E-ISSN: 2814 – 1822; P-ISSN: 2616 – 0668
REVIEW ARTICLE
Kabir Yahuza1*, Aliyu Umar M2, Baha’uddeen S. D.1, Atalabi E. T.2, Mukhtar Gambo Lawal1 and Bashir Abdulkadir1
1Department of Microbiology Umaru Musa Yar’adua University, PMB 2218, Katsina, Nigeria
2&3Department of Biological Sciences, Federal University Dutsinma, Katsina State, Nigeria
Plasmodium parasites are the principal causative agents of malaria, a highly infectious and sometimes fatal illness. It is a serious worldwide health risk, particularly in tropical and subtropical areas, where it has become a significant public health threat. Thus, its diagnosis must be timely, efficient, and accurate to allow suitable management and effective control of the disease. With recent technological advancements, it became possible to use current advances in image processing and machine learning to apply artificial intelligence (AI) for the detection /quantification of malaria parasites. The goal of this paper is to present a thorough analysis of the most advanced AI-assisted techniques available today, such as convolutional neural networks (CNNs), deep learning, and computer vision approaches, highlighting their strengths and limitations for identifying and quantifying malaria parasites in a variety of biological materials, including digital photos and blood smears. The review also discusses key challenges and future trends in AI-powered malaria detection, such as dataset scarcity, stability and robustness of algorithms, and scalability at a geographic level for resource-constraining areas. In conclusion, through critically examining the existing literature and research findings, this review showcases the potential of AI-driven technologies to revolutionize malaria diagnosis and surveillance with a view to guiding stakeholders in the choice of effective control strategies against this infectious disease.
Keywords: Artificial Intelligence, Convolutional Neural Networks, Detection, Quantification, Malaria, Plasmodium.
Malaria is a parasitic infectious disease that could be fatal caused by the Plasmodium genus. The disease has consequently been ranked among the biggest threats to public health in poorer nations (WHO, 2022). Of the different Plasmodium parasites that can cause human infection, P. falciparum has been described as responsible for the more severe forms of the disease and may eventually progress to cerebral malaria (CM) or multiple organ dysfunction syndrome (MODS) (Calderaro et al., 2024). There is a clear need for improved diagnostic and treatment approaches, given that there were an expected 249 million instances of malaria infections worldwide in 2022, with around 230,000 fatalities recorded (WHO, 2022).
The gold standard for diagnosing malaria is looking for malaria parasites under a microscope on stained blood films (Wardhani et al., 2020). However, conventional microscopy is difficult to apply in resource-constrained situations where the malaria burden is frequently highest since it is labor-intensive, time-consuming, and dependent on the knowledge of competent microscopists (Wardhani et al., 2020). Furthermore, the subjectivity of microscopy might result in inconsistent findings and possible diagnostic mistakes, highlighting the need for more trustworthy and objective diagnostic techniques (Mehanian et al., 2017).
Artificial intelligence (AI) has become a potential technology for the identification and quantification of malaria parasites in recent years. Artificial intelligence (AI) methods, in particular convolutional neural networks (CNNs), have demonstrated impressive powers in examining digital photos and recognising patterns suggestive of malaria parasites (Yang et al., 2019). AI models have the ability to overcome the drawbacks of conventional microscopy and improve the precision and efficacy of malaria diagnosis by automating the process of parasite identification and quantification via the use of machine learning algorithms and deep learning techniques (McDermott, 2020).
AI-powered apps such as mobile medical applications (MMAs) for diagnosing malaria go through several phases of development and validation, including building hardware and software components, refining image processing and collection techniques, and verifying parasite detection algorithms (Visser et al., 2021). Validation studies measure factors including sensitivity, specificity, and diagnostic accuracy by comparing AI models to reference techniques or traditional microscopy and assessing their performance against predetermined standards (Visser et al., 2021).
Although AI-powered applications have the potential to improve malaria detection, there are still a number of obstacles to their acceptance and deployment. To guarantee the general adoption and efficacy of AI models in malaria diagnosis, concerns including infrastructural constraints, availability to skilled individuals, and cost-effectiveness must be addressed (Alonso & Tanner, 2013).
This review attempts to give an overview of current developments in artificial intelligence for malaria parasite identification and quantification in light of these prospects and difficulties. Through an examination of recent papers categorised by topics like Deep Learning Methods, Point-of-Care Technologies, Molecular Detection Techniques, Transfer Learning Approaches, and Segmentation Methods, the review aims to synthesise current knowledge, identify gaps, and assess the implications of these cutting-edge technologies. This review seeks to improve malaria diagnostic techniques and support evidence-based decision-making for malaria control and elimination initiatives by critically analysing the literature.
Our review's scope is limited to scholarly publications that have been published (available in Google scholar/Google/Scopus/PubMed search engines) between 2017 to 2024, reporting various AI techniques utilised to diagnose malaria. The evaluation took into account the geographic and epidemiological diversity of malaria while examining a variety of methodological techniques, such as deep learning algorithms and point-of-care devices. However, we disregarded research that did not directly address the use of AI in the diagnosis of malaria or that did not fit within the parameters of the chosen subjects. The review’s scope is depicted in Figure 1.
Figure 1: Conceptual framework of the review
The reviewed papers utilized deep learning techniques to improve the detection and classification of malaria parasites in thin blood smears. Deep learning, especially Convolutional Neural Networks (CNNs), has proven effective in automating this traditionally manual process. Research aims to enhance accuracy, reduce reliance on human expertise, and speed up diagnostics. These techniques are discussed below:
A consensus among most of the papers is that CNNs are the primary deep-learning method for malaria detection and classification. Models like VGG19 (a CNN model with 19 layers), ResNet, and GoogleNet have been adapted for this application. The summary of the recent findings on CNN-Based Models for Malaria Detection and Classification is presented in Table 1.
Table 1: Common CNN-Based Models for Malaria Detection and Classification
| 2022 | CNN, ResNet50, VGG19, AlexNet | VGG19 was most efficient with 95.28% training accuracy and 93.89% testing accuracy | Kadiyala et al. (2022) |
|---|---|---|---|
| 2022 | Dilated CNN | Achieved 99.9% accuracy for parasite detection | Garba et al. (2022) |
| 2022 | CNN | Achieved 96.97% accuracy, 97.00% precision, and 97.00% sensitivity | Ozsahin et al. (2022) |
| 2022 | CNN with STM-SB-RENet | Achieved 97.98% accuracy | Khan et al. (2022) |
| 2022 | Transformer-based model with multi-headed attention | High accuracy and explainability through Grad-CAM | Islam et al. (2022) |
| 2023 | CNN-based models (VGG19, ResNet, GoogleNet) | Achieved 94.63% accuracy in detecting infected cells | Mohammed et al. (2023) |
| 2023 | Two-stage deep learning system | Notable results on public dataset for detection and life stage classification | Chen et al. (2022) |
| 2023 | Six deep learning models with data augmentation | Maximum accuracy of 96.73% | Babu et al. (2023) |
| 2023 | GoogleNet, VGG19 | GoogleNet achieved the highest accuracy in classifying malaria parasites at the schizont stage. | Jusman et al. (2023) |
| 2023 | Hybrid deep learning approach with CNNs, SVMs, etc. | Achieved high accuracy for malaria parasite classification | Jones et al. (2023) |
| 2023 | Deep learning-based web app using CNN models | Achieved high accuracy in detecting malaria parasites in granular blood samples | Santoshi et al. (2023) |
| 2023 | Automated parasite identification using CNNs | Addressed urgent need for accurate malaria diagnosis | Pandey et al. (2023) |
| 2023 | Optimal machine learning-based automated detection model | Achieved 98.5% accuracy | Kundu and Anguraj (2023) |
| 2023 | CNN-based model for diagnosis and classification | Achieved 94.7% accuracy | Shriya et al. (2023) |
| 2023 | CNN-based algorithm for vector and parasite identification | DenseNet-121 achieved 99.5% accuracy | Hasikin (2023) |
| 2023 | Edge-assisted framework using federated learning | Introduced a new framework for malaria parasite detection | Rajput et al. (2023) |
| 2023 | Mosquito Net model | Achieved 96.97% accuracy | Oladele et al. (2023) |
| 2023 | A CNN-based model with ten-fold cross-validation | High accuracy was achieved with 27,558 single-cell images | Nugroho and Nurfauzu (2023) |
| 2023 | Hybrid deep learning approach for legally blind accessibility | Achieved promising results in malaria parasite detection | Pandiarajan et al. (2023) |
| 2024 | Quantum-convolutional networks | Demonstrated high accuracy for malaria classification | Kumar et al. (2024) |
| 2024 | Web-based deep learning system using CNN models | Achieved high accuracy for malaria parasite detection | Babu et al. (2024) |
| 2024 | Efficient CNN architecture | Achieved 99.8%, 98.2%, and 97.7% accuracy for training, validation, and testing sets | Alraba'nah and Toghuj (2024) |
As can be seen from Table 1, one of the significant models was developed by Mohammed et al. (2023), which achieves 94.63% accuracy in detecting infected cells, emphasizing regions of interest for stained parasites. Similarly, Chen et al. (2022) present a two-stage deep learning system with detection and life-stage classification models, showing notable results on a public dataset. Babu et al. (2023) compare six deep learning models with data augmentation techniques, reporting a maximum accuracy of 96.73%. Jusman et al. (2023) use GoogleNet and VGG-19 pre-trained models for classifying malaria parasites at the schizont stage, with GoogleNet achieving the highest accuracy.
Jones et al. (2023) present a hybrid deep learning approach using bilateral filtering, CNNs, improved grey-wolf optimization, and support vector machines, achieving high accuracy for malaria parasite classification. Santoshi et al. (2023) propose a deep learning-based web app for detecting malaria parasites in granular blood samples using CNN models like ResNet50, VGG19, and custom CNNs, achieving high accuracy.
Pandey et al. (2023) developed an automated parasite identification method in blood smear pictures using CNNs, addressing the urgent need for accurate malaria diagnosis. Kundu and Anguraj (2023) propose an optimal machine learning-based automated malaria parasite detection and classification model using blood smear images, achieving 98.5% accuracy.
Kadiyala et al. (2022) evaluate the performance of pre-trained CNN-based models like AlexNet, ResNet50, and VGG19 as feature extractors for analyzing infected and non-infected cells, with VGG19 being the most efficient (training accuracy of 95.28% and testing accuracy of 93.89%). Ozsahin et al. (2022) propose a CNN capable of accurately predicting malaria parasites using microscopic images of thin and thick peripheral blood smears, with an accuracy, precision, and sensitivity of 96.97%, 97.00%, and 97.00% for thick smears.
Akruwala and Prajapati (2022) review work on reducing diagnosis time using deep learning and image processing techniques. Garba et al. (2022) present a Dilated CNN for malaria parasite detection and species classification using blood smear images, achieving 99.9% for parasite detection and varying accuracy for different species.
Raj et al. (2023) propose a CNN-based method for diagnosing and classifying malaria using microscopic blood-smeared images. Khan et al. (2022) introduce a novel CNN with STM-SB-RENet, achieving 97.98% accuracy. Shriya et al. (2023) employ CNNs and VGG models for malaria diagnosis with 94.7% accuracy. Amin et al. (2022) combine color-based k-means clustering for segmentation with deep feature extraction using pre-trained models, achieving 99.2% accuracy with the SVM classifier.
Islam et al. (2022) propose a transformer-based model with a multi-headed attention mechanism, achieving high accuracy and explainability through Grad-CAM. Rajput et al. (2023) introduce an edge-assisted framework for malaria parasite detection using federated learning.
Hasikin (2023) presents two case studies using deep learning for malaria parasite and vector identification. This research develops a neural network-based algorithm for identifying mosquito vectors and detecting Plasmodium parasites in microscopic blood smear images, with DenseNet-121 achieving 99.5% accuracy.
Pandiarajan et al. (2023) introduce a hybrid deep learning approach for malaria parasite detection to enable accessibility for the legally blind, achieving promising results. Kumar et al. (2024) demonstrate that quantum-convolutional networks can achieve high accuracy for malaria classification. Babu et al. (2024) propose a web-based deep learning system for malaria parasite detection in granular blood samples using CNN models like ResNet50, VGG19, and custom CNNs, achieving high accuracy.
Alraba'nah and Toghuj (2024) propose an efficient CNN architecture for diagnosing malaria, achieving high accuracy rates of 99.8%, 98.2%, and 97.7% for training, validation, and testing sets. Oladele et al. (2023) explore a Mosquito Net model trained on an augmented malaria cell image dataset, achieving 96.97% accuracy. Nugroho and Nurfauzu (2023) present a novel CNN-based model for automating the classification and prediction of infected cells in thin blood smears, achieving high accuracy through rigorous ten-fold cross-validation with 27,558 single-cell images.
Based on the reviewed articles above, the accuracy of different deep learning methods for malaria detection and classification varies significantly. CNNs remain the dominant deep-learning method for malaria detection and classification, with various models showing high effectiveness. Among the models, some like the Dilated CNN, DenseNet-121, and certain efficient CNN architectures achieve exceptionally high accuracy rates such as Garba et al. (2022) with 99.9% for parasite detection using a Dilated CNN, Hasikin (2023) with 99.5% using DenseNet-121 for malaria parasite and vector identification, Alraba'nah and Toghuj (2024) with 99.8% accuracy (training set) with an efficient CNN architecture and Kundu and Anguraj (2023) with 98.5% accuracy using an optimal machine learning-based model. The methods with the lowest accuracy, such as that of Shriya et al. (2023) with 94.7% using CNNs and VGG models for malaria diagnosis and Mohammed et al. (2023) with 94.63% accuracy in detecting infected cells, still perform well but are slightly less effective compared to the top performers.
Another cluster of works focuses on developing mobile-based and optimized deep learning frameworks fit for resource-constrained settings. Shahadat (2023) presents a mobile-embedded CNN architecture with a squeeze-and-excitation block and spatial 1D CNN layer, achieving a testing accuracy of 99.52% on a modified dataset. Similarly, Oladele et al. (2023) describe a mobile-based CNN framework for rapid detection using Android phones, with an accuracy of 95.81% on the testing set. Nugroho and Nurfauzu (2023) propose a hybrid method using optimized thresholding and deep learning for improved detection and segmentation on the PlasmoID dataset, achieving an F1-score of 0.91 in parasite detection.
Some papers present comparative analyses of different models and novel architectures to enhance detection performance. For example, Gill et al. (2023) discuss a sustainable classification model through transfer learning using VGG19, achieving an accuracy rate of over 90%. Saini et al. (2023) evaluate CNNs as feature extractors for classifying cells, obtaining accuracy values of up to 94.42%. Mahmood et al. (2023) explore a three-layered CNN combined with a two-layered dense neural network, achieving a remarkable accuracy of 96%. Lastly, Sherif and Mohammed (2023) propose a new CNN model for identifying infected cells, comparing its performance with other pre-trained models and reporting an accuracy of 97%.
The papers reviewed show a strong consensus on the effectiveness of CNNs in malaria detection and classification. There is variation in the specific architectures and pre-trained models used, but the underlying approach of using deep learning for feature extraction and classification is consistent. Studies also highlight the importance of model optimization and adaptation for mobile and resource-constrained environments. Disagreements are minimal and typically revolve around the best model architecture or pre-training strategy to achieve optimal results. Overall, advancements in deep learning methods show promise for improving malaria diagnosis accuracy and accessibility. Pandey et al. (2023) address the urgent need for accurate and easy malaria diagnosis using CNNs, demonstrating excellent sensitivity (95%) and specificity (92%) in parasite identification.
