Detection, Counting, and Classification of Blood Cells in Lens-Free Holographic Images

Point of care testing has become increasingly popular in recent years, as it provides convenience for patients as well as the ability for doctors to make better-informed treatment decisions. As new hardware is designed to collect data at or near the point of patient care, there is also a need to rapidly process this data and provide doctors and patients with meaningful results. This thesis focuses on designing computer vision algorithms for a novel point of care blood test. The blood test aims to produce the Complete Blood Count by acquiring lens-free holographic images of a blood sample and then analyzing the images to estimate the concentrations of the various types of cells in the patient's blood sample. It is therefore necessary to accurately detect, count, and classify blood cells in holographic images. However, there are many challenges associated with detecting, counting, and classifying blood cells in holographic images. In particular, most object detection and classification methods are not designed to work with low resolution, holographic images that contain thousands of objects. These unique challenges must be overcome in order to produce meaningful results for the point of care Complete Blood Count. This thesis focuses on methods for detecting, counting, and classifying white blood cells in lens-free holographic images. First, we propose a convolutional sparse dictionary learning and coding approach for detecting and counting instances of a repeated object in a reconstructed holographic image. Next, we propose an approach based on a probabilistic generative model for detecting, counting and also classifying white blood cell populations in holographic images, which capitalizes on the fact that the variability in a mixture of blood cells is constrained by physiology. Finally, we address cell detection, counting, and classification directly from holographic images. We describe a method for detecting objects directly in holograms while jointly reconstructing the image, which is achieved by assuming a sparse convolutional model for the objects being imaged and modeling the diffraction process responsible for generating the recorded hologram. We conclude with a convolutional neural network that can be used for jointly detecting and classifying white blood cells in holographic images. We successfully validate all of our methods using our datasets of holographic images of white blood cells, and we show that our proposed approaches outperform other methods for the tasks for detecting, counting, and classifying cells in holographic images