The overall aim of this project is to develop and evaluate deep learning methods for large-scale medical image analysis, in close collaboration with Sahlgrenska University Hospital and international partners. We focus on using CT, MRI, and PET data to improve diagnosis, outcome prediction, and treatment evaluation across several disease areas. Our work builds on access to large and diverse datasets from Swedish and international clinical patients, as well as openly shared datasets. In total, we expect to handle around 140,000 imaging studies, many with linked clinical, textual, and follow-up information. Methodologically, we will train and test a range of modern architectures, including convolutional neural networks (CNNs), vision transformers (ViTs), and large vision–language models adapted for medical imaging. We will also explore weakly supervised and self-supervised learning to make use of partially labeled data, and include uncertainty estimation and explainability to better understand and trust model predictions. A key part of our work is to evaluate how well these models generalize across scanners, hospitals, and populations. The project relies heavily on large-scale computational resources for both model training and experimentation. Training modern deep networks with millions of parameters requires extensive GPU time and efficient parallelization, especially when tuning hyperparameters and testing different architectures. Access to NAISS infrastructure is therefore essential for carrying out these experiments in a reasonable timeframe and at sufficient scale. The results of this work will contribute both to methodological advances in medical AI and to improved clinical tools for disease detection and prognostication in areas such as cardiovascular disease, oncology, and pulmonary medicine.