Table 1.
Reference β values expressed in the sRGB color space
Fig 1.
Schematic diagram of the OpenGL virtual transillumination rendering algorithm.
a) Data flow in the OpenGL algorithm showing pixel data generated by the scan engine and A/D being processed by shaders. Fluorescence data is loaded into GPU RAM as a texture, processed by a shader running on hundreds or thousands of GPU cores and the final result is stored in the frame buffer for display. b) The relationship between vertex and pixel shaders. The vertex shader defines the quad’s position on screen and provides a mapping to the texture coordinates. Gray squares c1 to c4 show texture coordinate locations in GPU memory, while the associated vertices v1 to v4 are shown in blue. Blue dotted arrows show the association of the texture coordinates to the vertices by the vertex shader. The pixel shader performs the computations according to Beer’s law individually for each displayed pixel. The green grid indicates the pixel grid of the final image, while the green dotted arrows show the transform by a vertex shader which is run for each pixel.
Fig 2.
Comparison between histopathology and virtual transillumination H&E image generated by MPM.
a) H&E histopathology image with apocrine metaplasia (green box) and benign breast duct (blue box). b) Corresponding virtual transillumination H&E image. The higher axial resolution of the MPM image better resolves individual collagen fibers as compared to the H&E section, an effect that could be reduced by using a lower NA objective. Due to minor tilting of the histological cutting plane, the left side of the H&E image is slightly deeper than the MPM plane and therefore transects more of the duct on the bottom left. Scale bar: 500 μm.
Fig 3.
Enlargement of green boxed region in Fig 2 showing apocrine metaplasia on histopathology (left) and virtual transillumination MPM (right).
In the low magnification view (top row), a distinct border (black arrows) separating metaplasia from normal breast tissue is apparent. In the enlarged view (bottom row), both modalities show large, uniform, round nuclei with vesicular chromatin and distinct nucleoli arranged in a thin sheet of hexagonally packed cells. Note that while MPM is non-destructive, the H&E sectioning has partially stripped away the metaplastic tissue and introduced crack artifacts (left). Scale bar: 75 μm (top), 20 μm (bottom).
Fig 4.
Enlargement of blue boxed region in Fig 2 showing a breast duct with some element of tangential sectioning.
H&E histopathology (left). Virtual transillumination H&E image from MPM (right). Both modalities reproduce the duct structure as well as the surrounding collagen. Scale bar: 75 μm.
Fig 5.
Virtual transillumination H&E images rendered from epi-fluorescence MPM imaging.
a) Virtual transillumination Beer’s law algorithm. b) An additive method with nonlinear transfer function. c) Additive method with linear transfer function. The virtual transillumination algorithm enables both high contrast, realistic color rendering and perfect white point. In contrast, the additive method with nonlinear transfer function has reduced color accuracy due to the nonlinear transfer function shifting color slightly and minor green-shifting the backlight color to RGB [0.9 1.0 0.9], while the additive linear method has good color accuracy and perfect white point, but lower contrast. Of note, all three images span the entire 0.0 to 1.0 contrast range, and additional contrast enhancement is not possible without applying a nonlinear transfer function.
Fig 6.
Enlargement of green boxed region in Fig 5 showing a cluster of darker staining nuclei.
a) Virtual transillumination Beer’s law method. b) Additive method with nonlinear transfer function. c) Additive method with linear transfer function. Neither additive method has sufficient dynamic range to render both the nuclei (red arrow) and surrounding collagen fiber (blue arrow) accurately because of the strong spectral overlap between eosin and hematoxylin. Scale bar: 50 um.