Although much progress has been made using CNNs and other deep learning methods in malaria detection, some gaps and future research areas remain.
Most studies have relied on publicly available datasets that may not fully represent the diversity of malaria infections across different geographic regions. The generalization capabilities of these models are a concern when exposed to data from various sources, different microscopy techniques, or unstained blood smear images. There is a need for more diverse and comprehensive datasets that showcase various forms of the parasite and different stages of infection. Future research should focus on collecting and curating such datasets to train more robust models.
While the proposed models have shown high accuracy in controlled experimental settings, there is often a lack of information about their performance in real-world clinical environments. These models need to be deployed in real healthcare settings, with subsequent validation studies to confirm their effectiveness and reliability. Models also need to be tested for their ability to detect multiple malaria species and differentiate between similar-looking parasites or artifacts in blood smears.
Although some studies, such as Shahadat (2023) and Oladele et al. (2023), have begun to address the need for mobile-based solutions, there is still a gap in developing computationally efficient models that can run on low-powered devices without compromising accuracy. Future research could focus on optimizing existing models or developing new architectures that balance performance and computational requirements for mobile deployment.
Interpretability is crucial in medical applications to build trust among healthcare professionals. Most of the reviewed models are black boxes, making it difficult to understand the decision-making process. Research into explainable AI models, such as the work by Pandiarajan et al. (2023), is essential to provide insights into the features and patterns the models use for classification, thereby increasing their acceptance in clinical practice.
There is a gap in research with regard to the integration of AI-based malaria detection systems into existing healthcare infrastructure and workflows. Seemingly, the integration needs to be seamlessly developed so that such systems complement and do not interfere with the diagnostic process. This takes on multiple aspects, including a user interface for healthcare workers, compatibility with lab equipment, and adherence to medical data standards and regulations.
The performance of the deep learning models may vary due to changes in the quality and consistency of the input images. The lighting condition, magnification, and staining procedures are some factors that influence how well the model perceives a parasite in an image. To this end, beneficial areas of research will include models robust to these variations or methods for standardizing image acquisition.
Cost-effectiveness analysis of the deployment of deep learning-based malaria detection systems, especially in low-resource settings where the burden of diseases is high, will also be an important aspect. This should include upfront costs of developing and deploying such a system as well as maintenance and support costs.
Most studies focus on malaria detection alone, while in practice, patients may have multiple infections with several pathogens at the same time. Such an integrated approach to data from multiple diagnostic tests might enhance the performance and detail in disease diagnosis. Similarly, developing models for detecting multiple pathogens from a blood smear would bring with it a great advantage for areas with high co-infection rates.
The malaria parasites evolve and become resistant, so deep learning essentially requires frequent updating. Consequently, this approach allows the updating of model parameters under the continuous learning regime without losing information that has been previously learned.
In summary, while the reviewed studies presented the potential of deep learning in malaria detection, addressing such research gaps will significantly enhance the practicality and effectiveness of this kind of system, leading to better disease management and patient outcomes.
A consensus from papers by Dev et al. (2023), Nascimento et al. (2023), and Wojtas et al. (2023) indicates that advanced computational methods, especially those involving neural networks, are most effective for the detection and identification of malaria from microscopic images of blood samples. These papers outlined the potential of machine learning and deep learning methods in improving accuracy and speed in the diagnoses of malaria, which is very critical for point-of-care technology.
In another study, researchers developed hybrid deep learning models, combining a convolutional neural network and various forms of recurrent neural networks, including LSTM, BiLSTM, and GRU, in the identification of malaria parasites. They reported a high accuracy and low error rate for the CNN-GRU-GRU and a fast-training time for the CNN-LSTM-LSTM, thus indicating models such as those to be useful for real-time, portable point-of-care devices.
The paper by Nascimento et al. (2023) introduced a completely different approach based on shallow neural networks combined with digital image processing techniques. With this technique, less complex and faster models will be generated and then deployed in mobile phones ideal for use in poor-resource locations. This would save both financial and human resources. As the authors have written, an F1-score outperforms any previous works; this proves that less complex models are able to give quality results because such critical applications are, after all, based on evergreen technologies, without any mystery around them, like detecting malaria.
A custom semantic segmentation neural network structure has been proposed for the rapid classification of microscopic images in these papers (Wojtas et al., 2023). For the clinician, this model provides a classification mask showing the position of healthy, infected cells and background in order to make a semi-automatic diagnosis of malaria. The results support the further use of neural networks in point-of-care devices, since the model has given very good recognition accuracy with very minimal computational power.
To conclude, the studies reviewed present various deep learning approaches for malaria detection (Hybrid Deep Learning Models combine convolutional neural networks (CNNs) with recurrent neural networks (RNNs) like LSTM, BiLSTM, and GRU; Shallow Neural Networks with Digital Image Processing; and Custom Semantic Segmentation Neural Network) each with distinct methodologies and strengths. However, the hybrid deep learning models (CNN combined with RNNs) seem to be the most effective due to their high accuracy and efficiency, especially in real-time and portable applications. These models are particularly promising for use in point-of-care devices where both speed and precision are critical.
The articles by Krishnadas et al..,2022 and Rees-channer et al.., 2023, address automated devices whose features have, by and large, facilitated the ease of detection of malaria for point-of-care applications where simplicity and fast diagnosis are important.
A case in point is a study by Krishnadas et al. (2022), which talks about handheld nucleic acid testing for point-of-need detection of malaria using a purification-free protocol and microfluidic technology for sample preparation and analysis of whole blood samples. The high sensitivity of the device allowed the detection of asymptomatic carriers of malaria, a great stride in disease eradication efforts.
In contrast, Rees-channer et al. (2023) evaluated the performance of an automated microscope device, EasyScan GO, when combined with machine learning software for the detection of malaria parasites. This system showed proficiency in detecting the parasites and distinguishing Plasmodium falciparum from the non-falciparum species compared to expert light microscopy, although it limited quantification of the low parasite densities and differentiation between the non-falciparum species. Both studies indicate that despite needing improvement to become more sensitive in all areas of detection and identification, the potentiality is there for automated devices to help detect malaria.
On the whole, the major findings from these two papers can be summarized as showing that there is a promising potential for improved malaria detection through integrated machine learning and automated devices at the point of care. There seems to be a consensus on the potential efficacy of the image analysis capabilities of neural network-based models and those of automated devices in making a more straightforward diagnostic process. However, further work needs to be done on variables such as precise measurement of parasite densities and the differentiation of species.
Although the discussed papers show lots of progress in this specific direction of malaria detection through machine learning and automated devices, there still exist some gaps and discrepancies that demand further research to optimize these technologies for widespread field usage. On this point, computational resources required for different approaches appear as some of the main discrepancies. The work of Dev et al. (2023) uses hybrid deep learning models that are naturally accurate but at the cost of demanding great computational resources, which are not readily available in most remote or resource-constrained settings. In contrast, Nascimento et al. (2023) will focus on using shallow neural networks, which are computationally less intensive and more suited to mobile devices. Future research will be dedicated to seeking an optimized compromise between model complexity and computational efficiency that allows for the most effective models to also be the most accessible.
As pointed out by Rees-Channer et al. (2023), the detection of low parasite densities is one of the weaknesses in achieving this goal. This has a relation to the capacity of the technique to pick up parasites at low densities for early diagnosis and the interruption of malaria transmission. The performance of EasyScan GO for low parasite densities has not been able to meet WHO competency standards under level 1 and thus suggests remarkable potential for improvements in this regard. Sensitivity should be increased without compromising specificity to allow improved detection, especially in asymptomatic carriers.
