Medical Image Analysis

点击翻译按钮获取中文摘要

In a data driven world, machine learning is expected to be a key tool for converting big data into tangible benefits. Neuroscience studies in particular collect large and complex data on a single patient, including brain imaging, functional scores, and clinical information, to adequately capture the complexities of brain diseases. Lately, advances in machine learning, and especially deep learning, have enabled the development of sophisticated computer-aided diagnosis tools for challenging tasks in many medical domains (e.g. dermatology or ophthalmology). Without doubt, novel machine learning methods also have huge potential for developing advanced methods for computer-aided data analysis (e.g. image segmentation), classification, and prognosis in neurological diseases. However, many datasets are typically needed to train robust and accurate machine learning models. This is a major reason preventing the development of sophisticated machine learning models for health care applications in general and in the neuroscience domain in particular because concerns about data privacy and sharing practices are becoming increasingly important while patients also have a better understanding of the value of their personal data and are increasingly concerned about data privacy. For this reason, we face the risk that novel data mining and machine learning approaches will not achieve their true potential for diagnosis and treatment support in patients with neurological diseases.Therefore, the aim of this CRC research program is to develop a novel distributed learning system for training machine learning models that does not require data sharing and centralized collection. Here, the patient data used for machine learning model training does not leave the contributing institution and is only used locally to train machine learning models on-site. The machine learning model, which does not contain any personally identifiable information, is then shared between institutions or with a central server where it is combined to a global model. Using this approach, many advanced machine learning models can be trained for computer-aided diagnosis of various diseases, which will have significant impact on neuroscience research and beyond.