Overview / Usage

The aim of this project is to build a machine learning model to label histologic images. We want to develop something which can have a look at a histologic section and tell which tissue it is (skin/ liver/ lung/ kidney etc).

The data is not directly available. So it is being prepared from the ground up. There are a few label noises as well which we are manually rectifying for the time being. But eventually, this will be handled by a robust model. Then the challenge is to develop a model with a mimimal amount of data.

Methodology / Approach

• Data augmentation to generate more samples of data
• ResNet34 as the main deep architecture to learn the feature maps present in the data
• Use of automatic mixed precision to enhance the training time of the model

Technologies Used

• Python (Intel)
• Keras, scikit-learn and fastai (main libraries and I have experimented with a number of modeling approaches)
• FloydHub

Repository

https://github.com/vaishleshik/histology_classifier/tree/master/Sayak_Notebooks