| 日時(Date) |
2025年02月04日 (火) / Feb. 4th, 2025. 4限 (15:10--16:40) / 4th period (15:10--16:40) |
|---|---|
| 場所(Location) | エーアイ大講義室（L1）+ ONLINE |
| 司会(Chair) | Yoshito Otake |
| 講演者(Presenter) | Miguel A. González Ballester, Visiting professor, Barcelona Centre for New Medical Technologies (BCN Medtech), Department of Engineering, Universitat Pompeu Fabra, Barcelona, Spain. |
| 題目(Title) | Interpretable deep learning for medical image analysis |
| 概要(Abstract) | Deep learning has had a profound effect on the state of the art of computer vision, leading to huge advances that were until recently unthinkable. The impressive performance of these methods is partly due to the development of very efficient methods to solve highly overparameterized neural networks. