Multi-Species Classification of Malaria Parasites using Ensemble Transfer Learning Technique

Malaria, a prevalent global health concern, necessitates accurate and prompt diagnosis for the treatment and control of disease. Microscopic examination of thin blood smear images is the definitive method for diagnosing Malaria. However, this method relies on skilled experts and is prone to human error. This paper introduces an innovative method for categorizing thin blood smear images of Plasmodium parasite species into their four types through an ensemble transfer learning technique. By utilizing the information gathered from pre-trained models on sizable datasets, our approach uses a deep learning framework built on transfer learning. A diverse ensemble of models, such as Xception, EfficientNet, MobileNet and DenseNet 121, was utilized to collectively capture intricate patterns and representative features of different Malaria parasite species. To develop and evaluate our method, we collected and preprocessed the MP-IDB dataset, which includes blood slide images of all four parasite types. The literature reveals that no research has explicitly focused on identifying and categorizing all four Plasmodium species in blood smear images. It signifies a gap in the current body of knowledge, emphasizing the need for comprehensive investigations in this specific domain. Through rigorous experimentation and evaluation, our proposed weighted ensemble transfer learning technique achieves an accuracy 98.9 in classifying Malaria parasite species into its four types. Our work contributes to advancing computer-aided diagnosis of Malaria, paving the way for more effective and efficient disease management strategies.