Development of Automated High-Throughput Digital Microscopy With Deep Learning for Enhanced Blood Smear Imaging

A microscope is essential in scientific and medical research, enabling the magnification of specimens too small for the naked eye. The conventional method of acquisition of images requires pathologists or technicians to manually focus the microscope and examine one slide at a time, making the process tedious, especially in health emergencies. However, manual focusing in traditional microscopes often leads to human errors, image drift, and fatigue. Therefore, the work aims to design and develop an automated high-throughput optical digital microscope-based device for scanning and capturing blood smear images of 10 peripheral blood smear slides sequentially in a batch, focusing on a particular field of view at an objective lens magnification of 40. Maintaining focus is especially challenging in high-magnification microscopes due to their sensitivity to vibrations, which can blur images. This study proposes an automated focusing system combining an optical assembly setup with a deep learning algorithm for real-time image acquisition. Leveraging transfer learning with pre-trained VGG-16 and Mobile Vision Transformer (ViT) models, the system overcomes the computational challenges of training CNNs from scratch. By fine-tuning the model's upper layers on a smaller dataset, it efficiently learns task-specific features while minimizing computation, time, and overfitting. The Mobile ViT model outperforms the VGG-16 model, achieving an accuracy of 99.28%, which establishes MobileViT as a more reliable and efficient model for the acquisition of well-focused images. The approach improves image sharpness, reliability, and stability, enhancing the efficiency of microscopic imaging. This advanced system offers practical solutions for biomedical imaging diagnostics and advancements in microscopy analysis.