Multiplex immunohistochemistry was carried out as described previously
18 (link) using an Opal Automation IHC Kit (Akoya Bioscience) and Bond RXm autostainer (Leica Biosystems). The sections were stained with DAPI and the following Abs: anti‐CD45 (clone D3F8Q; Cell Signaling Technology; 1:500), anti‐DIG (clone 9H27L19; Thermo Fisher Scientific; 1:500), anti‐Ki‐67 (clone D3B5; Cell Signaling Technology; 1:500), and anti‐pCK (rabbit poly; Bioss Antibodies; 1:250). Stained slides were mounted with ProLong Diamond (Thermo Fisher Scientific) and imaged with a Mantra Quantitative Pathology Workstation (Akoya Biosystems). The obtained images were analyzed with inForm Tissue Finder software (Akoya Biosystems). To calculate the percentage of Ki‐67 positive cancer cells, inForm software was trained to detect tissue and cell phenotypes using machine‐learning algorithms based on the following criteria: areas with pCK expression = tumor, other areas = stroma, pCK+ CD45− cells = cancer cells, pCK− CD45+ = blood cells, and pCK− CD45− = other cells. inForm software computed the percentage of Ki‐67‐positive cells among cancer cells. The average percentage was calculated from five images for each specimen.