Blood parasite identification using feature based recognition

Diagnosis of diseases like malaria are very dependent on the identification of parasites in blood. Various methods have been applied for this process. The majority uses machine learning to identify the parasites. This method has shortcomings in long training time and the need to be retrained if a new data emerged. Of the various methods that have been applied, identification using feature based recognition is still rarely used. This method is strong in the term that it does not require training process, but only an image sample from which the feature will be extracted. In this paper, we design an identification process for blood parasites using one of the famous local feature extraction algorithms, i.e. SURF (Speeded-Up Robust Features). In our experiment, we evaluate the system to identify Plasmodium parasites. In this experiment, we are focusing only on parasite's gametocyte stage. Here, we use the system to identify whether or not the parasite is Plasmodium falciparum, Plasmodium malariae, Plasmodium ovale, or Plasmodium vivax. From this, we obtained 60%, 33.33%, 75%, and 50% true positive rate for each type of parasite respectively. Although this method has its strength, the experiment shows that the result obtained have not reached the desired expectations. This may due to the fact that the identified parasites is from the same species so that it contains some common characteristics.