DEV Community

EGBE UFUOMA
EGBE UFUOMA

Posted on

Automating Malaria Diseases

Abstract:
Malaria is a significant global health concern, leading to numerous deaths worldwide. The disease is characterized by symptoms such as fever, headache, fatigue, and vomiting, and in severe cases, it can result in coma and death. Accurate and timely identification and diagnosis of malaria are essential to combat its impact. Traditional detection methods relying on manual analysis of blood cells under a microscope are prone to false diagnoses due to limited resources and expertise. This study aims to develop an automated system with high precision for malaria diagnosis. Two machine learning techniques, Multilayer Perceptron (MLP) and Convolutional Neural Network (CNN), are employed to classify scanned images as positive or negative for malaria. The models are evaluated using cross-validation and the F-measure score to determine their effectiveness and accuracy. The CNN model achieves the highest accuracy of 79% in this study. Although the CNN model demonstrates better performance in predicting positive instances, the confusion matrix reveals a higher number of false positives (194) compared to false negatives (88) with respect to malaria diagnosis. Further analysis and improvement are required to minimize false positives and enhance the overall precision of the model. The results of this investigation contribute to the development of automated systems that can aid in accurate malaria detection, thereby mitigating the impact of the disease and improving patient outcomes

1.0 Background
Malaria is a long-standing and pervasive disease that primarily affects tropical nations, imposing significant social, economic, and health costs (Moss, 2008). It is caused by the bite of a parasitic plasmodium, posing a threat to human health (Antinori et al., 2012). Climate change and global warming have been identified as contributing factors to the spread and impact of malaria (Arnott et al., 2012). Changes in environmental temperature can disrupt the life cycle of malaria vectors and parasites, affecting their distribution and transmission patterns (Walter and Chandy, 2022; White, 2011).
Accurate diagnosis plays a crucial role in effective treatment and recovery from malaria. Blood smears stained with specific chemicals are commonly used to identify the presence of malaria parasites, which are then examined under a microscope to detect malaria-infected blood cells (Tilley et al., 2011). Skilled physicians visually analyze these blood smears to identify malaria-infected red blood cells (Singh and Daneshvar, 2013). Despite global efforts to reduce malaria-related deaths, the disease remains a major cause of mortality worldwide, infecting approximately one out of every 21 individuals and claiming the lives of about one million people annually (WHO, 2018; WHO, 2021). Malaria continues to pose a significant public health concern, affecting around 40% of the global population in various countries, irrespective of geographical location (Despommier et al., 2019).
The global incidence of malaria has increased due to factors such as the emergence of vaccine-resistant parasite species, deteriorating disease containment measures, and the impact of activities like migration and international travel (Alonso and Noor, 2017). Consequently, there is an urgent need for comprehensive and effective research on the global prevalence of malaria to address the role of accurate diagnosis in reducing complications and mortality (Murray et al., 2008).
In resource-limited settings, the manual examination of blood films for malaria diagnosis is a common practice that heavily relies on the expertise of trained microscopists (Singh and Daneshvar, 2013). However, this diagnostic method is subjective and prone to errors, leading to inconsistent results. Microscopists often work in isolation without a robust system to ensure the maintenance of their skills and diagnostic quality, potentially leading to erroneous field diagnostic decisions (Singh and Daneshvar, 2013).
The limitations and challenges associated with malaria diagnosis, including potential false-negative and false-positive results, have motivated efforts to automate the process. Automation aims to improve accuracy, efficiency, and accessibility in malaria parasite detection. Machine learning algorithms, such as multilayer perceptron (MLP) and convolutional neural networks (CNN), are being explored as potential tools for automated malaria diagnosis (Research Objectives).
Overall, addressing the challenges of malaria diagnosis through automation has the potential to enhance diagnostic accuracy and contribute to effective malaria management and control strategies.
1.2 Research question
The research on automatic identification of malaria images using neural networks and machine learning has primarily utilized small image datasets, leading to limited robustness and accuracy in the segmentation models. These models require substantial computational power and memory, preventing real-time or online training. To address this gap, a new research proposal aims to evaluate the performance of a CNN and multi-layer perception (MLP) algorithm. MLP, being non-linear and requiring less computational power, can run in real-time, while CNN excels in handling larger datasets. The question remains as to how efficient this proposed model will be in overcoming the limitations of previous approaches.
2.0 methodology
This chapter outlines the necessary tools for building the proposed systems, including Python libraries, Jupyter working environment, and Google Colab. It also discusses the steps involved in constructing the systems, which encompass data collection, data pre-processing, feature extraction, and classification. The chapter provides code snippets, diagrams, and textual explanations to elucidate the methodology.
2.1 Research Design
The research design refers to the methods and techniques employed by the researcher to develop a study. A well-structured research design increases the likelihood of achieving successful outcomes. It provides a foundation for various research methods and ensures the alignment of data collection and analysis approaches with the study's objectives, theoretical foundations, and methodology. In this case, the research design is quantitative, as it aims to assess the sensitivity and accuracy of malaria diagnosis numerically.

