Based on the success of artificial intelligence (AI), its use for automated diagnostics of medical image data has become a major focus. Despite excellent results on prediction tasks involving big data, a naïve application of deep learning, i.e., without any prior knowledge of the domain, may not be the optimal solution when there are only small amounts of data for the prediction task at hand. This, however, is often the case in clinical studies and biological experiments. Therefore, it may be beneficial to integrate prior information into the learning technique. With that in mind, this book identifies novel macroscopic and microscopic imaging biomarkers for computed tomography and multiphoton microscopy by developing image processing and prior-informed learning techniques for research in pulmonology, oncology, and myology. A spectrum of learning methods is explored, ranging from the traditional, i.e., statistics or classical machine learning with handcrafted features, to the modern, i.e., deep learning and meta-learning, resulting in novel hybrid biomarker systems that seamlessly blend prior knowledge with the power of AI.
