Malaria Detection on Giemsa-Stained Blood Smears Using Deep Learning and Feature Extraction

Malaria is a severe and fatal disease leading to exacerbation of a person’s health and even death. Thus identifying whether a person is suffering from malaria is much preponderant. General and famous practice for the identification of malaria is usually the analysis done by microscope on the thin and thick stained blood smears. The identification of the parasitized blood cells is a laborious and challenging task as it involves the very convoluted methods such as spotting the parasite in the blood, counting the number of the parasites, pretending its type, etc. The fidelity for the stated task depends a lot on the experience of individual technician performance and relies a lot on him/her. The Convolution Neural Network models are famous for the automatic selection of these features. In this paper, 128 features from a custom-build Convolution Neural Model (CNN) were extracted and fed to the Support Vector Machine (SVM) classifier. The results were then compared with the custom-build, 17 layers deep CNN model and also with various Transfer Learning (TL) models, both as feature extractors and fine-tuned. After feeding the right features, the Support Vector Machine resulted in an accuracy of 98.8% on the dataset containing a total of 27,558 labeled red-blood-cell (RBC) images which were re-sampled to 40 * 40 * 3. Other performance matrices were also evaluated and are as sensitivity being 98.3%, specificity 97.6%, and F1 score resulting in 97.95%. Comparing with the state-of-literature, transfer learning models and custom-build model, CNN-based SVM classifier model performed the best.