Electroencephalography and Blood Cells Analysis for Cerebral Malaria Detection using Deep Learning and Neural Networks

Abstract

Cerebral malaria is a clinical syndrome that is marked by the asexual parasitic form - ‘plasmodium falciparum’. It has been a major health issue contributing to a huge number of deaths around the World, especially widespread in the suburban regions of Africa. Its mortality rate stands at 20% in adults and 15% in children. However, if diagnosed at an early stage, patients can receive proper treatment and recover quickly thus avoiding fatal and long-lasting neurological outcomes such as severe psychosis, metabolic acidosis and hypoglycemia. The symptoms associated with cerebral malaria often fall into the common bracket including fever and body ache which makes diagnosis difficult and possibly inaccurate. Therefore, it requires to be tested by a model that is specific to confirm the presence of the disease.

To achieve this, a bipartite framework is considered. The developed model aims to determine the contribution of this unicellular protozoan parasite and its impact on blood cells and neurological dysfunction. The design is divided into the following functions - the recognition of potential seizures in the patient and the identification of parasitic blood cells. The association of the results will help hypothesize and confirm cerebral malaria. In the first half, The deep-learning algorithm consists of a Neural Network Sequential model which is compiled under the ‘adam’ optimizer, ‘SparseCategoricalCrossentropy’ loss and matrices based on accuracy used in the framework in order to grasp the discriminative electroencephalogram (EEG) features of epileptic seizures recorded for each patient for 23.6 seconds [1]. Particularly, it works with Seizure-activity recordings to detect the various representations of the differing EEG patterns to reveal the correlation between successive data samples which is then utilised for training and classifications with 97.22% accuracy. The latter work focuses on inspecting the decompressed blood cells images that are fed into a deep convolutional neural network and different lossy image compression methods are examined as contrasting compression ratios impact the classification accuracies to distinguish parasitized cells from the healthy cells [2,3]. This is made possible using the sequential neural network model just like the previous model but this time it is compiled with ‘binary_crossentropy’ loss instead [4]. With this, the transfer of medical findings is made effortless since compression of images is done without loss of valuable information which is known as image augmentation done with the help of a library named ‘ImageDataGenerator’. As shown in figure 1, It helps in precise diagnosis with 94.7% accuracy.