Enhancing High-Resolution Malaria Parasite Detection In Blood Smears Using Deep Learning

Malaria remains a significant global health concern, with its debilitating effects claiming over 500,000 lives annually worldwide, and faces a lot of challenges stemming from delayed or erroneous diagnoses, primarily attributed to the reliance on manual microscopy. Although effective, this is a more time-consuming method and susceptible to human error. Recognizing the severity of this global health crisis, there is an imperative to automate the evaluation process, eliminating the need for extensive human intervention. To tackle this issue, cutting-edge technologies, including Convolutional Neural Networks (CNNs), Transfer learning and Adaptive Boosting are being used in microscopic blood slides to evaluate parasitemia. Through rigorous preprocessing, the dataset is enhanced to ensure diversity and robust model training. This innovative approach involves collecting images depicting infected and non-infected erythrocytes, which are then input into CNN models, specifically VGG16, in conjunction with transfer learning, enabling the model to leverage knowledge gained from a broader dataset, enhancing its ability to accurately classify malaria instances. While previous baseline papers using CNNs encountered drawbacks related to sensitivity and reliance on large labeled datasets, our study addresses these limitations through the incorporation of transfer learning, adaptive boosting and hyperparameter tuning techniques. These advancements aim to enhance adaptability, accuracy and robustness. The model training and evaluation process, incorporating CNN, transfer learning and post-adaptive boost, achieves an accuracy that surpasses 96%. This ensures a reliable and high-performance malaria detection system, offering a promising avenue for more effective and accurate disease detection in clinical settings.