This thesis presents an uncertainty quantification (UQ) system on medical classification imaging tasks and its practical use. Deep Neural Networks have shown tremendous success in numerous AI-related fields, for example, object detection, recognition, and health care. However, despite Deep Neural Networks exhibiting remarkable performance, we usually can not guarantee the modelling predictions to be absolutely correct. Therefore, estimation and quantification of uncertainty have become an essential parameter in Deep Learning practical applications, especially in medical imaging. Measuring uncertainty can help with better decision making, early diagnosis, and a variety of tasks.
    
In this thesis, we explore uncertainty quantification (UQ) approaches and propose an uncertainty estimation system for general medical imaging classification tasks. In experiments, we apply the UQ system for three medical imaging databases, including All-IDB2 (an acute lymphoblastic leukemia database), SARS-CoV2 (a coronavirus disease 2019 database) and BreaKHis (a breast cancer histopathological imaging database). Besides, we discuss how to apply UQ methods to obtain more information on the database and its modelling. We can capture the samples with the  uncertainty values and predict the most uncertain category. We also discover that we can receive more accurate results than initial modelling results by removing a percentage of data with higher uncertainty results. In summary, we find great potential for UQ research on complex medical classification tasks and consider it to become probably one of the future's essential research directions.