Research should be conducted on the differentiation of non-falciparum species. As observed during the assessment of the EasyScan GO Rees-channer et al. (2023), the device fails to identify the non-Plasmodium falciparum species and instead shows all as P. vivax. The use of such a device is, therefore, limited in settings with multiple circulating species. Future studies could aim at more sophisticated algorithms or use more data sources to enhance the capability of species differentiation.
The integration of these technologies into existing healthcare systems is not well addressed in the literature. Research is required on how these diagnostic tools can be translated for use in different healthcare setups, keeping in mind the infrastructural and workflow differences that exist between regions. This would involve user training, data management, and interoperability of other healthcare technologies.
The generalizability and robustness of the proposed models are not known. Research should focus on testing the generalizability and robustness of these models in datasets involving different staining techniques, different image qualities, and patient demographics. This would ensure that the models are not only effective in controlled study environments but also in the field where conditions are more variable.
Real-time analysis and feedback mechanisms are research areas that could be expanded upon. While Dev et al.., 2023, and Wojtas et al.., 2023 both demonstrate the possibilities of using real-time applications, neither explores how these could provide instant feedback or alerts to the healthcare worker or patient. Equally important are interfaces and feedback systems that would be user-friendly and thus supportive of clinical decision-making.
Finally, cost-effectiveness and the scale-up of these technologies are not well addressed. The technologies must also be affordable and scalable for them to be adopted widely, especially in low-resource settings. Cost-benefit and possible economies of scale in mass production and distribution of the diagnostic tools are yet to be researched.
In conclusion, although the above-reviewed studies have shown the potential of machine learning and automated devices in combating malaria, there exist major research gaps that need to be investigated. Filling these gaps will be the key to making diagnostic tools that are technologically capable as well as practical and usable in various settings where malaria is prevalent.
This paper brings out the main highlights identified from the recent research papers that covered the application of object detection models for malaria parasite detection in blood smear images. The papers are then categorized based on the detection techniques that the approaches have used, such as classic machine learning algorithms, convolutional neural network (CNN) models, and You Only Look Once (YOLO)-based models.
A typical CNN model is shown in Figure 2. It has an input section where the raw image is added and many convolutional layers that process the image to give and output in the classifier layer. A strong statement from various studies supports the claim that CNN -based methods are the strongest in detecting malaria parasites from blood smear images. The ensemble technique with CNN architectures gave high accuracy, precision, specificity, F1-score, and Cohen's Kappa coefficient in Ahmad et al., 2023. Similarly, Zhong et al., 2023, used the CNN method through a portable system integrated with a smartphone and microscope, giving high accuracy and F1-score. The study conducted by Efrizoni et al., 2023, also used CNN with ResNet architecture to give high accuracy. According to (Kundu and Anguraj, 2023), applications have reached the deployment of traditional machine learning algorithms even for the detection of the malaria parasite, although in a less sensitive and specific manner than deep learning models.
Figure 2: Typical CNN Model Architecture (Adoped from Zaniolo et al., 2020)
Another group that presents much promise regarding their effectiveness in malaria detection tasks is YOLO-based models (Figure 3). Zhang and Chen (2023) presented a more reliable and faster YOLO model called YOLO-VF. In their work, Koirala et al. (2022) developed a custom YOLO-mp model that significantly outperforms the standard YOLOv4 in terms of accuracy and computational efficiency. Similarly, other authors made use of YOLOv5 to compare to other architectures or make further improvements in relation to better accuracy through more modules used within YOLOv5 (Liu et al., 2023). In the work of Zedda et al. (2023), a novel YOLO-based architecture has been proposed to detect malaria parasites, integrated with the Transformer and attention mechanisms. An example would be the study by Krishnadas et al. (2022) which utilized YOLO for the proper classification of a malaria stage and type; however, it achieved mixed results in this regard.
Figure 3: A Typical YOLO Architecture (Adopted from Herfandi et al., 2024)
Some papers also tackled ensembling to provide more reliability and robustness in the process. In this study by Ozbilge et al. (2024), they first evaluated state-of-the-art models and then ensembled them; the leading model is YOLOv8s. In this way, ensemble strategies improved the performance of the detection when compared to that of single models. This research activity resulted in a big comparative analysis presented by (Rocha et al., 2022) for several object detectors with their different versions of the YOLO model, out of which the best-performing model was YOLOv5. The observed tendency in most reviewed papers shows rather the use of CNN and YOLO-based models for the recognition of malaria parasites on images of blood smears. There is fairly good consensus over model effectiveness, but some discrepancies remain in regard to the best architecture or approach, usually motivated by the particular requirements of the task in terms of accuracy, speed, or computational resources. Ensemble approaches and new mechanisms in neural network architectures, such as attention and transformers, open paths towards much stronger and more accurate models for medical diagnostics.
Even with such significant advances in applying deep learning models for malaria parasite detection, several research gaps and mismatches have been noted in the literature. Identification and addressing of these gaps are important for further improvements in diagnostic capabilities and practical deployment of these technologies. Several areas in which future research might have large impacts follow:
Most studies rely on data that is highly constrained in diversity, often sourced from a particular geographical location or demographic, leading to models that perform well on the data they were trained on but generalize badly to new, unseen data. Future work can be done by building and using a diverse blood smear dataset, with variations covering the scope of the majority of the blood smear image dataset from different regions, different ethnic groups, and species of the malaria parasite.
Some studies note high performance within controlled environments but do not specify applicability in real-world settings with diverse image quality and acquisition conditions. More research is needed on the models that can generalize between different microscopy setups, staining techniques, and image resolutions. Additionally, in this regard, the importance of transfer learning approaches in helping fine-tune pre-trained models with new datasets without too much additional training cannot be overstated. Despite having high accuracy, DL models can become an obstacle in interpretability and explainability due to black-boxedness. Physicians and health care workers will not likely accept a prediction unless it can be explained, so future research efforts to produce more interpretable/explainable AI models for malaria detection can hopefully increase user trust and clinical acceptance.
Most of the studies concerned themselves with providing high accuracy and sensitivity, but for practical implementation, real-time analysis is the key. Work needs to be done on models that are efficient enough to run on edge devices, for instance, smartphones and portable microscopes again, with no or minimal loss of accuracy, to give a diagnosis of malaria quickly in a remote and resource-constrained setting. Robustness to noise and artifacts may significantly affect the performance of a model in blood smear images. It would be beneficial, then, to test models that are robust to this or can automatically clean and pre-process images prior to analysis.
Rigorous validation studies in clinical practice are required to compare deep learning models with performance by human experts directly. Such studies will help build the clinical efficacy of these models and help identify areas where improvement is required. Integration with existing health care systems and EHR: One will build utility and access for health providers through integration research of malaria detection models with existing health care systems and EHR. While many models show potential, their cost-effectiveness and scalability have not been thoroughly evaluated. It would, therefore, be the key for any future research to analyze the cost-benefit ratio of deploying these models at scale and coming up with strategies for minimizing costs while maximizing reach and impact.
Research into integrating malaria detection models with existing healthcare systems and electronic health records (EHR) would enhance their utility and accessibility for healthcare providers.
While many models show promise, their cost-effectiveness and scalability have not been thoroughly evaluated. Future research should analyze the cost-benefit ratio of deploying these models at scale and develop strategies to minimize costs while maximizing reach and impact.
Malaria has different stages and species that require different treatments. Most current models focus on detecting the presence of parasites, but a detailed classification of stages and species is less explored. Future models should aim to provide more comprehensive diagnostic information.
Finally, ethical considerations and potential biases in AI models must be addressed. Ensuring that models do not perpetuate or exacerbate healthcare inequalities is essential. Research into fair and ethical AI practices in the medical domain is of paramount importance.
By addressing these research gaps, the field can move towards more robust, generalizable, and clinically applicable models for malaria parasite detection, ultimately contributing to better healthcare outcomes in malaria-endemic regions.
The integration of malaria detection into automated hematology analyzers and the use of machine learning for feature extraction and classification represents a novel approach to malaria diagnosis.
