Rapid 3D Imaging at Cellular Resolution for Digital Cytopathology with a Multi-Camera Array Scanner (MCAS)

Academic Background

Optical microscopy has long been the standard method for diagnosis in cytopathology. However, traditional whole-slide scanners, while capable of automatically imaging and digitizing large sample areas, are slow and expensive, thus not widely adopted. Particularly in the clinical diagnosis of cytology specimens, samples are often spread over large areas and are thick, requiring 3D imaging. Existing whole-slide scanning technologies often take several hours to complete scans of thick samples, significantly limiting their clinical application. Therefore, developing a technology capable of rapid and efficient 3D imaging of thick samples has become a critical challenge in the field of cytopathology.

This paper introduces a novel Multi-Camera Array Scanner (MCAS) aimed at addressing this challenge. MCAS, through a parallelized microscope design, can perform high-resolution 3D imaging of large-area, thick samples in a very short time, combined with machine learning techniques to assist pathologists in rapid diagnosis.

Source of the Paper

This paper was co-authored by Kanghyun Kim, Amey Chaware, Clare B. Cook, and others, from institutions such as Duke University, University of California Berkeley, and Ramona Optics Inc. The paper was published in npj Imaging in 2024.

Research Process

1. MCAS System Design

The MCAS system consists of 48 independent 13-megapixel CMOS image sensors, each equipped with a custom high-resolution objective lens. These sensors are tightly arranged in a 6×8 grid, with each camera simultaneously imaging different areas. Through this design, MCAS can capture 624 megapixels in a single snapshot, significantly faster than traditional whole-slide scanners.

The MCAS system is also equipped with sample stages and associated firmware, enabling rapid movement of up to three slides to complete 3D scanning. The system can scan thick samples at resolutions of 1.2 µm and 0.6 µm, with a scanning depth of up to 150 µm. This parallelized imaging approach theoretically increases imaging speed by 48 times.

2. Rapid Whole-Slide Imaging

To demonstrate the rapid whole-slide 3D imaging capabilities of MCAS, the research team scanned 16 Diff-Quik-stained lung cytology samples, including adenocarcinoma-positive and benign samples. Using a 0.3 numerical aperture (NA) objective lens and a 5 µm axial step size, MCAS completed the scanning of three slides in less than 5 minutes, equivalent to less than 2 minutes per slide. In comparison, existing whole-slide scanners typically take 1 minute for 2D scanning and tens of minutes for 3D scanning.

3. Adequacy Assessment of Thyroid Specimens

The rapid 3D data acquisition and GigaViewer navigation capabilities of MCAS enable remote examination of cytology smears. The research team applied MCAS to the task of adequacy assessment of thyroid fine-needle aspiration (FNA) smears. By digitizing 26 thyroid FNA smears using MCAS and displaying them to three pathologists for evaluation, the results showed that MCAS’s adequacy decisions were on par with current standards, with a sensitivity of 100% and specificity of 94.4%.

4. Adenocarcinoma Detection in Lung Specimens

To more efficiently analyze and examine 3D cytology smear data, the research team explored the application of machine learning algorithms to MCAS image data. Specifically, they used MCAS to image five lung FNA smears and trained an object detection model using the YOLOv7 algorithm to locate adenocarcinoma regions in the smears. The model achieved a mean average precision of 0.645 and a recall of 0.73.

5. Adenocarcinoma Classification in Lung Specimens

In the diagnosis of lung FNA smears, determining the spatial location of disease is not always necessary; the primary goal is to classify the type of cancer. Therefore, the research team designed a slide-level classification network based on Multiple Instance Learning (MIL). Through 4-fold cross-validation, the model achieved an average test accuracy of 0.930 and an AUC of 0.969.

Main Results

1. MCAS System Design: MCAS can perform rapid 3D imaging of thick samples at resolutions of 1.2 µm and 0.6 µm, significantly faster than traditional whole-slide scanners.
2. Rapid Whole-Slide Imaging: MCAS completed the scanning of three lung cytology slides in less than 5 minutes, demonstrating its efficient 3D imaging capabilities.
3. Adequacy Assessment of Thyroid Specimens: MCAS’s remote assessment capabilities are on par with current standards, effectively optimizing cytology workflows.
4. Adenocarcinoma Detection: Using the YOLOv7 algorithm, MCAS can rapidly locate adenocarcinoma regions in lung FNA smears, with a recall of 0.73.
5. Adenocarcinoma Classification: The MIL-based slide-level classification model performed excellently in 4-fold cross-validation, with an AUC of 0.969.

Conclusion

MCAS, through its parallelized microscope design, significantly improves the speed of 3D imaging of thick cytology samples, completing whole-slide scanning in just a few minutes. Combined with machine learning algorithms, MCAS not only assists pathologists in rapid diagnosis but also optimizes cytology workflows, improving the efficiency and accuracy of clinical diagnosis. In the future, with further optimization and cost reduction of the MCAS system, its application prospects in digital pathology will be even broader.

Research Highlights

1. Rapid 3D Imaging: MCAS can complete whole-slide 3D imaging of thick cytology samples in just a few minutes, significantly improving scanning speed.
2. Machine Learning-Assisted Diagnosis: Through the YOLOv7 and MIL algorithms, MCAS can rapidly locate and classify adenocarcinoma regions, assisting pathologists in diagnosis.
3. Remote Assessment Capabilities: MCAS’s GigaViewer navigation enables remote examination of cytology smears, optimizing cytology workflows.
4. Cost Efficiency: MCAS’s imaging speed is significantly faster than traditional whole-slide scanners, with lower costs, offering broad application potential.

Other Valuable Information

The design of the MCAS system is not limited to cytopathology but can also be extended to other fields requiring rapid, high-resolution 3D imaging, such as biomedical research and industrial inspection. In the future, the research team plans to further optimize MCAS’s illumination system and 3D data processing capabilities to enhance its performance in various application scenarios.