Please enable JavaScript.
Coggle requires JavaScript to display documents.
Malaria Parasite, Note: Training on False positives could help a network…
Malaria Parasite
Methods and Contributions
Improved public malaria dataset
Removed false positive images
Original total is 27,558
Reduced to 26,161
More efficient network
Deployment to low resource equipment
Data Augmentation
Random zoom
Width shift
Random rotation
Height shift
Shear intensity
Horizontal flip
Network Architecture
Use of Autoencoders
Used for dimension reduction
Addition of fully connected layer transform it to classifier
Model architecture
Encoder
Mapping to latent space
Rectified Linear Unit activation function
Max pooling
General Training
Comparison of data augmentation and original image
Use of CNN-SVM
Use of CNN-KNN
Distillation Training
For knowledge transfer
From complex model to smaller model
Autoencoder training
Outperformed other trainings
Life threatening disease
Caused by Plasmodium parasites
Curable but relies on early diagnosis
Detection Techniques
Microscopic diagnosis
Complete tests
Requires experts
Clinical Diagnosis & Polymerase Chain Reaction
Relies on human expertise
Lacking in remote areas
Rapid Diagnostic Test
Does not require human experts
Easy to deploy
Lacks needed tests
Lack of sensitivity
Quantification of density
Automatic malaria detection
Image extraction
blood smears
Microscope
Phones
Segmentation
Classification
Results and Discussion
Comparison of metric performances
Precision
Distillation (64) and Autoencoder (28) with Augmentation
99.29%
Sensitivity
General CNN with augmentation (32x32)
99.60%
F1 score
Autoencoder (28) with Augmentations
99.51%
Specificity
Distillation with Augmentation (64)
99.32%
Size
For deployment
73.70KB for autoencoder method
The proposed model outperformed state-of-the-art models
Accuracy
Autoencoder with Augmentations
99.23%
Note: Training on False positives could help a network if GANs are used
Fuhad, K. M. F., Tuba, J. F., Sarker, M. R. A., Momen, S., Mohammed, N., and Rahman, T. Deep learning based automatic malaria
parasite detection from blood smear and its smartphone based application.
Diagnostics 10, 5 (2020).