Classifying body fluids using deep learning
- 0 Collaborators
Investigation of different body fluids is a part and parcel of a pathologist's work-life. However, the manual process of classifying body fluids into categories like malignant and benign is quite time-consuming. The aim of this project is to better aid the pathologists who are working in this area. ...learn more
Project status: Under Development
Intel Technologies
AI DevCloud / Xeon,
Intel Python
Overview / Usage
The main objective of this project is to better aid the pathologist who are working in the area of identifying body fluids from pathology images and then classifying them. The project will enable to them simply drag and drop a pathology image slice to a web portal and then get the category with a certain confidence score.
Methodology / Approach
The main problem that arises when classifying several body fluids into categories is that pathology slices are very identical in nature. So, it gets really hard, even for a trained pathologist to distinguish between them. In this project, we are training deep learning models to do this task. Naturally, this empirical problem gets in the way of training and raises the question - how can we instruct the model to not get confused.
The solution that partially resolves this problem is aspect aware and domain-specific data augmentation which is one of the core components of this project. After applying such augmentations to the image data we have, we fine-tuned a pre-trained ResNet50 model for the representation learning and classification tasks.
Note: The images that were used in this project are pathology images.
Technologies Used
-
Python
-
FastAI
-
PyTorch
-
Matplotlib