Deep learning with PyTorch | Medical Imaging Competitions

Deep learning with PyTorch | Medical Imaging Competitions


  • Should have good understanding of python
  • Have basic theoratical knowledge of deep learning (CNNs, optimizers, loss function etc)
  • Have done atleast one project in machine learning or deep learning in any framework


Greetings. This course is not intended for beginners, and it is more practically oriented. Though I tried my best to explain why I performed a particular step, I put little to no effort into explaining basic concepts such as Convolution neural networks, how the optimizer works, how ResNet, DenseNet model was created etc. This course is for those who have worked on CIFAR, MNIST data and want to work in real-life scenarios

My focus was mainly on how to participate in a competition, get data and train a model on that data, and make a submission. In this course PyTorch lightning is used

The course covers the following topics

  1. Binary Classification
    1. Get the data
    2. Read data
    3. Apply augmentation
    4. How data flows from folders to GPU
    5. Train a model
    6. Get accuracy metric and loss
  2. Multi-class classification (CXR-covid19 competition)
    1. Albumentations augmentations
    2. Write a custom data loader
    3. Use publicly pre-trained model on XRay
    4. Use learning rate scheduler
    5. Use different callback functions
    6. Do five fold cross-validations when images are in a folder
    7. Train, save and load model
    8. Get test predictions via ensemble learning
    9. Submit predictions to the competition page
  3. Multi-label classification (ODIR competition)
    1. Apply augmentation on two images simultaneously
    2. Make a parallel network to take two images simultaneously
    3. Modify binary cross-entropy loss to focal loss
    4. Use custom metric provided by competition organizer to get the evaluation
    5. Get predictions of test set
  4. Capstone Project (Covid-19 Infection Percentage Estimation)
    1. How to come up with a solution
    2. Code walk-through
    3. The secret sauce of model ensemble
  5. Semantic Segmentation
    1. Data download and read data from nii.gz
    2. Apply augmentation to image and mask simultaneously
    3. Train model on NIfTI images
    4. Plot test images and corresponding ground truth and predicted masks

Who this course is for:

  • For itermediate users who know about python and machine learning
  • Have done cats and dogs classification problem but not sure how to handle a large data or problem
  • Want to step in medical imaging and build a portfolio
  • Want to win kaggle, codalab and grandchallenge comeptetions


We will be happy to hear your thoughts

Leave a reply

awsomenews Free Online Courses