Malaria Detection using DL and Medical Image Analysis

Inspiration

Malaria is one of the life threatening disease caused specifically by parasites. The parasites typically spread into human blood through the bites of infected mosquitoes. As per the WHO statistics, the total estimate of Malaria cases was around 229 million and the estimated deaths stood at 409000 in the year 2019. Amongst the most affected were the children below 5 years of age which accounted for most death percentage due to malaria.

What it does

After observing the statistics we can notice how threatening malaria can be and hence there is a need to develop a robust technique to identify malaria in the early stages so as to take appropriate measures. Our project thus aims at analysing the microscopic images of cells and classifying it into parasitized or uninfected category.

How we built it

The project was divided into major sections:
Section 1: Data Collection - We obtained the required microscopic cell images data from National Library of Medicine an open source medical library.
Section 2: Data Preparation - The original dataset contained images with 2 relevant classes i.e. parasitized and uninfected. In data preparation phase, our major focus was to split the data appropriately into Training, Validation and Testing. Further the images were scaled to appropriate dimensions for modeling purpose.
Section 3: Data Modeling - The major focus of this phase was to develop a robust deep learning model for accurate classification of microscopic cell images. We achieved the best accuracy score using pre-trained model.
Section 4: Model Evaluation - The testing data was evaluated using several classification metrics like accuracy score, precision, recall and F1-Score. Visualizations of accuracy and loss charts were also performed to get an understanding about underfitting or overfitting model.

Challenges we ran into

1) Detecting appropriate dimensions of the images to be used for modelling.
2) Detecting best model and their hyperparameter optimization.
3) Dealing with overfitting situation.
4) Less processing power (GPU, RAM, etc.)

Accomplishments that we're proud of

Accurately and quickly determining whether the person is parasitized or not, so that required measures can be taken based on the results obtained.

What we learned

1) Handle large image dataset
2) Efficiently applying deep learning for image classification to produce accurate results.

What's next for Malaria Detection using DL and Medical Image Analysis

Further in parasitized infection there are several parasites which needs to be analyzed and detected accurately due to their severe threat possessing capabilities. So once the parasitized class is identified we can focus our future research to obtain specific species of parasites through image analysis.