Enhanced Malaria Detection Using 2D CNN and Transfer Learning: An Efficient Deep Learning Solution

Abstract

Introduction: Effective treatment of malaria, a potentially fatal illness spread by mosquito bites and brought on by parasites, depends on prompt and precise diagnosis. Convolutional Neural Networks (CNN), a deep learning technology, have demonstrated significant promise in automating illness identification through medical imaging. Objectives: The primary objective of this study is to develop and evaluate a 2D CNN-based deep learning model combined with transfer learning to accurately detect malaria-infected blood smear images. The model aims to improve diagnostic accuracy while maintaining efficiency and generalizability. Methods: A dataset containing 13,000 blood smear images (Parasitized and Uninfected) from a reputed medical institution was utilized. Preprocessing techniques such as image resizing, normalization, and thresholding were applied to enhance the dataset. The model was trained using a 2D CNN architecture with transfer learning, optimized over 20 epochs using the TensorFlow library. An 80:20 split was used for training and testing to validate model performance, and early stopping was applied to prevent overfitting. Results: The proposed 2D CNN model achieved an impressive accuracy of 97.45% during training and 96.94% on testing data. Precision, recall, and F1-score for both Parasitized and Uninfected classes ranged between 0.93 and 0.97, demonstrating the model's robustness. The confusion matrix highlights minimal misclassifications, and the performance graphs reflect effective learning despite slight overfitting in validation loss. Conclusions: The study demonstrates that the 2D CNN with transfer learning is a reliable and efficient solution for automated malaria detection. The high accuracy and balanced performance metrics validate its potential for clinical implementation, particularly in resource-limited settings. Future research can focus on multi-class classification, model optimization, and real-world deployment to further enhance its applicability.