Classification of Blood Smear Images using CNN and Pretrained VGG16: Computer Aided Diagnosis of Malaria Disease
Prema T. Akkasaligar, Santosh Pattar, Sakshi Gupta, Divya Barker, Bhagyashree Gunayyanavarmath · 2024
Malaria is a life-threatening disease caused by the plasmodium parasite present in female anopheles mosquitoes. It remains a significant public health concern in many parts of the world, mostly impacting vulnerable populations like children under five year of age, pregnant women, and HIV/AIDS patients. An early and accurate diagnosis is crucial for its effective treatment. Instead of relying on traditional hand-crafted features, in this work of computer aided diagnosis, we are using the image recognition capabilities of Convolutional Neural Networks (CNNs) and Visual Geometry Groups (VGG16). By feeding them the preprocessed image the model will automatically extract key features like cell shape, texture (presence of parasite pigments), and color intensity differences. These features invisible to the human eye, become crucial for the model to distinguish infected and uninfected cells. We have trained a CNN and VGG16 model on the labeled data, allowing it to adjust its internal parameters (weights and biases). The goal is to achieve high accuracy in differentiating infected cells and minimizing the misclassifications. The experiments are conducted using the National Institute of Health (NIH) malaria dataset. CNN model achieved 97.36% accuracy for the correct classification of cells, compared to 90.5% for the VGG16 model. These results helps to empower medical professionals with automated analysis of blood smears, leading to faster, more accurate malaria diagnoses and improved patient treatment outcomes.