You are on page 1of 2

Abstract

In the field of biology, cell Detection and counting is a manual and tedious process which
involves human effort, it takes much time for Researcher or medical student to count the cells
and determines its viability i.e. either cell is alive or dead. Image processing techniques are
being used for the detection and counting of microscopic cells whose images are already been
taken. Reproductive studies showed that much research is already been done in the field of
image processing and different algorithms of counting and detection of microscopic cells have
already developed.

The focus of this research is to make an efficient and powerful machine learning algorithm
along with image processing techniques that helps in determining cell viability and cell count in
a given sample whose microscopic images has been obtained using some specific dyes. Image
processing is a software based technology that plays a major role in field of technology with
applications such as robotics, recognition, remote sensing etc. Machine Learning is the science
of getting computers to behave without being clearly programmed. In the past era machine
learning has been used in many artificial intelligence based projects like Speech Recognition,
Objection Recognition, Classification problems, speech recognition, weather prediction etc. The
knowledge of machine learning is used in many medical projects like detection of cancerous
cells, differentiation of human blood cells etc.

An Image labeling tool is developed to label alive, dead and no cells in an image dataset
containing 400 microscopic images of amoeba stained by using trypan blue dyes obtained from
university Biomedical laboratory. Machine learning Algorithm works in a way that it trains the
machine on the basis of dataset provided, the machine learns the data and by using the trained
data, the algorithm takes the decision for some unknown image of microscopic cells as alive
dead or no cell.

Data labeling tool is used to label two third of the dataset where the remaining one third of the
dataset is used for testing and validation. Each cell of microscopic image from the training
dataset is labeled as alive dead or no cell using data labeling tool. Data labeling tool works in
way that it captures the current axis of the cell using built in functions of Matlab. The
information is stored in memory of computer by image name in a “matrix” file. Information of
each cell is big enough for machine to handle so by using unsupervised machine learning
technique, the information data is passed to K-Means clustering algorithm so the data with
same values cluster in a single cell. Depending upon the requirement; K-Means clustering
shrinks the dataset into clusters of alive dead and No cells. The cluster of data cells are then
used to extract set of feature vectors by calculating correlation coefficient of each cell in the
image with the rest of all cells of data images forming training dataset. This stage is called
training phase.
In testing phase, sliding window of some specific size slides over the test image pixel by pixel to
get the small image of size equals to the centroid image. The resultant image is then passed to
the distance classifier that classifies either the region portion is alive, dead or no cell.
Corresponding label is assigned to result obtained from distance classifier and stored in a
matrix. The process repeats until whole test image is passed through the distance classifier. By
using image processing techniques the matrix file is converted into an image. Noise and false
positives removed and cell count of alive and dead cell is calculated.

The Experimental results obtained from this research prove to be beneficial for the automatic
cell detection and counting. It is to be believed that any dataset stained with trypan blue dyes
would be easily distinguished as alive and dead cell by using that proposed technique.

You might also like