Histopathological assessment of biopsies and resected tumors provides information that is central for clinical decision making. Today’s cancer diagnostics are performed manually by pathologists using a microscope or digitized images. However, human assessments are associated with a high degree of uncertainty and variability, thus providing unreliable information that lead to suboptimal treatment decisions. Today’s imprecise cancer diagnostics lead to under- and over-treatment that affects patient outcome and creates an unsustainable cost burden for the society. The most promising area for applied artificial intelligence currently in health-care is for medical image analysis applications. Image-based diagnoses are the central part of pathology. Therefore, we develop computer-based models using artificial intelligence (AI, deep learning) for assessment of histopathology images, aimed at outperforming routine clinical pathology assessment of the most problematic biomarkers in solid tumors (e.g: breast cancer, melanoma, urothelial carcinoma) providing consistent and well-calibrated model-based classifications directly from the standard Hematoxylin-Eosin (HE) and immunohistochemically (HC) stained stained tissue sections. Moreover, we also leverage AI-based models to capture and extract more clinically relevant information than human eyes can detect. Furthermore, we develop, validate and implement innovative multidisciplinary and multimodal AI models, trained on millions of images both HE and IHC stained sections), and on full clinical data aimed at increasing cancer survival through a quantum leap in diagnostic precision compared to the current standard qualitative pathology. The project aims to transition qualitative and manual aspects of pathology into highly quantitative outputs and, as a result, augment efficiencies of physician-led pathology workflows, quality assurance and personalized medicine efforts. The project will use large, retrospective, multi-institutional cohorts of TNBC, urothelial cancer, and melanoma. Our TNBC cohorts consist of all breast cancer patients in Sweden who received neoadjuvant immunotherapy (up to date approx. 400 TNBC patients). Furthermore, via collaborations we have collected up to date approx. 130 patients with urothelial cancer and 250 patients with melanoma who received immunotherapy. Our specific aims are to develop next generation AI models for computational histopathology to obtain feasibility for these aims as follows: 1) Predict biomarker status to improve the current diagnostics of the standard-of-care biomarkers ( e.g: ER, HER2, PgR, Ki67, sTILs) 2) Improved prognostic clinical/histopathology models based on AI derived biomarkers. 3) Novel predictive models for treatment response risk assessment for neoadjuvant immunotherapy treated patients with triple negative breast cancer, melanoma and urothelial carcinoma.