Effective diagnosis of malaria, including the detection of infections at very low density, is vital for the successful elimination of this deadly yet treatable disease. Unfortunately, existing technologies are either inexpensive but poorly sensitive - Rapid Diagnostic Tests (RDTs) and microscopy - or sensitive but costly - Polymerase Chain Reactions (PCR). Here, we demonstrate an AI-powered, reagent-free, and user-friendly approach that uses mid-infrared spectra acquired from dried blood spots to detect malaria infections with high accuracy under varying parasite densities and anaemic conditions. Our AI classifier initially trained on 4655 spectra from parasite-spiked blood samples from 70 adult volunteers, in controlled laboratory settings, attained 90% accuracy in detecting infections as low as one parasite per microlitre of blood, a sensitivity unattainable by conventional RDTs and microscopy. These classifiers seamlessly transitioned to field applicability, achieving over 80% accuracy in predicting natural Plasmodium falciparum infections in blood samples collected during a field survey in rural Tanzania. Crucially, the performance remained unaffected by various levels of anaemia, a common complication in malaria patients. These findings suggest that the AI-driven mid-infrared spectroscopy approach has the potential to evolve into a cost-effective and highly sensitive malaria-screening tool adaptable to field conditions, including low-resource settings, thereby accelerating malaria elimination worldwide.
Article
Using AI-Powered Mid-Infrared Spectroscopy for Malaria Screening Under Varying Parasite Densities and Anaemic Conditions
https://doi.org/10.21203/rs.3.rs-4111334/v1
This work is licensed under a CC BY 4.0 License
Version 1
posted
You are reading this latest preprint version
Malaria control has made significant progress in recent decades but there are still an estimated 600,000 deaths and 250 million cases annually, most of these in sub-Saharan Africa [1, 2]. To accelerate elimination efforts, effective strategies are required, both for control and for surveillance. There is an urgent need for simple, scalable and low-cost methods for monitoring key malaria metrics, so as to establish prevailing risk, evaluate impacts of control measures and assess overall progress against malaria. The World Health Organization (WHO) underscores the need for evidence-based and context-specific control strategies – which require sensitive, rapid, and affordable screening tools that are deployable at scale even in low-income or remote settings [3]. This includes the need for accurate detection of malaria infections – both at clinical points of care settings and in population surveys.
While rapid diagnostic tests (RDTs), microscopy, and polymerase chain reactions (PCR) have played a crucial role in the diagnosis and management of malaria across endemic and elimination situations [4], these approaches still have major limitations in most settings. For example, RDTs have transformed malaria case management across Africa due to their practicality and affordability [5, 6], yet studies show that as many as half of children in some poor communities still lack access to these tests [3, 7]. Their effectiveness is also increasingly threatened by deletions on the histidine-rich protein 2 and 3 (HRP II and III) genes, which code for the protein targets of most P. falciparum RDTs [8, 9]. Additionally, both microscopy and RDTs have poor sensitivity when parasite densities fall below 50–100 parasites/µL of blood [3, 4, 10–12]; and can miss 20–50% of infections in population surveys[13]. From an operational perspective, microscopy requires electrical power sources, can take as long as 30 minutes to run [12], is prone to subjective interpretations of results and can be expensive due to personnel costs and reagents [14]. Unfortunately, even the highly sensitive techniques, such as PCR, enzyme-linked immunosorbent assays (ELISA), and loop-mediated isothermal amplification (LAMP), which detect as few as 5 parasites/µL of blood, with greater accuracy, are expensive, and require highly skilled workers.
In recent years, infrared spectroscopy (IR) in both the near (NIR) and mid (MIR) infrared ranges has shown substantial promise for monitoring key entomological and parasitological indicators of malaria - including detecting malaria infections in human blood [15]. The technology is rapid, robust, reagent-free, and requires minimum skills to operate [16]. Infrared spectroscopy involves shining infrared light through biological specimens to infer their biochemical composition, which can be analysed using different mathematical techniques to distinguish meaningful traits such as infection status, age, and species. Given the complexities of assigning spectroscopic bands for interpreting patterns associated with malaria infections, recent studies have been integrating spectroscopy with either multivariate analysis or machine learning (ML) approaches to more effectively interpret the key biological traits [17, 18]. Infrared spectroscopy combined with machine learning (IR-ML), outperforms the traditional analyses in detecting malaria by efficiently deciphering complex, non-linear, multi-correlated spectral signals and exhibiting high accuracy [19]. For example, during the intra-erythrocytic cycle, malaria parasites generate distinctive biomarkers, which are evident on infrared spectra and can be used to identify infected individuals [16, 20–23]. Further, models have been reported to achieve over 90% precision in differentiating infected from uninfected blood samples [15, 23, 24].
Despite IR-ML techniques demonstrating potential in identifying malaria infections in whole blood or isolated RBCs, critical gaps remain to be addressed, one of which is the need to establish the lower limits of parasite detections and quantifications (LLOD & LLOQ) [15]. One study has shown that MIR could detect malaria infections at less than 5 malaria parasites/µL of blood in isolated red blood cells (RBCs) [20]. However, no tests have been done on the actual detection thresholds critical for applications of IR-ML in either point-of-care tests or population surveys using whole blood specimens.
Another key gap is the need to identify and estimate the impacts of confounding factors. In areas with high malaria transmission, infected individuals often exhibit decreased haemoglobin concentrations, other parasitic infections, nutritional deficiencies or other concomitant factors [25–27]; all of which may compromise tools such as IR-ML that are designed to detect biochemical changes resulting from malaria infections. Indeed, previous MIR-ML models trained for malaria detection may have been influenced by the association of co-variates (e.g., anaemia and immune response to parasites) rather than true parasite signals [15]. Immunological variations from persistent asymptomatic infections [28] may also affect performance of IR-ML results for malaria detection, in ways not detectable in studies that use blood bank samples from limited numbers of volunteers [20, 23]. ML algorithms should therefore consider these variabilities, e.g. by incorporating varying anaemic, nutritional or immunologic conditions but also train models with different parasite ‘contrasts’, to disentangle true infections more effectively from these noise signals.
The aim of this current study was therefore to assess the lower limit of malaria parasite detection on dried blood spots (DBS) using MIR-ML approaches, investigate the impact of anaemic conditions on detecting malaria infections and demonstrate the best ML training approaches, whether utilizing high, low, or combined parasite concentrations, and finally test them on spectra of actual malaria-infected patients. The DBS were generated using whole blood from 70 malaria-free volunteers to capture immunological variations, and were spiked with cultured ring-stage P. falciparum at various parasitaemia and haematocrit ratios to mimic anaemia. ML algorithms were then trained on MIR spectra acquired from laboratory-generated DBS, where specific malaria parasite signals could be learned. MIRs-ML, trained with high-contrast parasite concentrations against negatives, demonstrated a detection accuracy of over 90% in laboratory tests across various parasitaemia levels; in field-collected samples, the accuracy consistently remained above 80%.
