Artificial Intelligence-Based Approaches for Detection and Classification of Different Classes of Malaria Parasites Using Microscopic Images: A Systematic Review

Artificial Intelligence has played an essential role in detecting malaria, which aims to reduce the involvement of any human microscopist in order provide an accurate diagnosis with minimum interference from the human being. The main aim of this paper is to create an AI-based system that helps for the detection and classification of malaria parasites in microscopic images. The data has been collected from three different sources to create four classes, i.e., Falciparum, Malariae, Ovale, and Vivax. Later, data pre-processing is done to reduce an image's size and remove the noisy signals. Further, feature extraction techniques are applied to obtain the region of interest by calculating the morphological values, and generating the bounding box around the selected features after applying watershed segmentation. Deep learning models such as DenseNet121, DenseNet201, ResNet152V2, NasNetLarge, MobileNetV2, and the hybridized DenseNet201 and ResNet152V2 are being applied and trained with the dataset. These models have been evaluated using metrics such as precision, loss, F1 score, recall, and accuracy to examine their performance during the training and validation phase. At the time of experimentation, it has been found that MobileNetV2 computed the highest accuracy, and ResNet152V2 has the best loss value by 0.005. Regarding precision, recall, and F1 score, DenseNet121 computed the exact value of 1.00. Besides this, the models are also examined for different classes by computing the actual and predicted values through a confusion matrix.

Automated summary

Artificial Intelligence is used in detecting malaria to provide accurate diagnosis with minimal human interference. This paper aims to create an AI-based system for detecting and classifying malaria parasites in microscopic images. Different deep learning models are trained and evaluated for their performance in terms of precision, loss, F1 score, recall, and accuracy. MobileNetV2 achieves the highest accuracy, while ResNet152V2 has the best loss value of 0.005. DenseNet121 achieves perfect precision, recall, and F1 score of 1.00. The models are also evaluated using a confusion matrix to compute actual and predicted values for different classes.