Public Defence: Steven Alexander Hicks

Applying machine learning to problems in medicine is a rapidly growing trend in nearly all areas of healthcare. 

The immense performance attained by using deep learning on tasks like image and time series analysis can profoundly impact how computers are used in hospitals or clinics. 

There is a lot to gain in developing these systems, both monetary and societal, where deep neural network-based models may someday be in charge of monitoring our health. 

However, despite the massive responsibility that we give these models, the approach of developing and evaluating these methods is often not clear. 

Medical artificial intelligence (AI) research usually has imprecise method descriptions, private data, closed-source implementations, and incomplete evaluations. 

This thesis studies at how AI can be used in different areas within medicine, where a primary focus is to look at the current state of transparency within medical AI systems research and aims to contribute to a more open and public research community. 

To achieve this, we collected and published several medical datasets, developed several AI models in various medical domains, performed an assortment of different experiments to validate the collected datasets, organized many competitions on medical AI applications, and examined adequate model evaluation methods. 

The work was done across four fields of medicine to get a thorough understanding of how transparent AI can be applied to different medical domains, which includes cardiology, assisted reproductive technology, gastroenterology, and mental
health.