Generating samples of different parasite densities and anaemic conditions.
In order to identify threshold limits for detecting malaria parasites and assess the impact of anaemia on the predictive accuracy of MIRs-ML for malaria infections, we initially cultured two strains of P. falciparum, FCR3 and NF54. We then diluted their ring-stage with malaria-negative blood from seventy volunteers to generate 4 different parasite densities (Fig. 1). We selected the two lowest malaria parasitemias (0.002% or 50–100 parasites/µL and 0.00003% or 1–5 parasites/µL to correspond to the detection limits of RDT/microscopy and PCR, respectively [4, 12, 29]. Additionally, parasitaemias of 6% and 0.1% were chosen to capture the highest parasite contrasts in the training dataset. Haematocrit was set at 50, 25, and 12.5% to represent normal, moderate, and severe anaemia, respectively. Thus, a two-way matrix consisting of four malaria-positive parasitaemia levels (6%, 0.1%, 0.002%, and 0.00003%) and a negative class (0%), along with three anaemic classes, was generated (Table S1). In total, we created and scanned 4,655 DBS using an MIR spectrometer, which were used to train and evaluate ML classifiers. Samples were randomly selected to confirm the success of dilutions (Figure S2 and S3).
Selection and training of machine learning classifiers
We analysed 4,559 spectra with varying malaria parasite densities and hematocrit ratios. Of these, 12 spectra from the non-anaemic samples, 35 from moderately anaemic samples, and 49 from severely anaemic samples were discarded due to excessive water content resulting from plasma and atmospheric interference (Table S2). The mean spectra of each parasitaemia class at different anaemic conditions revealed characteristic biochemical signatures of P. falciparum infections such as amide, lipids, and hemozoin (Figure S1) [30]. To ensure ML classifiers would learn features associated with malaria infections, an infrared region without the key biochemical information (from 2799 to 1801 cm⁻¹) was eliminated [20], and the regions 3100–2800 cm⁻¹ and 1800–900 cm⁻¹ were selected to train and validate ML classifiers.
Model selection and training were performed using 70% of the data, keeping 30% of the data aside for validation. First, using 10-fold stratified shuffle split cross-validation (SSS-CV), we tested 7 ML classifiers to evaluate the best performer for classifying malaria infections. To establish the most suited data to train the models, we tested three different training sets: (i) high contrast at the highest parasitaemia (6% iRBCs) against the negative class; (ii) All concentrations where all malaria parasitaemia (6%, 0.1%, 0.002%, and 0.00003%) grouped as a positive class against the malaria-negative class; and (iii) low contrast (positive malaria = 0.00003%) against the negative class. Logistic regression (LR) achieved the highest SSS-CV accuracy score in all three datasets, achieving 97.61%, 71.14% and 70.87% in high contrast, all concentrations combined and low contrast training sets, respectively (Fig. 2a-c).
We retained the LR algorithm for further tuning, utilizing a grid search approach over an extensive range of hyperparameter values. The tuned model, when fitted to high-contrast scenarios characterized by a 6% parasitaemia level versus non-infected samples, demonstrated remarkable capacity to distinguish malaria-positive and -negative samples, achieving accuracies of 100% and 91%, respectively (Fig. 2d). When evaluating the performance of the tuned models across all parasitaemia concentrations, the accuracy rates for correctly identifying positive and negative samples were 82% and 73%, respectively (Fig. 3e). Even in the context of low-contrast conditions with the minimal 0.0003% parasitaemia level, the models maintained strong performance, with positive and negative detection accuracies of 78% and 65%, respectively (Fig. 3f).
Even after tuning the LR to optimize its hyperpartemeter, training with high contrast had the largest area under the curve (AUC) estimate, 0.98 (Fig. 3g), outperforming models trained using both the combined concentrations and low contrast data sets, for which the AUC scores were 0.85 and 0.82, respectively (Figs. 3h-i). The train-test size, precision, recall, and F1-score of the three-trained models on the validation set are summarized in Table 1.
|
Classifier 01: Trained on High contrast (6%) against 0% |
|||||
|---|---|---|---|---|---|
|
Sample size |
Precision score (%) |
Recall score (%) |
F1-Score (%) |
||
|
Train size |
460 |
Negative class |
92 |
100 |
96 |
|
Test size |
92 |
Positive class |
100 |
91 |
95 |
|
Classifier 02: Trained on all concentrations (6%, 0.1%, 0.002%, 0.00003%) against 0% |
|||||
|
Sample size |
Precision score (%) |
Recall score (%) |
F1-score (%) |
||
|
Train size |
440 |
Negative class |
75 |
82 |
78 |
|
Test size |
88 |
Positive class |
80 |
73 |
76 |
|
Classifier 03: Trained on Low contrast (0.00003%) against (0%) |
|||||
|
Sample size |
Precision score (%) |
Recall (%) |
F1-score (%) |
||
|
Train size |
452 |
Negative class |
69 |
78 |
73 |
|
Test size |
91 |
Positive class |
75 |
65 |
70 |
In addition, to understand the specific biochemical contribution influencing predictions, we also extracted the wavenumber values and corresponding coefficients that most influenced the performance of our three trained LR classifiers. All three classifiers have learned from wavenumber values associated with biochemical signals associated primarily with lipids (3100 − 2800 cm⁻¹) and proteins (1800 − 600 cm⁻¹) (Fig. 3a-c), consistent with signals expected from malaria infections [20].
The highest contrast model appears to have primarily learned to classify malaria infections from the wavenumbers associated with proteins, notably amide III and amide I vibrations (1185–1191 cm⁻¹, 1199–1201 cm⁻¹, 1619 − 1617 cm⁻¹ and 1687 − 1671 cm⁻¹) which are indicative of the secondary structure of proteins. Wavenumber values 997, 1125–1139 cm⁻¹ highlight the models’ use of C-N stretching vibrations of proteins (peptides) and phosphodiester stretching (1201, 1071, and 1263 cm⁻¹) indicative of nucleic acids (Fig. 4a). The high-contrast model effectively identified wavenumbers linked to lipids, alkanes, and carbohydrates, indicated by key C-H stretching vibrations at 2917 cm⁻¹, 2843 cm⁻¹, 2967 cm⁻¹, and 2955 cm⁻¹, and C-H bending vibrations at 1491 cm⁻¹ and 1493 cm⁻¹. Optimal wavenumber values for the other logistic regression models, covering all concentrations and low-contrast scenarios, are detailed in the supplementary materials.