Shambhu et al. (2023), Rehan et al. (2022), and Telang and Sonawane (2023) explore the use of automated hematology analyzers and machine learning for the identification of malaria parasites and COVID-19. Shambhu et al. (2023) present case series where Plasmodium species were successfully identified using an automated analyzer, while Rehan et al. (2022) investigate the utility of abnormal white blood cell differential scattergrams as an early indicator for malaria detection. Telang and Sonawane (2023) introduce a bins approach with statistical moments for detecting and classifying malaria and COVID-19 using machine learning.
Ohdar and Nigam (2023) propose a computer-based model utilizing convolutional neural networks (CNNs) and support vector machines (SVM) for feature extraction and classification of malaria species from blood smear images, achieving high accuracy and demonstrating the promise of deep learning in malaria diagnosis. Phumkuea et al. (2023) developed a simplified decision tree model with selective optimal features to reduce processing power and further shorten diagnostic time in resource-constrained areas.
Both approaches indicate a consensus on the potential of automated systems and machine learning to complement or enhance traditional malaria diagnostic methods. They agree that these technologies could be particularly beneficial in resource-limited settings where access to skilled microscopy is limited.
Hemozoin, a by-product of hemoglobin digestion by malaria parasites, serves as a unique biomarker for the detection of malaria. Studies in this group review and explore microdevices and biosensors designed to detect hemozoin for malaria diagnosis.
Baptista et al. (2022) provide a comprehensive review of biosensors and lab-on-a-chip devices targeting hemozoin as a biomarker for malaria detection. The paper discusses the potential of these devices to meet the demands for malaria management and emphasizes the advantages of hemozoin as an indicator of infection that is present across all parasite species and stages of disease progression.
This review suggests that there is a consensus on the value of hemozoin as a biomarker for malaria detection. The focus on hemozoin-based detection methods is due to their specificity to malaria infection, which could lead to the development of diagnostic tools that are both sensitive and easy to use at the point of care.
Despite advancements in malaria diagnosis, several research gaps and discrepancies between studies remain, which could guide future research efforts. The following points summarize key areas where further investigation is needed:
Certain populations, such as pregnant women and individuals with Duffy-negative blood groups, present unique diagnostic challenges. Further research is required to understand how malaria affects these populations and to develop diagnostics that are sensitive and specific to these groups.
Hemozoin is an established biomarker for malaria detection, but other potential biomarkers have been less explored. Research into alternative biomarkers that could be easily and rapidly detected at the point of care is necessary.
While studies suggest that automated systems and machine learning offer promising approaches to malaria diagnosis, more research is needed to validate these technologies in diverse settings and populations. There is also a need for the development of machine learning models that can be easily updated and maintained in resource-limited settings.
Discrepancies between studies on the performance of various diagnostic methods highlight the need for standardization in methodology. Comparative studies using standardized protocols and large, diverse populations can help determine the most effective diagnostic tools for different settings.
The adoption of advanced diagnostic methods in endemic regions is often limited by cost. Research into the cost-effectiveness of different diagnostic strategies and their impact on health economics is required to guide policy and funding decisions.
In regions where malaria is endemic, co-infections with other pathogens are common. Diagnostics that can differentiate malaria from other febrile illnesses or detect co-infections are necessary to ensure appropriate treatment.
There is a need for long-term monitoring and surveillance to assess the effectiveness of diagnostic methods over time, especially considering the dynamic nature of malaria transmission and the emergence of drug resistance.
This group of papers focuses on utilizing various pre-trained Convolutional Neural Network (CNN) models for the detection and classification of malaria parasites in blood smear images. The consensus among these works is that transfer learning significantly improves the accuracy and efficiency of malaria detection compared to traditional image processing and machine learning techniques.
The paper by Swastika et al. (2023) deploys EfficientNetB0 within a web-based application, achieving high accuracy and F1-score in detecting malaria parasites. Similarly, the paper by Swastika et al. (2022) compares three CNN architectures (ResNet50V2, EfficientNetB0, and InceptionV3) for malaria detection, with InceptionV3 showing slightly higher accuracy. Notably, all three architectures demonstrated the effectiveness of transfer learning in this domain.
In the research by Gupta et al. (2023), a sequential CNN model is introduced, yielding high accuracy and precision, highlighting the potential of sequential processing within CNNs for malaria detection. The approach by Kavitha et al. (2023) also advocates for CNN models, underlining their superiority in accuracy over other algorithms and suggesting that further feature extraction enhancements could improve results.
The following papers discuss the integration of optimization techniques and advanced models in transfer learning to refine the detection and classification of malaria parasites.
Dutta et al. (2022) present the BMODTL-BMPC model, which incorporates the Barnacles Mating Optimizer with the NasNetLarge model for feature extraction and an extreme learning machine for classification, showing superior performance. Alassaf et al. (2022) developed the IDTL-MPDC model, which utilizes the Res2Net model with hyperparameters optimized by the differential evolution algorithm, resulting in high accuracy and sensitivity.
Khan et al. (2022) explored four different pre-trained deep learning models with pre-processing and optimization, with the Inception-Resnet model exhibiting the best performance. This study emphasizes the importance of model selection and optimization in enhancing the accuracy of malaria detection.
The research by Mishra et al. (2022) stands out by offering a comparative analysis of different machine learning and deep learning techniques for malaria parasite detection. The paper illustrates the superiority of transfer learning with models like VGG19 and modified ResNet50 over conventional machine learning algorithms such as KNN, Decision Tree, Logistic Regression, and Random Forest, proposing that deep learning models offer better predictions.
In conclusion, the surveyed papers collectively affirm the efficacy of transfer learning in the detection and classification of malaria parasites using blood smear images. There is a clear consensus on the advantage of using pre-trained CNN models to enhance accuracy and reduce the need for expert intervention. While there are a variety of approaches, including the use of different CNN architectures, optimization algorithms, and machine learning models, the overarching agreement lies in the capacity of transfer learning to improve performance metrics in malaria detection tasks.
Despite the advancements in malaria detection using transfer learning and CNN architectures, several research gaps and discrepancies need to be addressed. These discrepancies highlight the diversity in methodological approaches and the potential for future research to optimize malaria detection systems further. The surveyed papers indicated gaps such as dataset standardization, computational efficiency, interpretability, clinical integration, robustness, error analysis, cross-validation, and resilience to image quality variations as areas that could significantly enhance the practical utility of malaria detection systems.
One research gap evident from the surveyed papers is the variability in datasets used for training and testing the models. While some studies may have used the same benchmark dataset, others might have employed different datasets or image pre-processing techniques, which can result in discrepancies in performance metrics. Future research could focus on creating a standardized dataset that encompasses a wide range of blood smear image variations to test the generalizability and robustness of the proposed models across various settings.
The optimization of hyperparameters and the selection of appropriate pre-trained models are crucial for achieving high efficiency and accuracy. However, the surveyed papers use different optimization algorithms and pre-trained models, leading to varied results. Furthermore, computational efficiency is a concern that is not uniformly addressed.
Another gap is the lack of interpretability and explainability in deep learning models. While CNNs have shown high accuracy in detecting malaria parasites, understanding the decision-making process of these models is limited. Research could focus on developing more interpretable models that provide insights into the features and patterns recognized by the network, which is particularly important for medical applications where diagnostic decisions must be justified.
The integration of automated malaria detection systems into clinical workflows is not extensively covered in the literature. Future studies should investigate the practical challenges of deploying such systems in healthcare settings, including acceptance by medical professionals, integration with existing laboratory information systems, and adherence to regulatory standards.
While some papers report high accuracy and F1-scores, there is often a lack of comprehensive error analysis. Understanding the types of errors made by models, such as false positives or false negatives, is crucial for clinical applications. Research should aim to not only improve overall accuracy but also reduce specific types of classification errors that could have significant implications for patient outcomes.
