A comparative study on malaria cell detection using computer vision

Abstract

The detection of malaria causing organism is done in labs under a microscope manually by people. In this case, this job being done manually there is a maximum risk of error and false detection can cause a life at stake. so, a fare detection of this disease can help control and cure the disease in time. The traditional way of performing this task includes a lot of manual work to be done by a human which takes a lot of time and efforts to complete it. In order to solve this problem a lot of researchers have proposed different algorithms and model in which they used algorithms and concepts of transfer learning, Deep learning and Computer Vision algorithms like Visual Geometry Group Network (VGG net), Convolution Neural Network, ResNet50, YIQ color space Faster-RCNN and many more to classify and check if the cell image belongs to Uninfected class or Parasitized Class. These approaches were found to be very efficient in terms of accurately classifying an image and fast in terms of time taken to provide results. In order to identify malaria cell though Computer Vision and Deep Neural Network in this paper we have conducted a comparative study among four most efficient and widely used algorithms. These four algorithms will be tested on several performance evaluation parameters like Confusion Matrix, Accuracy, True Positivity Rate and Precision. This will help us to check the different aspects of these algorithms. Also, we will be performing hyperparameter tuning in every algorithm which we use. This will help us to make sure that these algorithms work with their maximum potential. The algorithms used in this paper are Convolution Neural Network, Yolo version 4, Yolo version 5 and Single Shot Detector. In this we will be retraining the whole Single Shot Detector algorithm with our dataset. The reason to choose these algorithms is that they are some of the most popular and widely used algorithms also they are highly efficient, and the main reason is that the core principle on which these algorithms works is different from each other so it will be very useful t compare these algorithms and see how they performs.