Detection of malaria infections at different parasitaemia and anaemic conditions
After establishing that the high contrast training set had the best performance in predicting a dataset of the same parasitaemia levels, we validated the accuracy of the three-trained LR algorithms against all concentrations and anaemic conditions using 100 bootstrap random resampling. For this purpose, the 30% of the full dataset held out separately before training was used.
The MIRs-ML trained on the high contrast data set against the negative class identified malaria infections in non-anaemic samples with 100% accuracy for the two highest parasitaemia levels tested (6% and 0.1%). Accuracy dropped to 92 − 91% when classifying intermediate and low-level parasitaemia, and when evaluating combined spectra from all concentrations (Fig. 4a). Similarly, high accuracies were observed when the high-contrast model was used to predict parasite infections at moderate and severe anaemia (Figs. 4b-c).
We also validated the models trained on all combined concentrations and those trained using low contrast for predicting various parasitaemias in an unseen dataset. Regardless of anaemia, MIRs-ML trained on all concentrations accurately classified different parasitaemias, achieving an average accuracy of 75.33% [70.07–80.8%] (Fig. 4a-c). The model trained on the lowest contrast demonstrated an average accuracy of 62.73% [56.53-74.00%] across different parasitaemias and anaemia. When validating the low-contrast model on different parasitaemias and anaemic conditions in new data, we noticed increased accuracy in predicting parasitaemia levels resembling those in the training set (moderate and low parasitaemia) (Fig. 4a-c).Overall, the high-contrast model outperformed the other two models in predicting malaria infections at all tested parasitaemia and anaemic conditions (Fig. 4d-f), suggesting that this approach is the most robust for achieving generalizability.
We fitted a generalized linear model to estimate the impact of training methodologies, levels of parasitaemia, and the presence of anaemia on the predictive accuracy of MIRs-ML in diagnosing malaria infections. The analysis revealed that neither anaemia nor the different levels of parasitaemia significantly affected the capabilities of the MIRs-ML models for distinguishing between infected and non-infected samples (anaemic conditions: χ² = 0.01, p = 0.99; parasite intensities: χ² = 0.24, p = 0.99). However, the choice of training methodology significantly affected model performance (χ² = 201.62, p < 0.0001). Further analysis using Post-Hoc Tukey's test to compare the three training methodologies showed that training with high-contrast samples notably enhanced the predictive accuracy of MIRs-ML, yielding a mean difference of 19.93% (p < 0.001) compared to training with all concentrations, and a mean difference of 32.53% (p < 0.001) compared to training with low-contrast samples. Additionally, training with all concentrations demonstrated a significant positive effect, with a mean difference of 12.59% (p < 0.001), when compared to training with low-contrast samples.
Validation of the MIRs-ML for identifying malaria infections using spectra from field-collected dry blood spots.
We evaluated the performance of laboratory-trained models using realistic samples collected from patients. To facilitate comparison, we tested the created models under two scenarios. Firstly, we employed laboratory data, as detailed in the preceding sections. Secondly, we utilized a realistic dataset obtained from the field (see Fig. 5a). In simulations resembling field conditions, where all parasitaemia levels generated in the lab were combined and treated as positive, the high-contrast trained model distinguished positive from negative samples with accuracies of 88% and 92%, respectively (Fig. 5b).
Next, we obtained and scanned 252 DBS samples collected from a previous field survey in southeastern Tanzania. The malaria infection status in these samples had been confirmed using nested PCR [22]. In these tests, the model with the highest contrast predicted malaria infections with an accuracy of 83%, achieving 86% accuracy for the positive class and 79% accuracy for the negative class (refer to Figs. 5c).The precision, recall, and F1-score of the three trained models when tested with the field-collected dataset are summarized in Table 2.
|
Classifier 01: Trained on High contrast (6%) against 0% |
|||
|---|---|---|---|
|
Precision score (%) |
Recall score (%) |
F1-Score (%) |
|
|
Negative class |
84 |
86 |
85 |
|
Positive class |
81 |
79 |
80 |
|
Classifier 02: Trained on combined positive (6%, 0.1%, 0.002%, 0.00003%) against 0% |
|||
|
Precision score (%) |
Recall score (%) |
F1-score (%) |
|
|
Negative class |
58 |
59 |
58 |
|
Positive class |
51 |
50 |
50 |
|
Classifier 03: Trained on Lowest parasitaemia (0.00003%) against (0%) |
|||
|
Precision score (%) |
Recall (%) |
F1-score (%) |
|
|
Negative class |
57 |
58 |
57 |
|
Positive class |
57 |
56 |
57 |
To understand the potential applications of MIRs-ML for either point-of-care or population surveys, we computed false negatives and false positives of the highest contrast model in predicting all parasitaemias in laboratory specimens and field datasets (Fig. 5d). The model exhibited a prediction error of 14% for false negatives and 21% for false positives in field samples, preliminary suggesting that this approach could be most useful in population surveys, especially in low transmission settings, due to the lower rate of false negatives.
Effective malaria screening is crucial for guiding elimination efforts, especially in detecting low-density parasitaemia. However, current methods often face trade-offs between cost and sensitivity. Alternatively, infrared spectroscopy coupled with machine learning (MIR-ML) is showing great promise, particularly in laboratory settings. Here, we found that the MIR-ML approach can detect as few as 1–5 parasites/µL of blood with an overall accuracy exceeding 90% under laboratory settings. Further, ML algorithms trained using the highest plausible parasite concentrations against negative samples (“high contrast” training set) yielded a potent and robust ML classifier capable of predicting malaria infections at various parasitaemia levels in the presence or absence of anaemia. Anaemic conditions, although highly correlated with malaria infections, did not impact the performance of the approach for predicting malaria infections. Notably, the ML model trained with laboratory-generated high contrast dataset accurately classified over 80% of field-collected specimens obtained from malaria-infected patients. Since these specimens were entirely unseen by the model, this indicates good generalisability of this approach and holds great promise for its future optimization in malaria detection, especially for field surveys.
This study demonstrates the capability of MIRs-ML to detect low parasite concentrations that are often missed by methods such as microscopy and RDTs [31–41]. Additionally, it improves upon earlier studies that applied multivariate analysis, including partial least squares and classical regression approaches, to detect and quantify malaria infections in isolated infected and uninfected RBCs [20] by revealing that the integration of MIRs-ML can detect as low as 1 parasite/µL of blood with an accuracy exceeding 90% in DBS, even in the presence of anaemia, and requiring minimal sample handling.
While malaria is the primary cause of anaemia in regions with high transmission risks, other infections, including gastrointestinal helminths, can also cause anaemia, primarily affecting children and the elderly [42–44]. If MIRs-ML detects anaemia (e.g. where anaemia may be confounded with malaria presence), there could be a risk of misdiagnosis disproportionality affecting children, who are more vulnerable to the effects of misdiagnosed malaria [25]. We therefore experimentally manipulated anaemia levels in by adjusting plasma to RBC ratios using blood from volunteers. MIRs-ML trained on the high contrast dataset predicted malaria infections in the presence of severe anaemia, with accuracy exceeding 90% even at the lowest parasitaemia, indicating that the predictive capabilities of IR-ML are not affected by anaemia.
