Ecosyste.ms: Awesome

An open API service indexing awesome lists of open source software.

Awesome Lists | Featured Topics | Projects

https://github.com/jeevan-j/blood-cell-characterization

This project aims to develop an end-to-end machine learning model which counts the blood cells, classify White Blood cells into sub-categories and predict probable diseases based on blood smear images taken from blood slide scanner or profiler.
https://github.com/jeevan-j/blood-cell-characterization

Last synced: 3 days ago
JSON representation

Host: GitHub
URL: https://github.com/jeevan-j/blood-cell-characterization
Owner: Jeevan-J
License: gpl-3.0
Created: 2019-03-15T04:41:45.000Z (almost 6 years ago)
Default Branch: master
Last Pushed: 2019-03-15T09:59:51.000Z (almost 6 years ago)
Last Synced: 2025-01-06T04:07:35.484Z (10 days ago)
Size: 1.13 MB
Stars: 0
Watchers: 2
Forks: 0
Open Issues: 0
Metadata Files:
- Readme: README.md
- License: LICENSE

Awesome Lists containing this project

README

# Blood Cell Characterization

Problem Statement

- Most of the blood test results for diseases are based on the cell structure and cell count. Pathologists analyse blood samples under microscope and gives results which is time taking process and also error prone in some cases
- Since the development of digital microscopes and machine learning techniques, many attempts were made to automate the process
- The existing models in the market are costly. There are few startups working to develop a low cost slide scanner with a trade off between cost and quality of blood smear images
- Aim is to develop an end-to-end model, with image processing techniques and machine learning algorithms, which can process microscopic images to count blood cells, classify white blood cells and predict probable diseases.

Fig.1 - A typical blood smear slide image under microscope

Work Done

- Applied an image processing technique with few pre-processing steps to count Red Blood Cells (RBCs) in the blood smear image.
- Identified and segmented out White blood cells (WBCs) from blood smear image.
- Applied a machine learning algorithm to do semantic segmentation of White blood cells from segmented images.
- Developed a machine learning algorithm to classify segmented WBCs into four sub-categories.
- Worked on "Classification of Normal vs Malignant Cells in B-ALL White Blood Cancer Microscopic Images" : ISBI 2019 challenge hosted by SBI Lab, IIIT Delhi.
- Developed an ensemble based machine learning model to classify leukemic B-lymphoblast cells from normal B-lymphoid precursors from blood smear microscopic images

Fig.2 - Proposed end to end model

Results

- Presently, our image processing technique is able to count RBC up to 97% accuracy excluding overlapped cells.
- Developed ensemble based machine learning algorithm is able to produce an accuracy of 94% in classification of WBCs into sub-categories.
- The same model when trained for two class classification of **leukemic B-lymphoblast cells from normal B-lymphoid precursors** produced an accuracy of 82%.

Fig.3 - Performances of of different architectures/approaches on Classification of leukemic B-lymphoblast cells from normal B-lymphoid precursors

Datasets

- [C_NMC_2019 Dataset by SBI Lab, IIIT Delhi.](https://competitions.codalab.org/competitions/20429)
- [ALL_IDB1 and ALL_IDB2 datasets](https://homes.di.unimi.it/scotti/all/#datasets)
- [WBCs Dataset from Blood Cell Images, Kaggle](https://www.kaggle.com/paultimothymooney/blood-cells)

Future Work

- Acquire the real data-set of blood smear images from blood slide scanner/profiler
- Apply active learning based training for ensemble based architecture and analyse the results
- High level Work Plan: Build an end-to-end Convolutional Neural Network with detection & segmentation of cells followed by classiﬁcation of cells and prediction of few diseases with in Active Learning framework