Phase-Based Classification of Malaria-Infected Parasite Red Blood Cells using Digital Holographic Imaging
Charlotte Kyeremah, Aditya S. Paul, Daniel Haehn, Manoj T Duraisinghn, Chandra S. Yelleswarapu · Microscopy and Microanalysis · 2025
Malaria remains a significant global health burden, with over 200 million cases reported annually, particularly in sub-Saharan Africa and Southeast Asia [1]. The disease is caused by Plasmodium parasites, which infect red blood cells (RBCs) and induce morphological and biochemical changes [2,3]. Early and accurate diagnosis is essential for effective treatment and disease management [1,4]. The gold standard technique for malaria diagnosis is Giemsa-stained microscopy, where trained experts manually identify Plasmodium-infected RBCs [5]. However, this method is time-consuming, labor-intensive, and subject to inter-observer variability, making it unsuitable for large-scale screening or deployment in resource-limited settings. Rapid diagnostic tests (RDTs) are immunoassay-based diagnostic tools that detect malaria antigens in blood samples. While they offer a quick and field-deployable alternative to microscopy, they suffer from limited sensitivity, particularly for low parasite densities [6]. There are high false-positive rates due to persistent antigen presence after treatment. The inability of RDTs to quantify parasitemia levels is crucial for disease monitoring. Although molecular techniques like polymerase chain reaction (PCR) offer higher sensitivity, their cost and infrastructure requirements hinder widespread adoption in resource-limited settings [7]. With advancements in computer vision and machine learning, automated image analysis has been proposed for malaria detection [8-10]. Deep learning models have accurately classified infected and uninfected RBCs using brightfield and fluorescence microscopy images [8,12]. However, these approaches have notable drawbacks – they require large, annotated datasets for training, which are difficult to obtain, and rely on intensity-based features that can be affected by variations in lighting and staining conditions. Additionally, they often function as black-box models, making it difficult to interpret classification decisions [11,12]. Automated and label-free diagnostic methods have been explored to overcome the limitations of manual microscopy [14-16]. Digital holographic microscopy (DHM) is a promising quantitative phase imaging (QPI) technique that enables the label-free characterization of RBCs by measuring optical phase shifts induced by cellular structures [17-19]. Unlike traditional intensity-based imaging, phase imaging provides information on the refractive index distribution, which correlates with intracellular changes in RBCs. This allows for the detection of subtle alterations in infected cells without requiring staining or fluorescent labeling [20]. However, while DHM has been explored for malaria detection, existing studies have primarily focused on feature extraction rather than classification [21-23]. Therefore, there is a need for a robust, phase-based classification framework that leverages these biophysical properties for automated malaria diagnosis. This study presents a novel computational framework for malaria diagnosis based on lensless inline digital holographic microscopy (LiDHM) imaging that leverages the phase-support constraint on phase-only function (PCOF) for high-accuracy phase reconstruction, enabling the robust classification of malaria-infected red blood cells (iRBCs) using phase and morphological features [24]. We compare our method against existing brightfield microscopy, evaluating reconstruction accuracy, computational efficiency, and classification performance. Our results demonstrate that PCOF-based phase recovery offers a powerful alternative to traditional staining-based diagnostics, with potential applications in automated, AI-assisted malaria screening. This work contributes to advancing cost-effective, label-free imaging solutions for global health applications. A fraction of healthy or uninfected RBCs (uRBCs) from a donor is infected with the Plasmodium falciparum HB3 strain, adhering to strict ethical and biosafety standards. After 48 hours, the cells reach the trophozoite and schizont stages, where they exhibit magnetic susceptibility due to the accumulation of paramagnetic hemozoin. Using magnetic activated cell sorting (MACS), a technique developed by Miltenyi Biotec, iRBCs are separated. Both uRBCs and magnetically sorted iRBCs were imaged using a brightfield microscope and a custom-built LiDHM, as shown in Fig.1. For the brightfield imaging, thin blood smears were prepared and stained using Giemsa-dye to confirm the infection. The recorded holograms, as shown in Fig. 2(A), were processed using PCOF, and the reconstructed phase image (Fig. 2(B)) reveals the phase shifts induced by RBCs with enhanced contrast and detailed morphological information. Once the reconstructed uRBCs were segmented, as shown in Fig. 2(C), each cell was labeled with a unique identifier for further analysis, as shown in