Image description

                 Diagram of the  research 
Enter fullscreen mode Exit fullscreen mode

2.2 Multilayer Perceptron (MLP)
The Multilayer Perceptron (MLP) is a neural network that involves a non-linear mapping between inputs and outputs. It consists of input and output layers, with multiple neurons arranged in one or more hidden layers. The neurons in an MLP can employ any activation function. Inputs are combined with initial weights, subjected to the activation function, and propagated to subsequent layers. Each layer feeds its computation results to the next layer, culminating in the output layers.

Image description
diagram of how Multilayer perceptron work

2.3Convolutional Neural Network (CNN)
A Convolutional Neural Network (CNN) is a deep learning algorithm specifically designed for image processing tasks. It can analyze input images, assign importance to different features, and differentiate objects within the images. CNNs can learn filters and characteristics, requiring minimal preprocessing compared to other classification algorithms. The architecture of a CNN is inspired by the organization of the Visual Cortex and the connectivity pattern of neurons in the human brain.

Image description

             Convolutional neural network
Enter fullscreen mode Exit fullscreen mode

2.4 The Concept Method Proposed
The objective of the proposed system is to improve malaria prediction compared to conventional methods and automated research. The methodology of the system consists of eight phases, including dataset collection, holdout cross-validation, creating classifier networks, training MLP and CNN networks, preserving the best classifiers, testing and validating networks, comparing MLP and CNN results, and determining the best classifier.

Figure 3 presents a diagram illustrating the methodology of the proposed system.

2.5 Data Collection
The malaria dataset obtained from the Kaggle website is used for this model. The dataset comprises 4,000 images, with two folders: "parasitized" containing malaria-infected cell images and "uninfected" containing images of healthy cells. The dataset has an equal distribution of cell pictures, eliminating the issues associated with imbalanced data.

Image description

       Diagram of infected cell with malaria 
Enter fullscreen mode Exit fullscreen mode

Image description

     Diagram of uninfected cell with malaria
Enter fullscreen mode Exit fullscreen mode

2.6 Pre-processing
Pre-processing refers to techniques used to improve the quality of images or prepare them for further processing. In this case, grayscale conversion is employed for MLP, and the Keras library is used for CNN model pre-processing.

2.6.1 Grayscale
Grayscale refers to representing an image using a range of tones from black to white, typically using 8-bit depth. The conversion results in a grayscale image with 0 representing black and 255 representing white. Code of the gray in below
def convert_to_grayscale(image_path):
# Open the image file
image = Image.open(image_path)
# Convert the image to grayscale
grayscale_image = image.convert("L")
# Save the grayscale image
grayscale_image.save("grayscale_image.jpg")

Display the grayscale image

grayscale_image.show()
Enter fullscreen mode Exit fullscreen mode

# Call the function with the path to your image file
convert_to_grayscale("image.jpg")
2.6.2 Keras
Keras is a deep learning framework that simplifies the creation and training of models. It provides a user-friendly interface and abstracts complex details, making it easier to get started with deep learning.
Code for keras for image preprocessing
from tensorflow import keras
from tensorflow.keras.preprocessing.image import ImageDataGenerator

Define the image data generator

datagen = ImageDataGenerator(rescale=1./255)

Load and preprocess the images

train_generator = datagen.flow_from_directory(
'train_directory',
target_size=(300, 300),
batch_size=32,
class_mode='binary'
)

Build and compile the model

model = keras.Sequential()

Add model layers...

model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])

Train the model

model.fit(train_generator, epochs=5)

2.7 Create Neural Network
When implementing a neural network, important considerations include determining the network's structure (number of layers and neurons), initializing weights and biases, and specifying activation functions. We can see the code below.
model = Sequential()

model.add(Conv2D(filters=32, kernel_size=(3,3),input_shape=image_shape, activation='relu',))
model.add(MaxPooling2D(pool_size=(2, 2)))

model.add(Conv2D(filters=64, kernel_size=(3,3),input_shape=image_shape, activation='relu',))
model.add(MaxPooling2D(pool_size=(2, 2)))

