HoloHema: Digital Holographic Hematology Analyzer

TL;DR

This work evaluates digital holographic microscopy (DHM) for rapid differential white blood cell counting in point-of-care devices, introducing two prototypes: a lens-based system and a lensless system. The lensless variant achieves higher throughput and 3-part differential accuracy of $92.65\%$ and 5-part accuracy of $89.44\%$, while enabling monocyte distribution width (MDW) estimation as a sepsis biomarker. Across chapters, the study integrates pixel super-resolution and multi-wavelength holography to enrich cell information, and demonstrates a physics-informed neural network (PINN) for holographic reconstruction. A parallel CGH track, HoloTile, presents rapid, speckle-suppressed holography via sub-hologram tiling and PSF shaping, with potential for RGB color CGH applications. Overall, the thesis lays foundational work for DHM-based dWBC in PoC devices and proposes several avenues (SR, multi-wavelength, PINN, HoloTile) to enhance performance, robustness, and practical deployment in clinical settings.

Abstract

This industrial Ph.D. project, carried out in collaboration between Radiometer Medical ApS and SDU Centre for Photonics Engineering at the University of Southern Denmark, explored the use of digital holographic microscopy (DHM) for the purposes of differential white blood cell counts (dWBCs) in point-of-care (PoC) devices for acute care settings. Two DHM prototypes were developed; an initial lens-based system serving as the foundation for algorithm development, and experimental validation of the approach, achieving 89.6% classification accuracy on a 3-part differential, and a subsequent lensless system for simplified design and increased field-of-view (FoV). Both prototypes employed convolutional neural networks (CNNs) for cell classification. With further optimizations, the lensless system achieved classification accuracies of 92.65% and 89.44% on the 3-part and 5-part differential, respectively. With the lensless system, the derivation of the monocyte distribution width (MDW), a biomarker for sepsis, was also demonstrated. Additionally, pixel super-resolution and multi-wavelength DHM approaches were investigated to enhance the obtained cell information. Finally, a proof-of-principle physics-informed neural network (PINN) for holographic reconstruction was implemented, demonstrating the potential for machine learning (ML) reconstruction techniques. In summary, this work represents an initial exploration of DHM for dWBC in PoC devices, laying the groundwork for future research.

Figures (52)

• Figure 1: Simplified illustration of the hologram recording process. A monochromatic plane wave, $|\mathcal{R}_0|$, illuminates the complex transmission function $t(x_o, y_o)$ in the object plane. The transmitted field consists of both a plane wave component due to the background transparency in the object plane, and a scattered wave component due to the modulation by the objects. Following free-space propagation of the transmitted field a distance $z_r$, the intensity of the field, $\mathcal{I}_h$, is captured by a detector.
• Figure 2: Illustration of the formation of the twin-images in inline holography. A monochromatic plane wave illuminates the hologram which is expressed by the amplitude transmittance $t_h(x, y)$. The diffracted wave consists of terms corresponding to an image of the object transmittance term $t_o$, as well as a conjugate image, $t_o^*$. The images are accompanied by a constant background due to the background transparency of the hologram. To an observer, the two images appear as if located on either side of the hologram, separated by a distance of $2z_r$.
• Figure 3: Illustrations of the DHM configurations used in this thesis. (a) The lens-based setup as described in Chapter 3. (b) The lensless setup as described in Chapter 4. (c) The large FoV, unity magnification, lensless system as described in Chapters 5 and 6 for use with a pixel super-resolution algorithm and multicolor reconstructions.
• Figure 4: Illustration of the convolution between an input image and three convolutional kernels. The convolution between the input and each of the kernels results in separate feature maps madsen_algorithmic_2021-1.
• Figure 5: Illustration of the effect of the 2D convolution. A $3 \times 3$ kernel, is convolved with the pixelated digit "zero". The convolution results in a feature map that highlights the horizontal edges of the digit and suppresses the vertical madsen_algorithmic_2021-1.