Deep Learning for Blood Cells Classification Based on Multispectral Imaging for Improved Accuracy

Cell classification is critical in biomedical applications, where recent developments in Multi-spectral imaging (MSI) techniques can capture detailed information about different biological cell types. For such applications, convolutional neural networks (CNNs) are typically used based on monochromatic or RGB microscope images. Increasing the spectral channels has been shown to provide higher classification accuracy due to the increased features exposed by the additional wavelengths. In this paper, we employ 3D image tensors with both spatial and spectral information by employing 3D-CNNs that divide the 3D image data into small cubes, capturing distinguishing features in both spatial and spectral dimensions. In order to quantify the improvement caused by increasing the chromatic channels, we explore the use of MSI and CNNs for frog blood cells classification. In this study, we employ up to six distinct optical channels to gain an understanding of classification improvement. Furthermore, we evaluate different wavelength combinations, measuring the impact on classification accuracy. Results demonstrate the significant performance enhancement of CNNs with MSI, emphasizing the importance of selecting the right wavelength combination. Our cost-effective approach has the potential for cell classification in medical applications, benefiting rapid disease diagnosis and treatment.