Abstract:
Malaria is among the most important world public health issues, with considerable morbidity and mortality, particularly in World third countries. Therapy and disease control rely on fast and precise diagnosis. Traditional methods like microscopic observation of blood smears are time consuming and skill demanding, and these may not be easily available for poor-resource countries.
The proposed project is focused on automated malaria infected cell detection with the implementation of a deep learning model. We utilized convolutional neural network ANN, (CNN), VIT and the hybrid CNN+VIT model implementation with the support of transfer learning and scratch modelling for increased performance. These models were trained and tested using a custom dataset which was constructed by merging images of the NLM Malaria Dataset and the Malaria Cell Images Dataset on Kaggle, totalling 27,500 images divided into four malaria classes—Plasmodium Falciparum, Plasmodium Vivax, Plasmodium Ovale, Plasmodium Malariae—and one class for non-infected. The model was executed efficiently with high precision, accuracy, and recall and hence is a reliable system to support malaria diagnosis.
Also, a web interface was developed using React.js for uploading cell images and obtaining the malaria infection predictions, thereby making the technology convenient and accessible. The article reflects a good account of method, experiment, and result and demonstrates the effectiveness of deep learning in medical image processing.
