRI machine and using scanned images to be used as an input to our trained machine learning-based platform and getting results after being precisely analyzed within seconds. The machine uses CNN ( convolutional neural network) to analyze the image and give results using the prediction data to a maximum of 97 to 100 % accuracy. This can save the precious time of the patient while being cost-friendly by saving resources on unnecessary labor and multiple layers of the grueling procedure.

How we built it

Firstly We Gathered The Data-Set From Kaggle!
Then Did Data Cleaning,Segmentation,Labeling, Uniform formatting.
Now Converting Data Set To Numpy array of Shape(n,100,100).
Building TensorFlow Convoluted Neural Network.

 model = tf.keras.models.Sequential()
model.add(tf.keras.layers.Conv2D(100, (3, 3), activation='relu', input_shape=(100, 100, 3)))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(256, (3, 3), activation='relu'))
model.add(tf.keras.layers.MaxPooling2D((2, 2)))
model.add(tf.keras.layers.Conv2D(256, (3, 3), activation='relu'))
model.add(tf.keras.layers.Flatten())
model.add(tf.keras.layers.Dense(126, activation='relu'))
model.add(tf.keras.layers.Dense(2))

Compiling And Training Model.
Then We Saved the Model Tumor_CNN.hdf5 and Build a Web app with Streamlit on Google Colab.

Challenges we ran into

Data Cleaning Was the Most Intriguing Part.
Choosing the Best Optimized Model for The Project(CNN).
Deploying Web app through Google Colab.
Hyper tuning the Parameters For Best Results.

Accomplishments that we're proud of

We Trained Our Model With 90-99 Percent Accuracy.
Deployed the Web App Using Google Colab.
Successfully Implemented Convoluted Neural Network.

What we learned

Our Team Lead Together That

Data Cleaning Is the Most Time Consuming Part Of any Model Creation.
Convoluted Neural Network Gives Optimal Results on Image Classification.
How To Deploy Web App on Colab Using Streamlit & Ngork.

What's next for Early Stage Brain Tumor Detection With CNN

Currently Due To Smaller Data-Set, Our Model Is Not That Much Generalized. So Next Step Would be Training the Model with Much Larger Data Set and Integration Of Deep Learning