Background: Most prostate cancers diagnosed are low to intermediate risk and face the decision on whether active surveillance or active treatment represents the optimal treatment option. To incorporate a reproduceable parameter in treatment planning, prognostic marker assessment using multi-gene panels has been frequently proposed. However, the fluctuating tumor purity can reduce the predictive value of such tests. A more accurate prognostication would be the quantification of prognosis markers exclusively on cancer cells. Methods: To enable automated prognosis marker quantification, a framework for automated prostate cancer detection that comprises three different artificial intelligence analysis steps and an algorithm for cell-distance analysis have been developed and validated using 8 marker BLEACH&STAIN mfIHC (i.e., panCK, p63, PSA, PSMA, INSM1, AR, Ki-67, CD56) in a tissue microarray and prostate biopsy cohort of 10’234 prostate cancers. Results: The approach for automated prostate cancer detection mimics the way pathologists use for identification of cancer cells and identified epithelial cells (CKpan positive) that were non-adjacent to basal cells (p63 positive). Validation experiments revealed that the optimal distance between p63⁺ basal cells and benign panCK⁺ cells was 23 µm to exclude benign glands from the prognosis marker analysis. A distance of 23µm as threshold for automated cancer detection showed a 39% higher accuracy compared to the cutoff value 5 µm. To compare this automated mfIHC-based prognosis marker assessment with the frequently used bright field IHC-based QuPath software package, and the manual prostate identification by pathologists, a cohort of 30 biopsies from routine clinical practice was assessed. An excellent agreement between the mfIHC-based automated cancer detection and manual cancer identification by pathologists was found (intraclass correlation [ICC]: 0.94 [95% CI 0.87 – 0.97]), while the QuPath-based model showed a 20 % deviation from the gold standard (manual assessment by pathologists). Accordingly, the automated mfIHC-based approach was used to search for a prognosis marker panel for routine clinical practice. 5 out of 6 analyzed prognosis markers (PSA, PSMA, INSM1, AR, Ki-67) were significantly liked to biochemical recurrence in univariate analysis (p ≤ 0.021) and 2 of 6 parameters (AR, Ki-67) were independent in multivariate analysis (each p≤ 0.0078). Several prognosis scores of different marker combinations were found that showed strong prognostic relevance in univariate analysis (p < 0.001) and were independent from Gleason groups, pT, pN, serum PSA and R-Status (p = 0.001). Conclusions: Automated prostate cancer identification enables fully automated prognosis marker assessment in routine clinical practice using deep learning and mfIHC.