Table 1 A comparison of related works concerning malaria diagnosis using AI techniques.
From: Automated multi-model framework for malaria detection using deep learning and feature fusion
Study | Year | Dataset | Method | Results | Pros | Cons |
|---|---|---|---|---|---|---|
Muhammad et al.11 | 2025 | 772 (616 + 156) images for rouleaux morphology and 772 for normal morphology | XceptionNet, ResNet 50, DenseNet, EfficientNetB4, customized CNN | Accuracy 99% for DeseNet-201 | Rouleaux formation morphology of the RBCs is detected with DL algorithms. | Unbalance of Giemsa-stained images with field-stained images. |
Ozsahin et al.3 | 2024 | Related to 2207 patients | MLR, ANN, ANFIS, and RF | Model performance is evaluated in R, R2, RMSE, and MSE | Developing hybrid ML models for malaria diagnosis. | Using only ML algorithms. Generalization of the approach is limited. |
Hoyos and Hoyos6 | 2024 | 222 original images and 666 augmented images | YOLOv8 | An accuracy of 91% for original images and 95% for augmented images | Detecting malaria parasites and leukocytes with high accuracy. | Low diversity of trained images. Low number of used datasets. |
Khan et al.2 | 2024 | 12,500 augmented microscopic images | R-CNN & ResNet- 18, 50, 101, 152 & GoogleNet | R-CNN outperformed the other classifiers with an accuracy of 91.21% | A novel framework was presented for malaria detection. | Cell morphological changes were not considered. |
Hemachandran et al.14 | 2023 | 27,558 microscopic images | CNN & ResNet-50 & MobileNet-V2 | MobileNet-V2 outperformed by an accuracy of 97%. | Comparing the performance of many DL models on a large dataset. | A traditional CNN was used. |
Muhammad et al.10 | 2023 | 51 images (rouleaux morphology) and 180 images (normal morphology) | Customized CNN for normal morphology and rouleaux morphology | Accuracy: 90.95% (300 × 300) & 87.75% (500 × 500) | Evaluating the CNN performance in detecting abnormal cell morphology across varying image sizes. | A relatively low number of images (6088). Shallow depth of CNN layers (5 layers) |
Ozsahin et al.15 | 2022 | 300 infected thick smear images and 319 thin smear images | Customized CNN, ResNet-50, VGG16, and Inception V3 | Best performance of thick smear images with an accuracy of 96.97%. | Developing a novel model for malaria diagnosis based on DL techniques. | Relative law number of images. |
Sunarko et al.1 | 2020 | Related to 150 infected patients and 50 healthy patients (2000 images) | Otsu’s method + K -clustering | Accuracy: 94.6% Specificity:96.2% Sensitivity: 93% | A threshold-based segmentation method was adopted for malaria diagnosis. | Low amount of data. Detecting parasites in schizont stage is limited. |
Kassim et al.13 | 2020 | Polygen set was 34,226 RBC, and point set was 162,443 RBC | RBCNet based on U-Net and faster R-CNN | Cell detection with an accuracy of over 97% | A novel DL algorithm was presented for the RBCs segmentation using a large dataset. | No classification of infected and non-infected cells. |