A further potential problem for IR-ML technologies is poor generalizability to new datasets, as observed in entomological traits of malaria vectors due to genetic, environmental, and dietary variation [15] [17]. Previous studies involving the training of IR-ML approaches for malaria detection in blood used low contrast or less immunologically diverse blood to spike with parasite culture [20]. Here, we trained our models on blood samples from 70 individuals to capture population immunological variability, experimentally spiked with cultured ring-stage P. falciparum parasites. This allowed models to achieve over 80% accuracy in diagnosing malaria infections with field data without further training. To further improve generalizability, techniques such as transfer and partial learning [17, 45] could be useful for improving the predictive accuracy of this approach, especially in diverse population surveys.
The trained algorithms in this study learned from peaks associated with the biochemical signatures of malaria-infected samples, particularly those suggested to be linked with the byproducts of the malaria ring-stage parasite, such as lipids, proteins (hemozoin) [30], and parasite DNA. In this study, the blood sample was prepared by removing the buffy coat and white blood cells by suction when separating plasma from RBCs for anaemic adjustment. Therefore, the host cells present in the sample were RBCs, which are enucleated. Thus, the signal observed corresponding to phosphodiester stretching (997, 1201, 1071, and 1263 cm⁻¹) are likely due to parasite nucleic acids only. Further, differences in the peaks related to the amide groups of the proteins were identified. Specifically, the band between 3500 and 3000 cm⁻¹, commonly referred to as Amide A and associated with the stretching of N-H bonds; the band between 1700 and 1600 cm⁻¹ (Amide I), primarily linked to the C = O stretching vibration and sensitive to the secondary structure of the protein; the band between 1600 and 1500 cm⁻¹ (Amide II), mainly due to N-H in-plane bending and C-N stretching vibrations; and the band between 1350 and 1200 cm⁻¹ (Amide III), which is a combination of C-N stretching, N-H in-plane bending, and C-H in-plane bending vibrations, appear crucial in capturing the variations due to malaria infections. We also observed changes in the peaks associated with the C-H stretching modes of the lipids (2901, 2905, and 2859 cm⁻¹).
The advantage of using MIRs-ML for malaria parasite detection relies on its higher sensitivity and accuracy cost ratio, demonstrating potential applicability in low-income communities [15, 46]. Standard low-cost maintenance, such as providing desiccants to limit humidity effects, is the only requirement for operation, with no need for additional reagents. This study demonstrated an added advantage of the approach, its capability to detect as low as one parasite/µL of blood in the presence and absence of anaemia, highlighting the potential use-case of MIRs-ML for malaria parasite detection in field settings. One important question to consider is where MIRs-ML should be utilized, either as a point-of-care test or for population surveys. Using a field-collected dataset, this study revealed that MIRs-ML algorithms detected unseen field-collected specimens with a lower false-negative rate of 14% compared to a false-positive rate of 21%. This might primarily indicate a potential use case of MIRs-ML for population surveys that could complement PCR confirmation of positive MIRs-ML samples. However, these should be considered as preliminary observations and for the further integration of MIRs-ML into routine malaria surveillance, large-scale clinical trials are essential to validate the feasibility of MIRs-ML for malaria parasite detection in vertebrate hosts.
We solely focused on detecting the P. falciparum parasite, prevalent in endemic regions. It is crucial to explore detection limits and infrared signals associated with infections from P. ovale, P. vivax, and P. malariae. Future studies may also explore the performance of MIRs-ML in detecting malaria infections across various epidemiological strata, as the current tools exhibit different performance characteristics based on transmission patterns, with RDTs and microscopy showing poor performance in low-transmission areas. Additionally, it is important to further investigate malaria diagnostic interference of MIRs-ML due to other confounding factors, such as coinfection and nutritional factors.
To advance malaria elimination, there is a critical need for screening methods that combine cost-effectiveness with the ability to detect low-density parasitaemia, a gap not filled by current methods like RDTs, microscopy, or expensive PCR. In this study, we have demonstrated an AI-driven mid-infrared spectroscopy technique that excels in identifying malaria infections in dried blood spots, achieving high accuracy across various parasite densities. The performance of the technique was not compromised by anaemia, which is a frequent complication in malaria patients. Moreover, when MIRs-ML models initially trained on laboratory data were used to distinguish between infected and uninfected field samples collected from a field survey in rural Tanzania, the classification accuracies were maintained above 80%. These findings indicate the significant potential and viability of this AI-driven mid-infrared spectroscopy technique as an affordable and malaria-scalable screening tool in low-resource settings. Further validation in diverse areas and the consideration of additional confounders, such as co-infections, are necessary to further validate the approach in real-life settings.
Malaria parasite cultures
Sample collection and parasite cultures were completed at Ifakara Health Institute’s laboratories in Bagamoyo, Tanzania using a protocol adapted from Trager and Jensen [47], with slight adjustments. Group O + blood was obtained from four malaria-free volunteers, and kept in tubes containing the anticoagulant, ethylenediaminetetraacetic acid (EDTA) for continuous culturing of P. falciparum strains NF54 and FCR3. The blood was centrifuged at 2500 rpm at 24°C for 10 minutes to obtain RBCs. The RBCs were then washed, diluted to 50% haematocrit with uncomplemented media (Roswell Park Memorial Institute (RPMI) 1640 media supplemented with hypoxanthine, neomycin and 4-(2-hydroxyethyl)-1-piperazineethanesulfonic acid (HEPES), and stored at 4°C.
The washed RBCs were used to culture P. falciparum in vitro for up to seven days. The asexual malaria parasites were grown in uninfected washed O + RBCs as host cells at 5% haematocrit, maintained in RPMI-1640 medium supplemented with 25 mM HEPES, 100 µM hypoxanthine, neomycin, 5% Albumax II and 24 mM Sodium bicarbonate. The parasite culture was gassed with a mixture of 5% CO2, 3% O2, and 93% N2 and incubated at 37°C. The culture was examined daily for parasitaemia estimation using field-stained (Hemacolor® rapid staining) thin blood smears under a compound microscope in 10 fields. Two rounds of parasite synchronization was performed to ensure the remaining of only ring-stage P. falciparum [48]. The culture was kept until the ring-stage parasitaemia level reached > 6% and was used for experimental dilutions. Parallel malaria-free cultures with only media and O + RBCs from the same volunteers were kept to create controls. The process was repeated for up to 11 batches, until we recruited and diluted blood from 70 volunteers.