2.8 Training Neural Network
Training a neural network is the process of implementing algorithms that enable the network to learn and improve its performance on specific tasks. It involves adjusting the weights and biases of the network's neurons to minimize error and maximize accuracy. We will be using Multi-Layer Perceptron (MLP) and Convolutional Neural Network (CNN) models (Larochelle et al., 2009).
3.4.5 Testing of the Neural Network
Testing the neural network ensures its correct functioning and helps identify any issues. It also allows developers to evaluate the network's performance on new data, ensuring it continues to work correctly amidst changes (Larochelle et al., 2009).

2.9 Model Evaluation
The most commonly used metrics for evaluating a model include:
2.9.1 Confusion Matrix
The confusion matrix presents model predictions in a table format. It consists of rows and columns representing predicted and actual classes, respectively. The matrix provides the basis for evaluating performance metrics such as true positives (TP), false negatives (FN), false positives (FP), and true negatives (TN). (Larochelle et al., 2009).

Image description

         diagram of Confusion Matrix
Enter fullscreen mode Exit fullscreen mode

2.9.2 Accuracy
Accuracy measures the ratio of correct predictions to the total number of predictions.
Accuracy = (TP + TN) / (TP + FP + FN + TN) (W. McKinney, 2017).
2.9.3 Precision
Precision is the ratio of true positives to the sum of true positives and false positives.
Precision = TP / (TP + FP) (W. McKinney, 2017).
2.9.4 Recall
Recall is the ratio of true positives to the sum of true positives and false negatives.
Recall = TP / (TP + FN) (W. McKinney, 2017).
2.9.5 F1-score
The F1-score is a metric derived from precision and recall, representing their harmonic mean.
F1-score = 2 * (Precision * Recall) / (Precision + Recall) (W. McKinney, 2017).
2.9.6 Macro avg
Macro average is used in classification when observations across classes are imbalanced. It calculates the average of metrics such as accuracy across all classes. (W. McKinney, 2017).

2.9.7 Weighted Avg
Weighted average is the sum of all class F-scores obtained by multiplying their respective proportions.(W. McKinney, 2017).
2.9.8 SUPPORT
The support refers to the number of instances in the classification dataset.
SUPPORT = (W. McKinney, 2017).

2.9.9.Comparing MLP and CNN Results
Testing the neural network ensures its correct functioning and helps identify any issues. It also allows developers to evaluate the network's performance on new data, ensuring it continues to work correctly amidst changes (Larochelle et al., 2009).
3.0 Result
This chapter discusses the implementation of the proposed method. It covers the data collection, pre-processing, and classification processes. Additionally, it presents the results obtained from two different machine learning algorithms and compares their accuracy.

3.1 Experimental Setup
3.1.1 Python
Python version 3 was chosen as the programming language for this study. Python is commonly used for image dataset application development due to the availability of numerous third-party libraries. The following Python libraries were used in this project:
I. NumPy: The "Numerical Python" (NumPy) module is the standard for numerical operations in Python. It provides N-dimensional array object types and methods for array manipulation.
II. SciPy: It is a Python library for data science, offering a wide range of scientific calculation routines. It extends NumPy and is often used for scientific computations.
III. Pandas: Pandas are used for data-related activities such as cleaning and analysis. It provides data structures like data frames, designed for structured data manipulation.
IV. Matplotlib: Matplotlib is used for generating visually appealing and robust graphs and plots. It offers an object-oriented API for integrating charts into applications.
V. TensorFlow: TensorFlow is an open-source machine learning library developed by Google. It allows the creation of custom Convolutional Neural Networks (CNNs) and Multi-Layer Perceptron's (MLPs).
VI. PyTorch: PyTorch is a Python toolkit for machine learning that enables the construction of neural networks and complex models. It is widely used in deep learning applications.
VII. Keras: Keras is a high-level API for neural networks written in Python. It can work on top of TensorFlow, CNTK, or Theano and offers a simple model construction approach.
3.1.2 Jupyter Notebook
A Jupyter Notebook-based integrated development environment (IDE) was used for this study. Jupyter Notebook is a web-based, open-source, and free-to-use program that allows the creation and sharing of documents containing live code, formulas, visualizations, and annotations. It offers various functionalities such as data cleaning, mathematical modeling, predictive modeling, data visualization, and machine learning.
3.2 Machine Learning Classifier
In this section, the accuracy of the machine learning classifiers is measured. The classifier is trained using 80% of the data, and the remaining 20% is used for validation to assess the model's accuracy.
Two machine learning algorithms were employed to test the accuracy of the analysis. The results from both algorithms were compared to obtain more reliable outcomes.
3.3 Multilayer Perceptron (MLP)
3.3.1 Pre-Processing
The following modules were imported for pre-processing:
i. NumPy: Used for high-dimensional arrays and array manipulation.
ii. Pandas: Used to read, manage, and make changes to data frames.
iii. Scikit-image (skimage): Imported the color module to convert images to grayscale and used the resize function to resize images in the dataset.
iv. Matplotlib: Used for plotting graphs.
v. OS: Used to retrieve the file path of the data.
vi. Glob: Used to retrieve all the files in the specified path.
vii. Pickle: Used for object serialization before writing to a file.
3.3.2 Building the Model
The following modules were used to build the MLP model:
i. NumPy: Used to convert images to a numpy array.
ii. Pandas: Used to read and modify the data frame.
iii. Pickle: Used to load the preprocessed data file.
iv. MLP Classifier from scikit-learn's neural_network module: Used to create the MLP neural network.

