| 要旨 |
Deep learning, as one of the machine learning technologies, has been extensively adopted in the field of neuroimaging due to its robust performances and time-saving capabilities compared to traditional strategies. This is largely attributed to the availability of massive accumulated datasets, the development of fascinating architectures, and enhanced computing resources. Over the past decade, deep learning applications in neuroimaging have primarily focused on brain segmentation and parcellation of brain structures. Conventional methods often entail lengthy data analysis processes, relying on statistical approaches or even manual methods that are both time-consuming and prone to errors associated with human hand-drawing. Although recent studies have demonstrated that deep learning models offer faster and more accurate segmentation and parcellation compared to traditional methods, the performance of deep learning models depends on the data used for training. This necessitates a target-oriented approach to training. In this seminar, I will introduce two cases of target-oriented model training.
|