Recruitment of malaria-free volunteers
We recruited malaria-free individuals from tertiary-level colleges in Bagamoyo, eastern Tanzania, following sensitization meetings, during which we explained the objectives, procedures, potential risks and benefits of the study. Participants who expressed interest were contacted by phone and requested to provide informed consent. The consenting participants were screened for malaria parasites using RDTs by taking a finger-prick blood sample, followed by a confirmatory quantitative polymerase chain reaction (qPCR) [49, 50]. Participants who tested negative for malaria were enrolled in the study, while those who tested positive were treated following Tanzania's malaria treatment guidelines and excluded from the study [51]. A total of 70 volunteer participants were recruited, all between 20 and 40 years old. A total of 40 mL venous blood was drawn in an EDTA tube from each participant and used for laboratory dilutions of cultured malaria parasites to create different malaria parasitaemia and haematocrit ratios.
Haematocrit dilutions to mimic anaemic conditions
For each participant, we made two sets of haematocrit dilutions to simulate different anaemic conditions for both infected and un-infected blood. One set had malaria-free blood at 50%, 25%, and 12.5% haematocrit content, while the other comprised infected red blood cells (iRBCs) from cultured parasites, adjusted to 50%, 25%, and 12.5% haematocrit ratios using respective plasma. For uninfected blood, 40 ml of venous blood from each participant was split into 5 ml and 35 ml portions and centrifuged to separate plasma from RBCs. After separation, plasma was transferred to empty 1.5mL tubes for haematocrit dilutions. RBCs from the 5 ml portion were used to formulate a 50% haematocrit stock solution by adding plasma from the same volunteer. Serial dilutions was done by transferring 2.5 ml of the stock solution to a second tube, and adding 2.5 ml of previously obtained plasma to simulate moderate anaemia (25% haematocrit) and severe anaemia (12.5% haematocrit) conditions (Fig. 1). On the other hand, for infected blood, when the culture reached > 6% ring stage parasitaemia, it was centrifuged to separate iRBCs from the culture media and washed twice. Washed iRBCs volume was 0.5 ml, which was mixed with an equal volume of participant plasma (0.5 ml) to create a 50% haematocrit ratio stock solution; and serially diluted to 25% and 12.5% solutions (Fig. 1).
Serial dilutions for different parasite densities and controls
The parasite dilutions were performed at the respective haematocrit ratios. Initially, the cultured parasitaemia was standardized to 6% stock – and in cases where the initial densities were higher, this was lowered to the 6% densities. Serial dilutions were done to create three additional parasitaemia levels, i.e. 0.1%, 0.002% and 0.00003%, were created from the stock solution.
The control group included malaria-negative samples at haematocrits of 50%, 25%, and 12.5% prepared from uninfected RBC from the culture with plasma from same individual participants. To ensure RBC distribution matched malaria parasitaemia dilutions, uninfected haematocrits from the culture were diluted with uninfected haematocrits from participants in volumes equal to the parasitaemia dilutions (Table S1).
Preparation of dried blood spots
For each individual volunteer, we created five replicates of dry blood spots (DBS) for each parasite density at each haematocrit level, resulting in a total of 75 DBS per participant. For each DBS, 50 µL of blood was added on the circular spot on the Whatman™ paper cards. The cards were air-dried for up to 3 hours and labelled with batch number, date, parasitaemia levels, haematocrit ratio, and participant ID. To prevent cross-contamination, each card was sealed in a plastic bag. The cards were then grouped by participant ID and stored in a cool environment in a larger bag with desiccant packets and a humidity card, awaiting transportation to another Ifakara Health Institute’s facility, the VectorSphere, at Ifakara, Tanzania for infrared scanning. During transport, the bags were kept in a cooler box with ice packs separated by plastic sheeting.
Acquisition and pre-processing of infrared spectra
The DBS were individually scanned using a Fourier transform infrared spectrometer (FT-IR) with a wavenumber range of 4000 − 500 cm⁻¹ and a resolution of 2 cm⁻¹. The instrument employed was a compact Bruker Alpha FTIR interferometer, equipped with a Platinum-Attenuated Total Reflectance (ATR) module that incorporates diamond crystals. Scanning was done after 3–5 days of DBS storage. Each blood spot was punched and placed on the diamond crystal. For scanning, each blood spot was excised, positioned on the diamond crystal, and subjected to pressure using an anvil to enhance the contact area with the crystal, thereby optimizing the depth of light penetration. Each spot was scanned 32 times to obtain an averaged spectrum, which was labelled by project initial, study site, participant ID, haematocrit ratio, parasitaemia ratio, and dates.
To quantify the depth of light penetration (d) in whole blood, we employed a theoretical approach that considers the wavelength of light (λ), incidence angle (θ), and the refractive indices of whole blood (n₁) and the diamond crystal (n₂) used in the spectrometer. The penetration depth (d) was calculated using the formula.
Given that the incidence angle (θ) was fixed at 45º, the refractive index of whole blood (n₁) was determined using the Sellmeier equation (with λ expressed in micrometres) [52]:
Similarly, the refractive index of the diamond crystal (n₂) was ascertained through its corresponding Sellmeier equation [16]:
The acquired spectra were then pre-processed using a Python program to compensate for atmospheric interferences, water vapour and carbon dioxide (CO2). The pre-processed spectra were subsequently used for training, testing, and validating machine-learning algorithms.
Selection of machine learning models
Machine learning analysis was conducted using the Python programming language, version 3.9. Employing a supervised ML classification approach, seven classifiers were evaluated: Logistic Regression (LR), Support Vector Machine (SVM), Random Forest (RF), Gradient Boosting (XBG), Decision Tree (DT), Extra Tree (ET) Classifier, and Bagging Classifier (BC).
The non-anaemic class (50% haematocrit) was utilized for ML algorithm selection and training, while the two other haematocrit ratios (25% and 12.5%) were kept separate and used to assess the impact of anaemia on the ability of the models to classify infected versus non-infected specimen from sets of previously unseen spectra (Table S2). To do this, the non-anaemia data was shuffled and split into two portions; 70% for model selection, training, and testing, while 30% were kept separate as an unseen dataset for validating the trained model. Further, 70% portion were divided into 80–20% train-test split, respectively.
We employed stratified shuffle split, 10-fold cross-validation (SSS-CV) to select the best machine-learning algorithm for identifying malaria infections. The seven mentioned algorithms were evaluated, and the best one was selected based on accuracy scores to distinguish malaria infections within the non-anaemia class using three approaches: (i) Cross-validation using datasets with high contrast (Positive class = 6%, N = 230) against the negative class (Negative = 0%, N = 230); (ii) Cross-validation with all concentrations (6%, 0.1%, 0.002%, and 0.00003%) combined as the positive class (N = 220) against the malaria-negative class (N = 220); and lastly, (iii) Cross-validation with low contrast (positive class = 0.00003%, N = 226) against the negative class (N = 226) datasets.
