High-Precision Lensless Microscope on a Chip Based on In-Line Holographic Imaging

Xiwei Huang, Yangbo Li, Xuefeng Xu, Renjie Wang, Jiangfan Yao, Wentao Han, Maoyu Wei, Jin Chen, Weipeng Xuan, Lingling Sun2021 · DOI

TL;DR

This paper presents a novel, high-precision lensless microscope on a chip that utilizes in-line holographic imaging and an iterative phase recovery algorithm. This compact and low-cost device overcomes the resolution limitations of traditional lensless shadow imaging, achieving improved image sharpness and enabling accurate cell identification and counting for point-of-care testing applications.

What Problem Does This Paper Address?

Traditional optical microscopes, while powerful, are often large, expensive, and require specialized training, limiting their accessibility in resource-poor or remote settings. Lensless microscopes offer a promising alternative due to their small size, low cost, and ease of use. However, existing lensless imaging techniques, such as shadow imaging, often suffer from low resolution. While fluorescence imaging can offer better resolution, it requires additional components, increasing complexity and cost. This paper addresses the need for a high-resolution, portable, and cost-effective lensless microscopy solution, particularly for point-of-care testing (POCT), by enhancing in-line holographic imaging with advanced phase recovery techniques.

Methods

The core of the proposed system is an in-line holographic imaging setup. This involves illuminating a sample with a coherent light source (achieved by using an LED with a pinhole) and capturing the interference pattern (hologram) directly on a CMOS image sensor. The key innovation lies in the iterative phase recovery algorithm used to reconstruct the sample's image from the captured hologram. This algorithm, based on the principles of Fresnel integral diffraction and angular spectrum theory, iteratively refines the phase information lost during the holographic recording process. Specifically, it involves:

  1. Backpropagation and initial phase estimation: Propagating the recorded intensity to the object plane and assuming an initial phase.
  2. Constrained reconstruction: Using a thresholding method to define the object boundary and retaining information within it while replacing external information with background backpropagation.
  3. Forward propagation and amplitude/phase combination: Propagating the reconstructed field back to the sensor plane and combining the recovered phase with the amplitude from the original holographic image.

This iterative process, typically requiring less than 15 iterations, significantly improves the resolution and sharpness of the reconstructed image compared to traditional shadow imaging. For cell analysis, the researchers employed machine learning tools (specifically, the Trainable Weka Segmentation plugin in Fiji) for cell identification and segmentation, followed by a cell counting algorithm that involved binarization, hole filling, and separation of adhered cells.

Datasets

The paper utilizes two primary types of "datasets" for evaluation:

  • United States Air Force (USAF) resolution plate: This standard optical target was used to quantitatively assess the resolution capabilities of the developed microscope system.
  • Mixed solution of mouse ascites tumor cells and polystyrene particles: This biological sample was prepared and used to demonstrate the system's ability to image and analyze biological entities, specifically for cell counting.

The paper does not introduce a new benchmark dataset but rather uses these established targets and biological samples to validate its proposed method against existing techniques.

Key Insights and Ideas

  • Holography's inherent advantage: In-line holography inherently captures both amplitude and phase information, providing a richer dataset for reconstruction compared to shadow imaging, which only captures amplitude.
  • The power of iterative phase recovery: The iterative phase recovery algorithm is crucial for overcoming the limitations of direct holographic reconstruction, particularly the "twin image" artifact, and significantly enhancing image resolution and clarity.
  • Lensless microscopy for POCT: The combination of a compact, lensless design with advanced image processing makes this technology highly suitable for portable, low-cost point-of-care diagnostic applications, especially in resource-limited settings.
  • Machine learning for automated analysis: Integrating machine learning for cell identification and counting automates a critical step in biological analysis, improving efficiency and potentially accuracy.

Key Results

  • Resolution Improvement: The iteratively phase-recovered holographic image achieved a resolution 1.41 times higher than traditional shadow imaging. Specifically, the system could resolve group 6, element 4 (90.5 Line Pairs/mm) on the USAF resolution plate, whereas shadow imaging could only resolve group 6, element 1 (64 Line Pairs/mm).
  • Image Quality: The holographic recovered image showed improved sharpness and clarity compared to the shadow image, as quantified by image quality assessment metrics (Kblur and Q values).
  • Cell Counting Accuracy: The automated cell counting using the developed platform showed results similar to manual counting, demonstrating its potential for accurate biological sample analysis. The system successfully identified and counted a mixed sample of mouse ascites tumor cells and polystyrene particles.

Limitations and Future Work

  • Limitations:
    • The paper acknowledges that for smaller or more scattered samples, simple backpropagation might suffice, but for more complex samples, the twin image interference can be problematic, necessitating the phase iterative recovery method.
  • While the system is described as low-cost, specific cost figures are not provided.
  • The current implementation uses a basic yellow LED, and further improvements in light source coherence could potentially enhance performance.
  • Future Work:
    • The authors suggest that the platform has specific application prospects in preventing the spread of diseases and early detection, implying a focus on clinical applications.
  • The success with machine learning for cell counting suggests potential for further integration of AI for more complex biological analyses and diagnostics.
  • The compact and low-cost nature of the system points towards further development for widespread adoption in resource-limited areas.

Summary updated

BibTeX
@misc{huang2021high,
  author = {Xiwei Huang and Yangbo Li and Xuefeng Xu and Renjie Wang and Jiangfan Yao and Wentao Han and Maoyu Wei and Jin Chen and Weipeng Xuan and Lingling Sun},
  title = {{High-Precision Lensless Microscope on a Chip Based on In-Line Holographic Imaging}},
  year = {2021},
  doi = {10.3390/s21030720},
}
Figure