Image description

       Diagram of the building of MLP model  built 
Enter fullscreen mode Exit fullscreen mode

3.3.3 Result of MLP
In the MLP model, two rounds of testing were performed. In the first round, the model had 10 perceptrons with a single hidden layer and was tested on a dataset consisting of 4,000 samples, with 1,928 infected and 2,072 uninfected samples. The results showed an accuracy of 0.64, a precision of 0.62 for infected samples, and a precision of 0.67 for uninfected samples. The recall was 0.64 for infected samples and 0.64 for uninfected samples.
In the second round, the number of hidden layers was increased to
3 layer. The results showed an improvement in precision, with a precision of 0.65 for infected samples and 0.69 for uninfected samples. The accuracy also improved to 0.67. This indicates that increasing the number of hidden layers can lead to improved accuracy.
Comparing the two instances, in the first instance, the model had a precision of 62% for infected predictions and 67% for uninfected predictions. The accuracy of negative predictions was 67% and positive predictions was 62%. This suggests that the model is more likely to produce false negatives than false positives. The overall accuracy of the model was 64%.
In the second instance, the model had a precision of 69% for infected predictions and 65% for uninfected predictions. The accuracy of negative predictions was 68% and positive predictions was 66%. This suggests that the model is more likely to produce false negatives than false positives. The overall accuracy of the model improved to 67%.
The results from the second instance of the MLP model indicate that hyperparameter tuning can further improve the performance of the MLP model.

Image description
diagram of both the first and second performance

3.4 In the CNN approach, image preprocessing was performed using an image generator. The images were rescaled with a parameter of (1./255) to enhance the image features. The dataset was split into training and validation sets, with 80% used for training and 20% for testing. The training dataset was divided into batches of 16 to control the accuracy of the error gradient estimate during CNN training.

The CNN model was built using a sequential model. The model included 2D convolutional filters to enable the layers to learn features from the images. Max pooling was applied to reduce the dimensionality during training. Dropout layers were added to prevent overfitting. The model was then flattened to convert the input data into a single dimension, and dense layers were added to receive input from the previous layers. The overall structure of the CNN model is depicted in the image below.

           Building CNN Model 
Enter fullscreen mode Exit fullscreen mode

The results of the CNN model showed an accuracy of 79%. The model performed well in predicting positive cases, with 91% accurate predictions. However, it struggled in predicting negative cases, with only 62% accurate predictions. This indicates that the model is more likely In the CNN approach, image preprocessing was performed using an image generator. The images were rescaled with a parameter of (1./255) to enhance the image features. The dataset was split into training and validation sets, with 80% used for training and 20% for testing. The training dataset was divided into batches of 16 to control the accuracy of the error gradient estimate during CNN training.

The CNN model was built using a sequential model. The model included 2D convolutional filters to enable the layers to learn features from the images. Max pooling was applied to reduce the dimensionality during training. Dropout layers were added to prevent overfitting. The model was then flattened to convert the input data into a single dimension, and dense layers were added to receive input from the previous layers. The overall structure of the CNN model is depicted image below

Image description
Building CNN Model
The results of the CNN model showed an accuracy of 79%. The model performed well in predicting positive cases, with 91% accurate predictions. However, it struggled in predicting negative cases, with only 62% accurate predictions. This indicates that the model is more likely to produce false positives than false negatives. The confusion matrix revealed a higher number of false positives (194) compared to false negatives (63).

