Computed tomography generates three-dimensional microscopic images of cells

Studying tissue with conventional optical microscopy suffers from a fundamental problem: using two-dimensional (2D) representations of inherently three-dimensional (3D) materials. A true 3D image of the architecture of individual cells and their components would provide pathologists with crucial information—such as shape and size—for diagnostic purposes. Here, we describe a microscope that enables quantitative imaging of 3D samples.

Preparing traditional specimens involves cutting thin (4–7μm) sections of tissue to minimize the overlap of cells during visualization. This thickness is less than the diameter of most cells; it is even less that then the diameter of the nucelus. As a result, images consist of incomplete cells, allowing only approximations of any quantitative characteristics of intact cells, such as their total amount of DNA. To enable 3D imaging of thick (20μm) sections of tissue—which provide pathologists with quantitative cellular characteristics and architectural features—we developed an optical computer-tomography (CT) microscope. This is an optical analogue of x-ray computer tomography.

In CT, a series of 2D images gets combined into a 3D image. More precisely, a 3D spatial distribution of an object's linear attenuation coefficients is constructed from projections through the sample recorded at different angles of the transmitted illumination beam.¹ Each element in the recorded projection corresponds to a line integral of the attenuation coefficient, which represents the total attenuation of the light beam as it goes along a straight line through the sample. In 1985, a team of scientists proposed applying CT to optical microscopy.²

Recently, we developed two new kinds of optical-CT microscope.^3,4 The first employs a digital micromirror device (DMD)—an array of hundreds of thousands of tiny micromirrors that can be individually controlled to illuminate the specimen at a specific angle—as a spatial light modulator. Another form of the instrument uses a computer controlled-optical scanner to move the light beam across the sample (see Figures 1 and 2). This system consists of two high numerical-aperture objective lenses, an optical scanner, a light source, and a light detector. A two-axis mirror equipped with motorized linear actuators serves as the optical scanner. The angular range is limited by the numerical aperture of the objective.

Fourier-related transformations, such as the Radon transformation, are currently used in almost all applications of straight-ray tomography with complete projection data. However, direct implementation of transform-based algorithms for our limited-angle (135°) data does not provide a satisfactory reconstruction. Consequently, we developed an algorithm that attempts to incorporate the advantages of transform-based and iterative approaches to CT reconstruction.⁵ Briefly, we iteratively apply feedback correction to the reconstructed image based on projection-error measures. The projection error is calculated as the difference between the original projection and the projection from the reconstructed image. We find that this algorithm improves the accuracy of reconstructions.

Using our optical-CT microscope and algorithm, we made quantitative reconstructions of normal and abnormal absorption-stained cells (Figure 3). According to our measurments, the axial and lateral resolution of this microscope is on the order of 1μm. Moreover, 3D images acquired with this microscope can be virtually sliced in any orientation or rotated in 3D space to analyze the tissue architecture (Figure 4).

Overall, our optical-CT microscope provides 3D images of absorption-stained, thick-tissue sections with a resolution sufficient to see subcellular structures. Moroever, we can measure histopathological features—such as the total amount of DNA—that are used in cancer detection and diagnosis. In addition, quantitative analysis has demonstrated good discrimination between normal and cancer cells.

This study was supported by the Canadian Institute of Health Research grant # 15349.