Discrepancies between studies in terms of the models employed and their respective performance metrics suggest a need for cross-model validation. Research could explore ensemble methods that combine the strengths of multiple models to improve detection rates. Additionally, validating the models across different architectures and datasets would ensure the robustness of the findings.
The quality of thin-blood smear images and the methods used for acquisition can greatly affect model performance. There is a gap in research addressing the impact of image quality, staining techniques, and microscopic settings on the efficacy of malaria detection models. Future work could focus on developing models that are resilient to variations in image quality and acquisition methods.
The studies by Nugroho and Nurfauzi (2023) and Shambhu et al. (2022) both explore advanced computational methodologies for detecting and segmenting malaria parasites in blood smear images. Nugroho and Nurfauzi (2023) address the challenge of semantic segmentation in malaria parasite detection, especially the issue of large parasites overshadowing smaller ones. They propose a novel dataset, PlasmoID, and a combined approach using Faster RCNN for detection and semantic segmentation techniques for improved performance. Their results show that their scheme outperforms existing semantic segmentation methods like UNet, ResFCN-18, DeepLabV3, DeepLabV3plus, and ResUNet-18 in both detection and segmentation tasks.
On the other hand, Shambhu et al. (2022) provide a comprehensive review of computational methods for automated analysis of malaria parasites using blood smear images. Their work covers the entire process from image acquisition to classification, with a focus on segmentation techniques as an integral component. They highlight the importance of accurate segmentation in the overall process of malaria parasite detection and the potential of computational methods to offer high accuracy and reduce observer variations.
Both studies agree on the critical role of segmentation in the accurate detection of malaria parasites and the potential of deep learning methods to enhance performance. They also highlight the importance of large and varied datasets for training robust models.
In the work by Shambhu et al. (2023), an edge-based segmentation method is proposed to detect malaria parasites in blood smear images. They employ gamma equalization for lighting adjustment, followed by the Fuzzy C-Means (FCM) soft clustering method and the Modified Prewitt (MPP) algorithm to enhance edge-based segmentation. Their method achieves high accuracy and is especially effective at segmenting red blood cells, which is a critical step for the subsequent detection and classification of malaria parasites.
This paper differs from the previous two in that it focuses on edge-based segmentation rather than semantic segmentation. However, it shares the common goal of improving malaria parasite detection through better segmentation techniques. The high accuracy reported by the edge-based method suggests that it could be a promising alternative or complementary approach to deep learning-based semantic segmentation methods.
While the studies by Nugroho and Nurfauzi (2023), Shambhu et al. (2022), and Shambhu et al. (2023) provide significant insights into malaria parasite detection using computational methods, there are several research gaps and discrepancies that merit further investigation.
One notable gap is the limited exploration of the impact of image quality and variation in blood smear preparation on segmentation and detection performance. While the studies have proposed advanced methods for segmentation, the robustness of these methods against varying image quality and preparation inconsistencies is not fully addressed. Future research could focus on developing algorithms that are resilient to such variations, thereby ensuring accurate detection even with suboptimal image conditions.
Another discrepancy lies in the type of segmentation technique employed. While Nugroho and Nurfauzi (2023) emphasize semantic segmentation with deep learning techniques, Shambhu et al. (2023) focus on edge-based segmentation. This methodological difference may lead to variations in performance, particularly in handling overlapping cells and differentiating between parasitized and non-parasitized cells. A comparative study directly assessing the strengths and limitations of both approaches in a variety of scenarios could be beneficial.
The studies also differ in their datasets; for instance, Nugroho and Nurfauzi (2023) created the PlasmoID dataset with images from rural Indonesia, which may have unique characteristics compared to datasets used in other studies. The generalizability of the proposed methods to other datasets and their performance across diverse geographical regions with different malaria parasite strains are not fully explored.
Furthermore, while the studies have proposed automated methods for segmentation and detection, there is a lack of comprehensive evaluation against manual segmentation by expert microscopists. Establishing a benchmark involving human experts could provide a more accurate assessment of the computational methods' performance and their practical utility in clinical settings.
Additionally, there is a research gap in the end-to-end automation of malaria detection. Most studies, including the ones reviewed, focus on segmentation and detection, but an integrated system that includes automated counting, morphological analysis, and species identification of malaria parasites remains underexplored. Such a system would be invaluable for scaling up malaria diagnosis in resource-limited settings.
Lastly, the scalability and implementation of these computational methods in real-world clinical workflows have not been thoroughly investigated. Issues such as computational cost, ease of use, and integration with existing laboratory information systems are crucial for the adoption of these technologies in routine diagnostics.
To summarize, while the reviewed studies have advanced the field of malaria parasite detection, there are significant opportunities for future research to address the gaps in image quality and preparation variance, comparative analysis of segmentation techniques, dataset generalizability, benchmarking against manual segmentation, end-to-end automation, and practical implementation challenges. These research directions could ultimately lead to more accurate, robust, and clinically viable solutions for malaria diagnosis.
In summary, the reviewed studies represent significant advancements in the application of deep learning techniques such as convolutional neural networks (CNNs), deep learning, and computer vision approaches for malaria parasite detection. However, several research gaps and discrepancies between studies remain that could guide future work. Addressing areas such as dataset variability, model optimization, interpretability, clinical integration, robustness, and error analysis will be important for developing fully-realised diagnostic technologies. Both theoretical understanding and practical impact could be further strengthened through efforts to standardize evaluation protocols, validate models across diverse settings, and integrate systems-level thinking from fields like healthcare, public health, ethics, and policy. Continuous innovation supported by multidisciplinary collaboration shows promise for eventually realizing accurate, scalable, and accessible malaria diagnostics with transformative potential. Overall, the implications of this research underscore its ability to enhance global health outcomes by advancing both scientific knowledge and real-world disease management capabilities.
This research was supported by the Tertiary Education Trust Fund (TET Fund) University-based Research Grant (UBR) 2023 under Umaru Musa Yar’adua University Katsina, Nigeria. We gratefully acknowledge the funding provided, which was instrumental in the successful completion of our research and this review.
Ahmed, K., Rahman, Z., Shaikh, R., & Hossain, S. (2023). Malaria parasite detection using CNN-based ensemble technique on blood smear images. Biomedical Signal Processing and Control, 75, 103573. [Crossref].
Akruwala, Y., & Prajapati, K. (2022). Malaria parasite detection using deep learning. Journal of Medical Imaging and Health Informatics, 12(8), 1832-1840. [Crossref].
Alassaf, A., & Sikkandar, M. Y. (2022). Intelligent deep transfer learning-based malaria parasite detection and classification model using biomedical image. Computers, Materials & Continua, 72(3), 5639-5652. [Crossref]
Alonso, P. L., & Tanner, M. (2013). Public health challenges and prospects for malaria control and elimination. Nature Medicine, 19(2), 150-155. [Crossref]
Alraba'nah, Y., & Toghuj, W. (2024). A deep learning based architecture for malaria parasite detection. Computer Methods and Programs in Biomedicine, 232, 107285. [Crossref].
Amin, J., Anjum, M., Sharif, A., Raza, M., Kadry, S., & Nam, Y. (2022). Malaria parasite detection using a quantum-convolutional network. Computers in Biology and Medicine, 145, 105498. [Crossref].
Babu, K., Reddy, M., Bhavanam, S., Chimma, A., Nuthalapati, P., & Kovvuri, S. (2023). Malaria parasite disease classification using deep learning neural networks. Journal of Artificial Intelligence and Applications, 17(4), 254-269. [Crossref].
Babu, M., Naveen, T., Pradeep, V., Varma, G., & Charan, K. (2024). Web-based deep learning system for malaria parasite detection in granular blood samples. Journal of Medical Systems, 48(1), 25-36. [Crossref].
Baptista, V., Peng, W., Minas, G., Veiga, M., & Catarino, S. (2022). Review of microdevices for hemozoin-based malaria detection. Sensors, 22(6), 2145. [Crossref].
Calderaro, A., Piccolo, G., & Chezzi, C. (2024). The laboratory diagnosis of malaria: A focus on the diagnostic assays in non-endemic areas. International Journal of Molecular Sciences, 25(2), 695. [Crossref].
Chen, X., Hsu, Y., Chen, Y., Goh, C., & Yu, W. (2022). Deep learning based malaria-infected cell detection and parasite life stage classification method. Biomedical Engineering Online, 21(1), 13-27. [Crossref].
Dev, A., Fouda, M. M., Kerby, L., & Fadlullah, Z. M. (2023, August). On improving malaria parasite detection from microscopic images: A comparative analytics of hybrid deep learning models. In 2023 11th International Conference on Information and Communication Technology (ICoICT) (pp. 417-422). IEEE. [Crossref]
Dutta, A. K., Mageswari, R. U., Gayathri, A., Dallfin Bruxella, J. M., Ishak, M. K., Mostafa, S. M., & Hamam, H. (2022). Barnacles mating optimizer with deep transfer learning enabled biomedical malaria parasite detection and classification. Computational Intelligence and Neuroscience, 2022(1), 7776319. [Crossref]
Efrizoni, L., Amin, R., & Rizali, A. (2023). Detection of malaria parasites in human blood cells using convolutional neural network. International Journal of Computer Science and Network Security, 23(2), 147-157. [Crossref].
Garba, S., Abdullahi, M., Bashir, S., & Abisoye, O. (2022). Implementation of malaria parasite detection and species classification using dilated convolutional neural network. Healthcare Technology Letters, 9(3), 85-92. [Crossref].
Gill, K.S., Anand, V., & Gupta, R. (2023). An Efficient VGG19 Framework for Malaria Detection in Blood Cell Images. In Proceedings of the 2023 3rd Asian Conference on Innovation in Technology (ASIANCON). IEEE. [Crossref].
Gupta, M., Dungarwal, Y., Sabarish, Y., & Kumar, K. (2023). A novel approach towards detecting malaria parasite in thin blood smears using a sequential convolutional neural network (CNN) model. Computers in Biology and Medicine, 147, 105845. [Crossref].
Hasikin, K. (2023). Automated mosquito vector identification and malaria parasite detection using deep learning approach. IEEE Transactions on Image Processing, 32, 2051-2062. [Crossref].
Herfandi Sitanggang, Ones Nasution, Muhammad Nguyen, Huy Jang and Yeong Min. (2024). Real-Time Patient Indoor Health Monitoring and Location Tracking with Optical Camera Communications on the Internet of Medical Things. Applied Sciences. 14. 1153. [Crossref].
Islam, M., Nahiduzzaman, M., Goni, M., Sayeed, A., Anower, M., Ahsan, M., & Haider, J. (2022). Explainable transformer-based deep learning model for the detection of malaria parasites from blood cell images. Sensors, 22(12), 4358. [Crossref]
Jones, C., & Murugamani, C. (2023). Malaria parasite detection on microscopic blood smear images with integrated deep learning algorithms. Microscopy Research and Technique, 86(4), 295-305.
Jusman, Y., Aftal, A., Tyassari, W., Kanafiah, S., Hayati, N., & Mohamed, Z. (2023). Classification of parasite malaria in schizont stage with GoogleNet and VGG-19 pre-trained models. Journal of Information Technology and Software Engineering, 13(2), 45-56. [Crossref]
Kadiyala, S., Kotecha, S., & Kulkarni, S. (2022). CNN based deep learning approach for automatic malaria parasite detection. International Journal of Engineering and Technology, 11(3), 199-207. [Crossref]
Kavitha, P., Manimala, M. G., & Sanchana, M. R. (2023, December). Malaria detection using neural network. In 2023 Intelligent Computing and Control for Engineering and Business Systems (ICCEBS) (pp. 01-05). IEEE. [Crossref].
Khan, D., Shah, N., Nuzhat, R., Majid, A., Alquhayz, H., & Khan, A. (2022). Malaria parasite classification framework using a novel channel squeezed and boosted CNN. Computers in Biology and Medicine, 145, 105401. [Crossref]
Koirala, A., Jha, M., Bodapati, S., Mishra, A., Chetty, G., Sahu, P., Mohanty, S., Padhan, T., Mattoo, J., & Hukkoo, A. (2022). Deep learning for real-time malaria parasite detection and counting using YOLO-mp. IEEE Access, 10, 48731-48742. [Crossref]
Krishnadas, P., Chadaga, K., Sampathila, N., Rao, S., SwathiK., S., & Prabhu, S. (2022). Classification of malaria using object detection models. Healthcare Technology Letters, 9(4), 119-128. [Crossref]
Kumar, A., Nelson, L., Rasher, S., & Surendran, R. (2024). MosquitoNet based deep learning approach for malaria parasite detection using cell images. Biomedical Signal Processing and Control, 80, 104137. [Crossref]
Kundu, T., & Anguraj, D. (2023). A performance analysis of machine learning algorithms for malaria parasite detection using microscopic images. Journal of Computer Science and Technology, 38(2), 409-423. [Crossref]
Kundu, T., & Anguraj, D. (2023). Optimal machine learning-based automated malaria parasite detection and classification model using blood smear images. Journal of Medical Systems, 47, 202. [Crossref]
Liu, Z., Liu, H., & Sun, Y. (2023). Detection and classification of malaria parasite based on improved YOLOv5 model. IEEE Transactions on Biomedical Engineering, 70(2), 364-375. [Crossref]
Mahmood, S. N., Mohammed, S. S., Ismaeel, A. G., Clarke, H. G., Mahmood, I. N., Aziz, D. A., & Alani, S. (2023, June). Improved malaria cells detection using deep convolutional neural network. In 2023 5th International Congress on Human-Computer Interaction, Optimization and Robotic Applications (HORA) (pp. 1-4). IEEE. [Crossref]
McDermott, J. (2020). Convolutional neural networks—Image classification with Keras. LearnDataSci. Retrieved from https://www.learndatasci.com/tutorials/convolutional-neural-networks-image-classification-keras/​
Mehanian, C., Jaiswal, M., Delahunt, C., Thompson, C., Horning, M., Hu, L., ... & Bell, D. (2017). Computer-automated malaria diagnosis and quantitation using convolutional neural networks. In Proceedings of the IEEE International Conference on Computer Vision Workshops (pp. 116-125). [Crossref]
Mishra ,R.; S. S. Saranya and M. Shafahad, (2022). Analysis of different Machine Learning and Deep Learning Techniques for Malaria Parasite Detection. Fourth International Conference on Emerging Research in Electronics, Computer Science and Technology (ICERECT), Mandya, India, 2022, pp. 1-7. [Crossref]
Mohammed, W., Taha, D., & Abduljabbar, H. (2023). Deep learning for malaria diagnosis: Leveraging convolutional neural networks for accurate parasite detection. Wasit Journal for Pure Sciences, 2(3). [Crossref]
Nascimento, M. S., Costa, M. G., & Costa Filho, C. F. F. (2023, November). Detection of malaria parasites in thick blood smear images using shallow neural networks and digital image processing techniques. In 2023 19th International Symposium on Medical Information Processing and Analysis (SIPAIM) (pp. 1-4). IEEE. [Crossref]
Nugroho, H. A., & Nurfauzi, R. (2023). A combination of optimized threshold and deep learning-based approach to improve malaria detection and segmentation on PlasmoID dataset. FACETS, 8, 1-12. [Crossref]
Ohdar, K., & Nigam, A. (2023). A robust approach for malaria parasite identification with CNN based feature extraction and classification using SVM. Biomedical Signal Processing and Control, 81, 104445. [Crossref]