Training, testing and validation of machine learning models to identify malaria parasite presence in non-anaemic spiked blood
The best ML algorithms selected through SSS-CV were then trained on 80% of the spectra data from non-anaemic blood using three distinct approaches: (i) High Contrast: Models were trained using infrared spectra derived from DBS with the highest parasite densities (6%) as positive samples and non-infected blood as negatives. (ii) All Concentrations: Models were trained using combined infrared spectra data from DBS with any parasite densities as positives, and non-infected DBS served as negatives. (iii) Low Contrast: Models were trained using infrared spectra derived from DBS with the lowest parasite densities (0.00003%) as positives, and non-infected blood as negatives. The trained models underwent fine-tuning through Grid Search for optimal hyperparameter optimization.
For testing purposes, the final tuned classifiers were tested on a similar parasitaemia class used for training. For instance, the model trained on 80% of the data with high contrast at 6% against negatives was also tested on the remaining 20% of the data at the highest contrast against negative classes. In addition to accuracy, other evaluation measures such as sensitivity, specificity, recall, and F1-score on the test set (20%) were calculated. Finally, the best classifiers were validated on a completely unseen dataset, the 30% kept separate at the start. Beginning with non-anaemic classes at different parasitaemia levels, the total number of DBS included for model validation in the non-anaemic class with various parasitaemia levels were as follows: validated model performance on predicting malaria infections at non-anaemic conditions when positive at 6% parasitaemia (N = 82) against negative 0% (N = 82); then validated positive at 0.1% (N = 82) against negative 0% (N = 82), followed by positive at 0.002% (N = 82) against negative 0% (N = 82), and finally, positive at 0.00003% parasitaemia (N = 82) against negative 0% (N = 82).
Evaluating the effect of anaemia on performance of MIR-ML for distinguishing between blood samples with and without malaria parasites
The best classifiers developed for predicting malaria infections without considering anaemia were evaluated on a new dataset comprising cases of moderate anaemia (with a haematocrit level of 25%) and severe anaemia (with a haematocrit level of 12.5%). This evaluation was structured across four distinct categories: (1) malaria-positive with a 6% parasitaemia rate (N = 101) versus malaria-negative with a 0% parasitaemia rate (N = 101), (2) malaria-positive with a 0.1% parasitaemia rate (N = 101) versus malaria-negative (N = 101), (3) malaria-positive with a 0.002% parasitaemia rate (N = 101) versus malaria-negative (N = 101), and (4) malaria-positive with a 0.00003% parasitaemia rate (N = 101) versus malaria-negative (N = 101). We computed the accuracy, sensitivity, and specificity of these models using a bootstrap method with 100 iterations and establishing a 95% confidence interval for these metrics. Finally, a generalized linear model was used to test the statistically significant effect of anaemia and parasitaemia on the performance of MIRs-ML approaches in predicting malaria infections.
Acknowledgements
We would like to express our gratitude to our project administrators Rukiyah M. Njambalaha, Rose P. Mallya, and Ritha R. Kidyala for their assistance with project logistics. Special thanks go to our medical personnel, Isaac Rutha, Rajabu M. Sayi, Mohamed S. Chabo, and the nurses at the clinical trial facility of IHI in Bagamoyo for their invaluable support in blood collection from our participants. Heartfelt appreciation is extended to our colleagues in the molecular laboratory facility, especially Ibrahim Sasamalo, Wilmina Kalinga, Reward Goodluck, Alfred B. Lazaro and Nassoro Lilolime, for their technical assistance in ensuring the quality assurance of the samples generated throughout the study. This work involved the contributions of many colleagues from Tanzania, particularly Bagamoyo. Primarily, we express our gratitude to all tertiary colleges’ students who voluntarily donated blood for our laboratory dilutions; their willingness and friendliness made all these demonstrations possible. Recognizing that quality science is a collaborative effort, we extend our thanks to everyone involved.
Author contributions
Conceptualization: I.H.M., F.O.O., F.B. and S.A.B. Data curation: I.H.M., F.J.M, R.Y.M., L.M.H., P.A.K. and M.M.T., Formal analysis: I.H.M., S.A.B., F.O.O., F.B and M.G.-J.
Funding acquisition: F.O.O, S.A.B., F.B. and I.H.M. Investigation: I.H.M., S.A.B., F.O.O., F.B, M.G.-J, F.J.M, R.Y.M., R.H.., P.K. and M.M.T. Methodology: I.H.M., S.A.B., F.O.O., F.B. Project administration: I.H.M., F.J.M., S.A.B., F.O.O., F.B. Resources: S.J.M., R.Y.M., L.M.H., P.A.K., M.M.T., E.P.M, D,J,S., I.H.M., F.J.M., S.A.B., F.O.O. and F.B. Software: I.H.M, S.A.B and M.G.-J. Supervision: S.A.B., F.O.O. and F.B. Validation: S.A.B., F.O.O. and F.B. Visualizations: I.H.M., S.A.B., F.O.O. and F.B. Writing original draft: I.H.M., S.A.B., F.O.O. and F.B. Writing review & editing all authors.
Funding
This work was supported in part by the Bill & Melinda Gates Foundation (Grant number Grant No. OPP 1217647 to Ifakara Health Institute). Under the grant conditions of the Foundation, a Creative Commons Attribution 4.0 Generic License has already been assigned to the Author Accepted Manuscript version that might arise from this submission. Support was also received from Rudolf Geigy Foundation through Swiss Tropical & Public Health Institute (to Ifakara Health Institute), Howard Hughes Medical Institute (HHMI) and Gates Foundation (Grants: OPP1099295) awarded to FOO and HMF, Royal Society (Grant No ICA/R1/191238 to SAB University of Glasgow and Ifakara Health Institute), Academy Medical Science Springboard Award (ref: SBF007\100094) to FB.
Data availability
The dataset supporting the findings is available upon a reasonable request, which can be directed to either the corresponding author, the Ifakara Health Institute ethical review board in Tanzania or National Health Research Ethics Committee in Tanzania with reference to ethical clearance certificate of NIMR/HQ/R.8a/Vol. 1X/3735.
Code availability
All code and their respective output for machine learning and statistical analysis are available at https://github.com/Issamshani17/MIRs-ML-Limit_of_detection
Ethics approval and consent to participate
This work was approved under Ifakara Health Institute Review Board (Ref: IHI/IRB/No: 1 2021) and the Medical Research Coordinating Committee (MRCC) at the National Institute for Medical Research-NIMR (NIMR/HQ/R.8a/Vol. 1X/3735).
