The authors introduced the public MP-IDB: Malaria Parasite Image Dataset comprising blood samples infected with malaria. The dataset is meticulously crafted to facilitate the assessment and comparison of algorithms tailored for segmenting and classifying malaria parasite species. Each image is accompanied by its corresponding ground truth and classification detailing the type and stage of the parasite’s life cycle. The authors primary objective is to furnish a novel benchmarking resource to the image processing and pattern recognition communities, with the ultimate goal of fostering advancements in computer-assisted analysis of malaria parasites.

Motivation

Visual inspection of peripheral blood samples is a critical procedure conducted by skilled hematologists to diagnose pathologies and assess patients’ health conditions. However, manual analysis of blood smears is arduous, time-consuming, repetitive, and subject to variable precision due to operator proficiency. Leveraging image processing techniques can streamline cell analysis in human blood samples while offering accurate insights into cell morphology. Peripheral blood smear analysis serves as a widely used and cost-effective diagnostic method, providing valuable health insights to expert pathologists. Furthermore, blood cell images captured through microscopy may exhibit variations in illumination and coloration conditions, further emphasizing the need for automated analysis methods.

Different illumination conditions could generate unconventional colour schemes in images. This is due to the absence of a standardized acquisition procedure. From left to right: same smear acquired with four microscope brightness levels.

Blood cell images typically comprise three primary components: platelets (or thrombocytes), red blood cells (RBCs or erythrocytes), and white blood cells (WBCs or leukocytes). It’s important to note that these cells exhibit diverse shapes, characteristics, and colorations based on their types. Various tests, such as Complete Blood Count (CBC), are designed to ascertain the quantity of erythrocytes and leukocytes in the blood, along with parameters like volume, sedimentation rate, and hemoglobin concentration of red blood cells (blood count). Additionally, specific tests are employed to categorize blood based on particular red blood cell antigens or blood groups. CBC includes different calculations such as red blood cell count (RBCC) or white blood cell count (WBCC) per cubic millimeter of blood, a differential white blood cell count, hemoglobin assay, hematocrit, red cell volume calculations, and platelet count. Human malaria infection, caused by parasitic protozoans of the Plasmodium genus, necessitates distinct tests for identification. Transmission occurs through the bites of infected female Anopheles mosquitoes, known as malaria vectors. Among the five parasite species causing malaria in humans, Plasmodium falciparum and Plasmodium vivax pose the greatest threat, while Plasmodium ovale, Plasmodium malariae, and Plasmodium knowlesi are less perilous.

Morphological scheme of human malaria parasites types and stages of life.

Computer vision techniques have emerged as a promising frontier for early malaria detection and medical imaging, offering solutions to the challenges associated with manual analysis conducted through visual examination of blood smears. This approach entails distinguishing between non-parasitic stained components (such as red blood cells, white blood cells, platelets, and artifacts) and malarial parasites using visual cues. In cases where a blood sample is diagnosed as positive (indicating the presence of parasites), further capabilities are required to differentiate between species and life stages, thereby specifying the infection. Various automated methods for malaria diagnosis have been proposed to address these challenges, utilizing images extracted from blood smear pictures captured through microscopy following a staining process. Staining techniques, notably Giemsa and Leishman, play a crucial role, along with considerations regarding the thickness of the blood slide, which may be thin or thick. Thin smears typically enable the identification of specific parasitic stages and quantification of malaria parasites, whereas thick smears are preferable for the initial identification of malaria infection through blood pathology.

Malaria infected blood smears types. This image shows a comparison between staining colouration procedures and smears thickness. From left to right: thick smear with Giemsa stain, thin smear with Giemsa stain, thick smear with Leishman stain, thin smear with Leishman stain.

Dataset description

Automated analysis of malaria parasites in blood smears typically involves four distinct tasks:

• Image preprocessing: This step involves normalizing the coloration of the images, which can vary significantly from one image to another, and enhancing the contrast of different regions of interest.
• Segmentation: Algorithms are employed to separate red blood cells and/or parasites from the background and white blood cells. These algorithms utilize various cell characteristics such as shape, color, and texture.
• Feature extraction: Relevant characteristics, including shape, color, and texture, are extracted from different regions of interest to train an automatic parasite analyzer.
• Classification: Multiple classification schemes are utilized in a hierarchical manner. Initially, cells are categorized into red blood cells and white blood cells. Subsequently, red blood cells are classified based on whether they are affected by parasite(s) or not. Finally, parasites are classified according to their type and life stage. It’s important to note that parasites may also be present outside the cells, requiring a more specialized and dedicated analysis approach.

The dataset images were captured using a Leica DM2000 optical laboratory microscope at the Centre Hospitalier Universitaire Vaudois (CHUV), equipped with an integrated camera and software. This process was conducted under the supervision of seasoned radiologists, led by Dr. Guy Prod’Hom. Each image is stored in PNG format, with a resolution of 2592 × 1944 pixels and a color depth of 24 bits. All images were captured at a consistent magnification of 100×. The dataset comprises a total of 229 images, featuring four distinct types of malaria parasites: Plasmodium falciparum (122 images), Plasmodium malariae (37 images), Plasmodium ovale (29 images), and Plasmodium vivax (46 images). Each image contains at least one identifiable parasite. This dataset, curated by the authors, serves as a valuable resource for evaluating the segmentation capabilities of algorithms or classification systems. It encompasses approximately 48,000 blood cells, with malaria parasites meticulously labeled by expert radiologists. The Malaria Parasites Image Database (MP-IDB) contains a total of 840 annotated candidate parasites. Every image contain information about parasite life stages: ring stage, schizont stage, trophozoite stage and gametocyte stage.

These dataset images were all captured using the same microscope. However, many of them exhibit various issues such as non-uniform background illumination and overexposed borders, resulting from the microscope lamp’s illumination. Additionally, differences in the coloration of regions of interest may arise due to the age of the analyzed smears. Consequently, it necessitates a rigorous pre-processing step to standardize the image conditions as closely as possible, facilitating automated procedures. While the images remain comprehensible, conventional segmentation methods, such as thresholding, may be impacted by these issues.

Types of malaria parasites: from top left, clockwise, P. Falciparum in its schizont stage, P. Vivax in a gametocytes specimen, P. Malariae in its schizont stage, P. Ovale in its ring stage. All parasites have been surrounded with a yellow box. Underneath, from left to right: crops of P. Falciparum schizont, P. Vivax gametocyte, P. Ovale ring and P. Malariae schizont, taken from the boxes.