Chest radiography is the most common imaging examination globally, critical for screening, diagnosis, and management of many life threatening diseases.
My aim is to provide automated COVID-19 chest radiology analysis comparable to professional practicing radiologists. In addition, I hope to assist various medical settings such as improved workflow prioritization and clinical decision support. I have faith that my application will advance global population health initiatives through large-scale screening.
Main steps:
Task | Time | Progress |
---|---|---|
Data collection | 1 days | x |
Data preprocessing | 1 days | x |
Building Model | 5 days | x |
Build Flask App | 1 day | x |
Total | 8 days |
This dataset contains 1000 images of COVID-19 and Normal of 200 patients.I had Provided the COVID-19 dataset,Which all preprocessing done...!! dataset link: https://drive.google.com/drive/folders/1sIm7jJ_OcTIeNYfx_Yu54c_5whN0-UVL?usp=sharing
- Python and some neccessary libraries such Tensorflow, keras, pandas, numpy, keras, tensorflow, flask.
- Google Cloud Platform to train models.
- The training labels in the dataset for each observation are either 0 (COVID +ve), 1 (Normal). Explore different approaches to using the uncertainty labels during the model training.
Binary classification CNN model to recognise 2 different medical observations corona +ve or -ve.
- Building CNN model
The architecture is built by Tensorflow and Transfer Learning techniques. More details can be found in COVID-19_detector.ipynb
.
densenet = tf.keras.applications.densenet.DenseNet121(weights='imagenet',input_shape=(224,224,3),include_top=False)
densenet.trainable=False
model = tf.keras.Sequential([
densenet,
tf.keras.layers.GlobalAveragePooling2D(),
tf.keras.layers.Dropout(0.25),
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='elu'),
tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(64, activation='elu'),
tf.keras.layers.Dropout(0.5),
tf.keras.layers.Dense(14, activation = 'sigmoid')])
base_learning_rate = 0.0001
model.compile(optimizer=tf.keras.optimizers.Adam(lr=base_learning_rate),
loss='binary_crossentropy',
metrics=["accuracy"])
Our model has the AUC of ~97 % for the train dataset and ~90 % for the validation dataset.
Successfully built a deep neural network model by implementing Convolutional Neural Network (CNN) to automatically interprete ChestX-ray_Based-COVID detection. In addition, we also built a Flask application so user can upload their X-ray images and interpret the results.