Advanced Deep Neural Network Approaches for Classification and Denoising of Microscopic Malaria-Infected Cells in Thin Blood Smear Imaging
Main Article Content
Abstract
Image denoising poses a major challenge in the field of medical imaging, particularly when analyzing thin blood smear images to detect malaria-infected cells. Most traditional noise-freezing methods cannot accomplish image noise-freezing because the various noise types in medical images are too complex. This study presents a dedicated noise-freezing approach that leverages state-of-the-art deep neural network architectures to noise-freeze images of malaria-infected cells in thin blood smears. It uses a two-network method. The first network is designed to identify and classify some of the most prevalent noise types in the medical imaging field: speckle noise, salt and pepper noise, Poisson noise, and Gaussian noise. The first network helps to classify the noise, and based on the previous classification, the second network performs noise-freezing using the information from the first network to effectively remove the identified noise types. This customized method approach aims to provide a solution to these challenges while dealing with the specific noise types provided by images of malaria-infected cells. The results show that in addition to producing higher accuracy in classifying noise, the high quality of noise-free images has also improved significantly.