- WHO. Global Malaria Elimination programme, Certification of malaria elimination, 1955–2023.
- World Health Organization (WHO). World malaria report. 2023.
- UNITAID. 2015 Malaria Diagnostics Landscape Update. 2015.
- Moody A. Rapid diagnostic tests for malaria parasites. Clin Microbiol Rev. 2002;15: 66–78.
- Cunningham J, Jones S, Gatton ML, Barnwell JW, Cheng Q, Chiodini PL, et al. A review of the WHO malaria rapid diagnostic test product testing programme (2008-2018): Performance, procurement and policy. Malar J. 2019;18. doi:10.1186/s12936-019-3028-z
- New perspectives : malaria diagnosis : report of a joint WHO/USAID informal consultation, 25-27 October 1999. Geneva PP - Geneva: World Health Organization;
- National Bureau of Statistics, United Nations Children’s Fund. Multiple Indicator Cluster Survey 2021 (MICS) survey findings report.
- Gatton ML, Chaudhry A, Glenn J, Wilson S, Ah Y, Kong A, et al. Impact of Plasmodium falciparum gene deletions on malaria rapid diagnostic test performance. Malar J. 2020;19: 1–11. doi:10.1186/s12936-020-03460-w
- World Health Organization. False-negative RDT results and P. falciparum histidine-rich protein 2/3 gene deletions. Malar J. 2019;10: 1–12.
- Abuaku B, Amoah LE, Peprah NY, Asamoah A, Amoako EO, Donu D, et al. Malaria parasitaemia and mRDT diagnostic performances among symptomatic individuals in selected health care facilities across Ghana. BMC Public Health. 2021;21: 239. doi:10.1186/s12889-021-10290-1
- Milne LM, Kyi MS, Chiodini PL, Warhurst DC. Accuracy of routine laboratory diagnosis of malaria in the United Kingdom. J Clin Pathol. 1994;47: 740–742. doi:10.1136/jcp.47.8.740
- Murray CK, Gasser Jr RA, Magill AJ, Miller RS. Update on rapid diagnostic testing for malaria. Clin Microbiol Rev. 2008;21: 97–110.
- Okell LC, Bousema T, Griffin JT, Ouédraogo AL, Ghani AC, Drakeley CJ. Factors determining the occurrence of submicroscopic malaria infections and their relevance for control. Nat Commun. 2012;3: 1–9. doi:10.1038/ncomms2241
- Odhiambo F, Buff AM, Moranga C, Moseti CM, Wesongah JO, Lowther SA, et al. Factors associated with malaria microscopy diagnostic performance following a pilot quality-assurance programme in health facilities in malaria low-transmission areas of Kenya, 2014. Malar J. 2017;16: 1–10. doi:10.1186/s12936-017-2018-2
- Mshani IH, Siria DJ, Mwanga EP, Sow BBD, Sanou R, Opiyo M, et al. Key considerations, target product profiles, and research gaps in the application of infrared spectroscopy and artificial intelligence for malaria surveillance and diagnosis. Malar J. 2023;22: 346. doi:10.1186/s12936-023-04780-3
- González Jiménez M, Babayan SA, Khazaeli P, Doyle M, Walton F, Reedy E, et al. Prediction of mosquito species and population age structure using mid-infrared spectroscopy and supervised machine learning. Wellcome Open Res. 2019;4: 76. doi:10.12688/wellcomeopenres.15201.3
- Siria DJ, Sanou R, Mitton J, Mwanga EP, Niang A, Sare I, et al. Rapid age-grading and species identification of natural mosquitoes for malaria surveillance. Nat Commun. 2022;13: 1501. doi:10.1038/s41467-022-28980-8
- Goh B, Ching K, Soares Magalhães RJ, Ciocchetta S, Edstein MD, Maciel-De-freitas R, et al. The application of spectroscopy techniques for diagnosis of malaria parasites and arboviruses and surveillance of mosquito vectors: A systematic review and critical appraisal of evidence. PLoS Negl Trop Dis. 2021;15: 1–24. doi:10.1371/journal.pntd.0009218
- Lambert B, Sikulu-Lord MT, Mayagaya VS, Devine G, Dowell F, Churcher TS. Monitoring the Age of Mosquito Populations Using Near-Infrared Spectroscopy. Sci Rep. 2018;8: 5274. doi:10.1038/s41598-018-22712-z
- Khoshmanesh A, Dixon MWA, Kenny S, Tilley L, McNaughton D, Wood BR. Detection and Quantification of Early-Stage Malaria Parasites in Laboratory Infected Erythrocytes by Attenuated Total Reflectance Infrared Spectroscopy and Multivariate Analysis. Anal Chem. 2014;86: 4379–4386. doi:10.1021/ac500199x
- Adegoke JA, De Paoli A, Afara IO, Kochan K, Creek DJ, Heraud P, et al. Ultraviolet/Visible and Near-Infrared Dual Spectroscopic Method for Detection and Quantification of Low-Level Malaria Parasitemia in Whole Blood. Anal Chem. 2021;93: 13302–13310. doi:10.1021/acs.analchem.1c02948
- Mwanga EP, Minja EG, Mrimi E, Jiménez MG, Swai JK, Abbasi S, et al. Detection of malaria parasites in dried human blood spots using mid-infrared spectroscopy and logistic regression analysis. Malar J. 2019;18: 1–13. doi:10.1186/s12936-019-2982-9
- Webster GT, de Villiers KA, Egan TJ, Deed S, Tilley L, Tobin MJ, et al. Discriminating the Intraerythrocytic Lifecycle Stages of the Malaria Parasite Using Synchrotron FT-IR Microspectroscopy and an Artificial Neural Network. Anal Chem. 2009;81: 2516–2524. doi:10.1021/ac802291a
- Heraud P, Chatchawal P, Wongwattanakul M, Tippayawat P, Doerig C, Jearanaikoon P, et al. Infrared spectroscopy coupled to cloud-based data management as a tool to diagnose malaria: A pilot study in a malaria-endemic country. Malar J. 2019. doi:10.1186/s12936-019-2945-1
- White NJ. Anaemia and malaria. Malar J. 2018;17: 371. doi:10.1186/s12936-018-2509-9
- Calis JCJ, Phiri KS, van Hensbroek MB. Severe anemia in Malawian children-Reply. N Engl J Med. 2008;358: 2291.
- Crawley J. Reducing the burden of anemia in infants and young children in malaria-endemic countries of Africa: from evidence to action. Intolerable Burd Malar II What’s New, What’s Needed Suppl to Vol 71 Am J Trop Med Hyg. 2004.
- Kangoye DT, Noor A, Midega J, Mwongeli J, Mkabili D, Mogeni P, et al. Malaria hotspots defined by clinical malaria, asymptomatic carriage, PCR and vector numbers in a low transmission area on the Kenyan Coast. Malar J. 2016;15: 1–13.