Image description

From the results, a subset of 1200 malaria samples was used due to computational limitations. The dataset contained 690 infected cell samples and 510 uninfected cell samples. The macro average was 0.75, indicating a balanced performance between the two classes. The accuracy of the infected samples was 0.75, while the accuracy of the uninfected samples was 0.83. This suggests that the model had a lower true positive rate for infected samples compared to uninfected samples. Based on the system's results, the model should focus on identifying whether a person is malaria-free, as the uninfected class has a higher chance of being correctly classified.
In conclusion, this study successfully developed machine learning models for automated detection of malaria using MLP and CNN approaches. The CNN model achieved a higher accuracy of 79% compared to the MLP model. However, the CNN model exhibited a higher rate of false positives. The research contributes to the knowledge of malaria diagnosis automation and suggests future studies to improve the models by adding more data and optimizing hyperparameters. The limitations of the research include long training and testing times and the cost associated with increasing data and hidden layers to produce false positives than false negatives. The confusion matrix revealed a higher number of false positives (194) compared to false negatives (63).
Figure 12: Result of CNN model
3.4.2 From the results, a subset of 1200 malaria samples was used due to computational limitations. The dataset contained 690 infected cell samples and 510 uninfected cell samples. The macro average was 0.75, indicating a balanced performance between the two classes. The accuracy of the infected samples was 0.75, while the accuracy of the uninfected samples was 0.83. This suggests that the model had a lower true positive rate for infected samples compared to uninfected samples. Based on the system's results, the model should focus on identifying whether a person is malaria-free, as the uninfected class has a higher chance of being correctly classified.
In conclusion, this study successfully developed machine learning models for the automated detection of malaria using MLP and CNN approaches. The CNN model achieved a higher accuracy of 79% compared to the MLP model. However, the CNN model exhibited a higher rate of false positives. The research contributes to the knowledge of malaria diagnosis automation and suggests future studies to improve the models by adding more data and optimizing hyperparameters. The limitations of the research include long training and testing times and the cost associated with increasing data and hidden layers.

