Interpolation of Erythrocytes and Leukocytes Microscopic image using MsR-CNN with Yolo v9 Model

In the field of hematology, pathologists often use different lenses on a microscope to carefully inspect samples, switching between different magnifications as needed. Typically, they start with low magnifications and then switch to higher magnifications as necessary to identify the shape of Erythrocytes. However, this process can be time-consuming as it is subjective in nature and may also lead to different perceptions. Therefore, deep learning-based models are needed to have a computer-aided diagnostic that can assist pathologists with higher accuracy. Before magnification, accurately identifying the location of erythrocytes is essential. Traditionally, this step was also performed manually under a microscope, making the process laborious and prone to errors. However, with advancements in deep learning-based detection algorithms, cell detection has become faster, more reliable, and less prone to human error. The objective of this research is to first detect cells and then magnify the detected cells at different scale factors. For detection, YoLo v9 is used, and for magnification, our proposed method (MsR-CNN) is employed. A comparative analysis of available methodologies with our proposed method (MsR-CNN) was conducted. Various experiments were performed using techniques such as KNN, Random Forest, and CNN to improve image quality. After applying each methodology, results were compared using metrics such as PSNR and SSIM. It was found that the proposed methodology, MsR-CNN, performed better than the aforementioned methods, achieving 35.09 PSNR and 0.96 SSIM. Additionally, semantic segmentation was performed using Otsu’s method to segment the ROI(nucleus of the erythrocytes).