- Okell LC, Ghani AC, Lyons E, Drakeley CJ. Submicroscopic infection in plasmodium falciparum-endemic populations: A systematic review and meta-analysis. J Infect Dis. 2009;200: 1509–1517. doi:10.1086/644781
- Banas AM, Banas K, Chu TTT, Naidu R, Hutchinson PE, Agrawal R, et al. Comparing infrared spectroscopic methods for the characterization of Plasmodium falciparum-infected human erythrocytes. Commun Chem. 2021;4: 129. doi:10.1038/s42004-021-00567-2
- Payne D. Use and limitations of light microscopy for diagnosing malaria at the primary health care level. Bull World Health Organ. 1988;66: 621.
- Ranadive N, Kunene S, Darteh S, Ntshalintshali N, Nhlabathi N, Dlamini N, et al. Limitations of Rapid Diagnostic Testing in Patients with Suspected Malaria: A Diagnostic Accuracy Evaluation from Swaziland, a Low-Endemicity Country Aiming for Malaria Elimination. Clin Infect Dis. 2017;64: 1221–1227. doi:10.1093/cid/cix131
- Prosser C, Gresty K, Ellis J, Meyer W, Anderson K, Lee R, et al. Plasmodium falciparum histidine-rich protein 2 and 3 gene deletions in strains from Nigeria, Sudan, and South Sudan. Emerg Infect Dis. 2021;27: 471.
- Gendrot M, Fawaz R, Dormoi J, Madamet M, Pradines B. Genetic diversity and deletion of Plasmodium falciparum histidine-rich protein 2 and 3: a threat to diagnosis of P. falciparum malaria. Clin Microbiol Infect Off Publ Eur Soc Clin Microbiol Infect Dis. 2019;25: 580–585. doi:10.1016/j.cmi.2018.09.009
- Beshir KB, Epúlveda N, Bharmal J, Robinson A, Mwanguzi J, Busula AO, et al. Plasmodium falciparum parasites with histidine-rich protein 2 (pfhrp2) and pfhrp3 gene deletions in two endemic regions of Kenya. Sci Rep. 2017;7: 1–10.
- Beshir KB, Parr JB, Cunningham J, Cheng Q, Rogier E. Screening strategies and laboratory assays to support Plasmodium falciparum histidine-rich protein deletion surveillance: where we are and what is needed. Malar J. 2022;21: 201.
- Nyataya J, Waitumbi J, Mobegi VA, Noreddin A, El Zowalaty ME. Plasmodium falciparum histidine-rich protein 2 and 3 gene deletions and their implications in malaria control. Diseases. 2020;8: 15.
- Koita OA, Doumbo OK, Ouattara A, Tall LK, Konaré A, Diakité M, et al. False-negative rapid diagnostic tests for malaria and deletion of the histidine-rich repeat region of the hrp2 gene. Am J Trop Med Hyg. 2012;86: 194.
- Murillo Solano C, Akinyi Okoth S, Abdallah JF, Pava Z, Dorado E, Incardona S, et al. Deletion of Plasmodium falciparum histidine-rich protein 2 (pfhrp2) and histidine-rich protein 3 (pfhrp3) genes in Colombian parasites. PLoS One. 2015;10: e0131576.
- Kobayashi T, Sikalima J, Parr JB, Chaponda M, Stevenson JC, Thuma PE, et al. The Search for Plasmodium falciparum histidine–rich protein 2/3 Deletions in Zambia and Implications for Plasmodium falciparum histidine-rich protein 2-Based Rapid Diagnostic Tests. Am J Trop Med Hyg. 2019;100: 842.
- Kaaya RD, Kavishe RA, Tenu FF, Matowo JJ, Mosha FW, Drakeley C, et al. Deletions of the Plasmodium falciparum histidine-rich protein 2/3 genes are common in field isolates from north-eastern Tanzania. Sci Rep. 2022;12: 5802.
- Schellenberg D, Menendez C, Kahigwa E, Font F, Galindo C, Acosta C, et al. African children with malaria in an area of intense Plasmodium falciparum transmission: features on admission to the hospital and risk factors for death. Am J Trop Med Hyg. 1999;61: 431–438.
- Kahigwa E, Schellenberg D, Sanz S, Aponte JJ, Wigayi J, Mshinda H, et al. Risk factors for presentation to hospital with severe anaemia in Tanzanian children: a case–control study. Trop Med Int Heal. 2002;7: 823–830.
- Furtado LFV, Alves WP, da Silva VJ, Rabelo ÉML. Hookworm infection as a model for deepen knowledge of iron metabolism and erythropoiesis in anemia. Cytokine. 2024;177: 156559.
- Mwanga EP, Siria DJ, Mitton J, Mshani IH, Jimenez MG, Selvaraj P, et al. Using transfer learning and dimensionality reduction techniques to improve generalisability of machine-learning predictions of mosquito ages from mid-infrared spectra. BMC Bioinformatics. 2023;24: 11. doi:10.1186/s12859-022-05128-5
- Garcia GA, Kariyawasam TN, Lord AR, da Costa CF, Chaves LB, Lima-Junior J da C, et al. Malaria absorption peaks acquired through the skin of patients with infrared light can detect patients with varying parasitemia. PNAS Nexus. 2022.
- Trager W, Jensen JB. Human malaria parasites in continuous culture. J Parasitol. 2005;91: 484–486. doi:10.1645/0022-3395(2005)091[0484:HMPICC]2.0.CO;2
- Lambros C, Vanderberg JP. Synchronization of Plasmodium falciparum erythrocytic stages in culture. J Parasitol. 1979; 418–420.
- Kamau E, Alemayehu S, Feghali KC, Saunders D, Ockenhouse CF. Multiplex qPCR for detection and absolute quantification of malaria. PLoS One. 2013;8: e71539.
- Hofmann N, Mwingira F, Shekalaghe S, Robinson LJ, Mueller I, Felger I. Ultra-Sensitive Detection of Plasmodium falciparum by Amplification of Multi-Copy Subtelomeric Targets. PLOS Med. 2015;12: e1001788.
- National guidelines for malaria diagnosis and treatment, 2006 / United Republic of Tanzania, Ministry of Health and Social Welfare. [Dar es Salaam] : National Malaria Control Programme, [2006].
- Liu S, Deng Z, Li J, Wang J, Huang N, Cui R, et al. Measurement of the refractive index of whole blood and its components for a continuous spectral region. J Biomed Opt. 2019;24: 1–5. doi:10.1117/1.JBO.24.3.035003
There is NO Competing Interest.
- SupplimentarytablesandFiguresMshanietallLLODofmalariaparasitebyMIRsML.docx
Supplimentary tables and Figures