References
1.Moss W., Shah S., Morrow R. The History of Malaria and its Control. Int. Encycl. Public Health. 2008:389–398. doi: 10.1016/B978-012373960-5.00374-9.
2.Antinori S., Galimberti L., Milazzo L., Corbellino M. Biology of human malaria plasmodia including Plasmodium knowlesi. Mediterr. J. Hematol. Infect. Dis. 2012;4:e2012.013. doi: 10.4084/mjhid.2012.013.
3.Arnott A, Barry AE, Reeder JC (2012). "Understanding the population genetics of Plasmodium vivax is essential for malaria control and elimination". Malaria Journal. 11: 14. doi:10.1186/1475-2875-11-14
4.Walter, Kristin; John, Chandy C. (2022-02-08). "Malaria". JAMA. 327 (6): 597. doi:10.1001/jama.2021.21468
5.Tilley L, Dixon MW, Kirk K (2011). "The Plasmodium falciparum-infected red blood cell". International Journal of Biochemistry & Cell Biology. 43 (6): 839–42. doi:10.1016/j.biocel.2011.03.012
6.Singh B., Daneshvar C. Human Infections and Detection of Plasmodium knowlesi. Clin. Microbiol. Rev. 2013;26:165–184. doi: 10.1128/CMR.00079-12.
7.WHO (2021). World Malaria Report 2021. Switzerland: World Health Organization. ISBN 978-92-4-004049-6.
8.World Health Organization. World Malaria Report 2018. WHO; Geneva, Switzerland: 2018.
9.Despommier DD, Griffin DO, Gwadz RW, Hotez PJ, Knirsch CA (2019). "9. The Malarias". Parasitic Diseases (PDF) (7 ed.). New York: Parasites Without Borders. pp. 110–115.
10.Alonso P., Noor A.M. The global fight against malaria is at crossroads. Lancet. 2017;390:2532–2534. doi: 10.1016/S0140-6736(17)33080-5
11.Murray C.K., Gasser R.A., Magill A.J., Miller R.S. Update on Rapid Diagnostic Testing for Malaria. Clin. Microbiol. Rev. 2008;21:97–110. doi: 10.1128/CMR.00035-07.
12.Sarkar PK, Ahluwalia G, Vijayan VK, Talwar A (2009). "Critical care aspects of malaria". Journal of Intensive Care Medicine. 25 (2): 93–103.
13.World Health Organization False-negative RDT results and implications of new reports of P. falciparum histidine-rich protein 2/3 gene deletions. WHO. 2017 doi: 10.1186/1475-2875-10-166
14.Loddo, C. Di Ruberto, and M. Kocher, “Recent advances of malaria parasites detection systems based on mathematical morphology,” Sensors, vol. 18, no. 2, p. 513, 2018.
15.Z. Jan, A. Khan, M. Sajjad, K. Muhammad, S. Rho, and I. Mehmood, “A review on automated diagnosis of malaria parasite in microscopic blood smears images,” Multimedia Tools and Applications, vol. 77, no. 8, pp. 9801–9826, 2018.
16.V. N. Orish, V. F. De-Gaulle, and A. O. Sanyaolu, “Interpreting rapid diagnostic test (RDT) for Plasmodium falciparum,” BMC Research Notes, vol. 11, no. 1, pp. 850–856, 2018.
a.Vijayalakshmi and B. R. Kanna, “Deep learning approach to detect malaria from microscopic images,” Multimedia Tools and Applications, vol. 79, no. 21, pp. 15297–15317, 2020.
17.Ghate, A.M. (2012). Automatic Detection of Malaria Parasite from Blood Images.
18.D. Yang, G. Subramanian, J. Duan et al., “A portable image-based cytometer for rapid malaria detection and quantification,” PLoS One, vol. 12, no. 6, Article ID e0179161, 2017.
19.S. S. Savkare and S. P. Narote, “Automated system for malaria parasite identification,” in Proceedings of the 2015 international conference on communication, information & computing technology (ICCICT), pp. 1–4, Mumbai, India, January 2015.
20.B. Maiseli, J. Mei, H. Gao, and S. Yin, “An automatic and cost-effective parasitemia identification framework for low-end microscopy imaging devices,” in Proceedings of the 2014 International Conference on Mechatronics and Control (ICMC), pp. 2048–2053, Jinzhou, China, July 2014.
21.A.-N. Aimi Salihah, M. Yusoff, and M. Zeehaida, “Colour image segmentation approach for detection of malaria parasites using various colour models and k-means clustering,” WSEAS Transactions on Biology and Biomedicine, vol. 10, 2013.
22.Abdul-Nasir, A.S., Mashor, M.Y., & Mohamed, Z. (2017). Clustering approach for unsupervised segmentation of malarial Plasmodium vivax parasite.
23.Saraswat, S., Awasthi, U., & Faujdar, N. (2017). Malarial parasites detection in RBC using image processing. 2017 6th International Conference on Reliability, Infocom Technologies and Optimization (Trends and Future Directions) (ICRITO), 599-603.
24.Parveen, R., Jalbani, A.H., Shaikh, M., Memon, K.H., Siraj, S., Nabi, M., & Lakho, S. (2018). Prediction of Malaria using Artificial Neural Network.
25.Sajana, T.K., & R.Narasingarao, M. (2018). Classification of Imbalanced Malaria Disease Using Naïve Bayesian Algorithm. International journal of engineering and technology, 7, 786.
26.Opeyemi A. Abisoye, Rasheed G. Jimoh, " Comparative Study on the Prediction of Symptomatic and Climatic based Malaria Parasite Counts Using Machine Learning Models", International Journal of Modern Education and Computer Science(IJMECS), Vol.10, No.4, pp. 18-25, 2018.DOI: 10.5815/ijmecs.2018.04.03
27.Usha D, Mallikarjunaswamy M. S, 2017, Detection of Malaria Based on the Blood Smear Images Using Image Processing Techniques, INTERNATIONAL JOURNAL OF ENGINEERING RESEARCH & TECHNOLOGY (IJERT) NLPGPS – 2017 (Volume 5 – Issue 21)
28.Memeu, D.M. (2014). A Rapid Malaria Diagnostic Method Based On Automatic Detection And Classification Of Plasmodium Parasites In Stained Thin Blood Smear Images.
29.Ahirwar, N., & Acharya, B. (2012). ADVANCED IMAGE ANALYSIS BASED SYSTEM FOR AUTOMATIC DETECTION AND CLASSIFICATION OF MALARIAL PARASITE IN BLOOD IMAGES.
Pandit, P., & Anand, A. (2016). Artificial Neural Networks for Detection of Malaria in

Top comments (0)