Fig. 2(D). The segmented image was then multiplied to the reconstructed phase image to isolate the uRBCs from the background, as depicted in Fig. 2(E). We extracted various morphological and phase features from each labeled uRBCs (Fig. 2(F)), including projected surface area, sphericity, eccentricity, extent, total mass, and phase. To study the phase distribution of the iRBCs, we first quantified and standardized the phase variation of healthy RBC samples without any infection so we could use that as the baseline for the phase-based classification algorithm. We then calculated the phase distribution of the uRBCs and applied Youden’s index, full-width half maximum (FWHM), 1/e2 width, and phase mean and standard deviation to find the optimal feature values that maximized sensitivity and specificity. These feature thresholds were then used to classify iRBCs in subsequent datasets, as shown in Fig 3. The classification was validated using different datasets obtained from the same sample preparation. The feature-based classification and filtering process can be represented as Filter out irrelevant data: Keep rows where ϕ≥ϕmin Classify as “Uninfected” if: (Amin≤A≤Amax)∧(emin≤e≤Amax)∧(extmin≤ext≤extmax) Classify as “Infected” if: ϕ≥ϕmax Classify as “Not Classified” if any of the conditions are met Parasitemia: Pinfected=countofinfectedcellscountofclassifiedcells×100 where ϕ is the phase, A is the surface area, e is the eccentricity, ext is the extent, and min and max are the minimum and maximum values. Figure 3 compares the probability density distributions of the extracted features between uRBC samples (blue curves representing healthy samples with no infections) and infected RBC samples (red curves representing healthy cells with infections). The plot highlights differences between uninfected and infected RBC samples across several features. Notably, phase, mass, and eccentricity exhibit strong discriminatory potential, making them effective markers for differentiating infected cells. The pixel-related morphological parameters, such as surface area and shape, show minimal variation between infected and uninfected cells, while the phase profile exhibits a clear distinction, making it a more reliable indicator for classification. The results were analyzed and validated using descriptive statistics and a Random Forest classifier. It is shown that uRBCs appear as biconcave discs with a lighter central area due to their thinner central region. Their size and shape are uniform, and they have a consistent color due to the uniform distribution of hemoglobin, whereas malaria-infected RBCs, particularly at the trophozoite and schizont stages, exhibit size, shape, and internal structure alterations. Trophozoites are typically identified by the characteristic “ring” structure within the cell, while schizont stages present a segmented appearance as the parasite divides, which results in a decrease or increase in the size and shape. These variations in the cell’s morphology alter the phase of the light transmitted through the cells. The phase-based classification method demonstrated robust performance in distinguishing iRBCs from uRBCs using phase information. In conclusion, recognizing the crucial role of phase information in imaging techniques, we developed a novel phase feature-based classification approach for effectively distinguishing malaria-infected RBCs from uninfected RBCs using lensless inline digital holography. Our extensive analysis and quantitative evaluations across diverse infected RBC samples demonstrate that this method outperforms conventional classification techniques, offering improved accuracy and robustness in malaria diagnosis. (A) Preparation and imaging of iRBCs. Top-left and right: MACS column setup used for the enrichment of iRBCs. Bottom-left: Prepared samples labeled with corresponding experimental conditions. Bottom-right: Bright-field image of fixed and stained RBCs. (B) Schematic illustration of the sample preparation process for measuring the percentage of infected RBCs (% iRBCs). The enriched sample is transferred onto a slide for analysis and imaging using (C) the custom-built lensless inline digital holographic microscope. Image processing of the proposed method. (A) Hologram of the cropped region of interest (ROI) of RBC sample (B) Reconstructed unwrapped phase image of (A) using the PCOF method, (C) Binary image, (D) Area divided marker image by the watershed method (E) Markered image multiplied by the reconstructed phase image to isolate each cell from the background, and (F) Labeled phase unwrapped image. Kernel Density Estimation (KDE) curves illustrate the comparison of several features of Pure RBC Samples (blue curves) and Infected RBC Samples (red curves). The features include surface area, eccentricity, extent, sphericity, mass, and phase. The density plots illustrate the probability distributions of these features, highlighting differences between the two groups. Vertical lines indicate threshold values (lower and upper) for each feature, aiding in understanding typical ranges and abnormalities.