Oladele, M.O., Adepoju, T.M., Badrudeen, A.A., Olalude, A.O., & Festus, B. (2023). Development of a Mobile-Based Convolution Neural Network Framework for Rapid Malaria Detection from Blood Smears. FUOYE Journal of Engineering and Technology, 8(3), 324-328. [Crossref]
Ozbilge, E., Güler, E., & Ozbilge, E. (2024). Ensembling object detection models for robust and reliable malaria parasite detection in thin blood smear microscopic images. Expert Systems with Applications, 236, 118116. [Crossref]
Ozsahin, D., Mustapha, M., Duwa, B., & Ozsahin, I. (2022). Evaluating the performance of deep learning frameworks for malaria parasite detection using microscopic images of peripheral blood smears. Computers in Biology and Medicine, 144, 105336. [Crossref]
Pandey, S., Dingankar, R., Singh, K., Rose, A., & Singh, T. (2023). Automated malaria parasite detection in blood smear images with CNNs. Journal of King Saud University - Computer and Information Sciences, 35(2), 233-241. [Crossref]
Phumkuea, T., Nilvisut, P., Wongsirichot, T., & Damkliang, K. (2023). A new computer-aided diagnosis of precise malaria parasite detection in microscopic images using a decision tree model with selective optimal features. International Journal of Medical Informatics, 169, 104772. [Crossref]
Pandiarajan M. Professor, M., Sorna, J., & Scholar, I. (2023). Automated malaria parasite detection for legal blindness accessibility using hybrid deep learning techniques. Procedia Computer Science, 207, 567-574. [Crossref]
Raj, H., Revathi, D., & Srivastava, S. (2023). Malaria Detection in Blood Smeared Images Using Convolutional Neural Networks. International Journal for Research in Applied Science and Engineering Technology, 11(5), 88-93. [Crossref]
Rajput, A., Saxena, S., & Adhikari, M. (2023). Edge-Assisted Framework for Malaria Parasite Detection on Cell Images Using Federated Learning. Journal of Artificial Intelligence and Technology, 3(1), 45-51. [Crossref]
Rees-Channer, R., Bachman, C., Grignard, L., Gatton, M., Burkot, S., Horning, M., Delahunt, C., Hu, L., Mehanian, C., Thompson, C., Woods, K., Lansdell, P., Shah, S., & Chiodini, P. (2023). Evaluation of an automated microscope using machine learning for the detection of malaria in travelers returned to the UK. Malaria Journal, 22(1), 134-142. [Crossref]
Rehan, M., Khalid, A., & Nasreen, F. (2022). White blood cell differential fluorescence abnormal scattergram: A useful indicator for early detection of malarial parasite. Journal of Clinical Pathology, 75(11), 772-777. [Crossref]
Rocha, M., Claro, M., Neto, L., Aires, K., Machado, V., & Veras, R. (2022). Malaria parasites detection and identification using object detectors based on deep neural networks: a wide comparative analysis. IEEE Transactions on Biomedical Engineering, 69(9), 2385-2395. [Crossref]
Saini, A., Guleria, K., & Sharma, S. (2023). Deep Learning Based Model for Malaria Disease Detection Using Convolution Neural Network. International Journal of Advanced Computer Science and Applications, 14(4), 183-191. [Crossref]
Santoshi, K., Saranya, G., Reddy, C., Reddy, C., Gyananandu, K., & Tej, G. (2023). Deep Learning based Web App for Malaria Parasite Detection in Granular Blood Samples. Journal of Medical Systems, 47(2), 112-120. [Crossref]
Shahadat, N. (2023). Mobile-Based Deep Convolutional Networks for Malaria Parasites Detection from Blood Cell Images. Biomedical Signal Processing and Control, 80, 104059. [Crossref]
Shambhu, S., Koundal, D., Das, P., Hoang, V. T., Tran-Trung, K., & Turabieh, H. (2022). Computational methods for automated analysis of malaria parasite using blood smear images: recent advances. Computational Intelligence and Neuroscience, 2022(1), 3626726. [Crossref]
Shambhu, S., Koundal, D., & Das, P. (2023). Edge-Based Segmentation for Accurate Detection of Malaria Parasites in Microscopic Blood Smear Images: A Novel Approach using FCM and MPP Algorithms. Biomedical Engineering Letters, 13(3), 263-275. [Crossref]
Shankaralingappa, A., Poongodi, R., & Babu, T. (2023). Successful detection and species differentiation of malarial parasite using an automated hematology analyser: A case series. Journal of Laboratory Medicine, 47(1), 112-117. [Crossref]
Sherif, F., & Mohammed, A. (2023). Detection of Malaria Infection Using Convolutional Neural Networks. Procedia Computer Science, 207, 191-198. [Crossref]
Shriya G. J., D., Babu, D., Kumar, S., & Kishore, R. (2023). Automatic detection of Plasmodium parasite using convolution neural network. IEEE Access, 11, 27836-27845. [Crossref].
Swastika, W., Pradana, B., Widodo, R., Sitepu, R., & Putra, G. (2022). CNN Based Transfer Learning for Malaria Parasite Detection Using Thin-Blood Smear Images. Journal of Computational Medicine, 12(3), 201-210. [Crossref]
Swastika, W., Pradana, B., Widodo, R., Sitepu, R., & Putra, G. (2023). Web-Based Application for Malaria Parasite Detection Using Thin-Blood Smear Images. Software: Practice and Experience, 53(1), 45-52. [Crossref]
Telang, H., & Sonawane, K. (2023). COVID-19 and Malaria Parasite Detection and Classification by Bins Approach with Statistical Moments Using Machine Learning. Journal of Computational and Graphical Statistics, 32(1), 123-134. [Crossref]
Visser, T., Ramachandra, S., Pothin, E., Jacobs, J., Cunningham, J., Menach, A. L., ... & Aidoo, M. (2021). A comparative evaluation of mobile medical apps (MMAS) for reading and interpreting malaria rapid diagnostic tests. Malaria Journal, 20(1), 1-12. [Crossref]
Wardhani, P., Butarbutar, T. V., Adiatmaja, C. O., Betaubun, A. M., & Hamidah, N. (2020). Performance comparison of two malaria rapid diagnostic tests with real-time polymerase chain reaction and gold standard of microscopy detection method. Infectious Disease Reports, 12(S1), 8731. [Crossref]
Wojtas, N., Wieczorek, M., & Bełkot, Z. (2023). Malaria detection using custom Semantic Segmentation Neural Network Architecture. Journal of Imaging Science and Technology, 67(3), 1-10. [Crossref]
World Health Organization. (2022). WHO malaria policy advisory group (MPAG) meeting, October 2022.
Yang, F., Poostchi, M., Yu, H., Zhou, Z., Silamut, K., Yu, J., ... & Antani, S. (2019). Deep learning for smartphone-based malaria parasite detection in thick blood smears. IEEE Journal of Biomedical and Health Informatics, 24(5), 1427-1438. [Crossref]
Zaniolo, Luiz & Marques, Oge. (2020). On the use of variable stride in convolutional neural networks. Multimedia Tools and Applications. 79. [Crossref]
Zedda, L., Loddo, A., & Ruberto, C. (2023). YOLO-PAM: Parasite-Attention-Based Model for Efficient Malaria Detection. Artificial Intelligence in Medicine, 132, 102464. [Crossref]
Zhang, X., & Chen, B. (2023). Research on Malaria Parasite Detection in Thick Blood Smears Based on YOLO-VF. Journal of Image Processing and Computer Vision, 31(4), 225-235. [Crossref]
Zhong, Y., Dan, Y., Cai, Y., Lin, J., Huang, X., Mahmoud, O., Hald, E., Kumar, A., Fang, Q., & Mahmoud, S. (2023). Efficient Malaria Parasite Detection From Diverse Images of Thick Blood Smears for Cross-Regional Model Accuracy. IEEE Transactions on Medical Imaging, 42(6), 1345-1354. [Crossref]