to BioTechniques free email alert service to receive content updates.
Use and validation of epithelial recognition and fields of view algorithms on virtual slides to guide TMA construction
Sanford H. Barsky1,2, Lynda Gentchev3, Amitabha S. Basu4, Rafael E. Jimenez3, Kamel Boussaid4, and Abhi S. Gholap4
Full Text (PDF)
Supplementary Material

Quality of image evaluation

Each grid was first analyzed for image quality. Not every acquired image was of sufficient quality to be further evaluated. There were several criteria that determined image quality: luminosity, sharpness, and contrast. A number of techniques could be used to improve image quality and normalize image variations (8,9,10). With these techniques, some inferior images could be improved to the point where they could be evaluated. Still, some images could not be improved sufficiently and had to be discarded.

For images that were acceptable, preprocessing algorithms reduced the effect of variations in staining intensity and the effects of colored masks and other anomalies. These preprocessing algorithms consisted of four distinct functions: verifying the image content, mask removal, contrast enhancement, and background removal (Supplementary Materials; Table S3 and “Details of image enhancement and ERA” section).

Epithelial recognition algorithms

Once the image was optimized through the various preprocessing algorithms, the image was analyzed through ERAs which utilized both colorimetric (RGB) as well as intensity (grayscale) values (Supplementary Table S2). Epithelial cell detection initially utilized the Gaussian kernel. The ERAs initially created had to be further refined to more precisely define the area of identification (AOI) (Supplementary Materials; Tables S2, S4, S5, and S6 and “Details of image enhancement and ERA” section).

Calculating the epithelial area percentage

Once the epithelial area was defined in each FOV, the epithelial area percentage was calculated according to the following schema (Supplementary Table S7). The stromal area could be calculated as the reciprocal of the epithelial area. Epithelial percentage could vary from 0 to essentially 100%. The algorithm then ranked the FOVs by epithelial percentage and identified the highest 3–5 and earmarked these on the virtual slides for future TMA construction (Figure 2). Since some FOVs contained minimal tissue compared with the overall grid area, in those FOV the epithelial percentage could be artifactually elevated. To control for this possibility, FOVs whose total tissue area was <10% of the FOV area were automatically removed from further consideration (Supplementary Table S7).

Construction of TMAs

The TMAs were made from paraffin blocks containing human breast, colon and lung cancer tissues. We made three different types of TMAs and termed them manual TMA (manually selected cores, manually assembled into a TMA block), algorithmic TMA (algorithmically selected cores, manually assembled into a TMA block), and robotic TMA (algorithmically selected cores, robotically assembled into a TMA block). The manual TMA was constructed solely manually by the Ohio State University Research Pathology Core from the 300 retrieved cases. This manual selection required archival slide retrieval, re-review, manual circling of regions of interest, and alignment with the original tissue block from which the TMA was to be constructed. The TMA was made manually by selecting the three best areas, which were rich in cancer cells by matching the areas encircled on the glass slide with corresponding regions of the paraffin block (Figure 3). Each core measured 1 mm in diameter and was spaced 0.8 mm apart on a single glass slide. Three 1-mm tissue cores from each formalin-fixed paraffin embedded donor block were selected and precisely arrayed into a new recipient paraffin block. A previous study showed that =3 cores from each sample gave acceptable statistical analysis in TMAs in diverse tumor types (7); as such, the present work utilized similar multiple core sampling. The constructed TMAs were also subsequently scanned into virtual slides.

The algorithmic TMA was constructed from the same paraffin blocks under the guidance of the algorithms identifying those FOVs on the virtual slides with the highest cancer cell density. A trained technician used this grid as a template in making these algorithmically selected, manually assembled TMAs. TMA construction proceeded from this point in the same manner as manual TMA.

The robotic TMA was constructed using a robotic hardware device termed TMAker, which is operated under the control of the ERAs and FOV algorithms (Supplementary Materials; “Demonstration of TMAker” section). The device consists of a stacked carousel which can house up to 500 paraffin blocks and presents each block sequentially to two working stations: (i) a reading station where the identity of the block can be identified based on a barcode and the shape of the tissue contained within the block (by illumination) and compared with the corresponding virtual slide image contained in the database, and (ii) a coring station where an actuator arm, under the guidance of the algorithms, cores the areas of the block with the highest cancer cell density. The cores obtained are robotically deposited into a receiving block which eventually holds all of the cores of the TMA. This block is sectioned into slides, stained, and then scanned into virtual TMAs (Figure 3) in a manner identical to that used for both the manual TMA and the algorithmic TMA.

  1    2    3    4    5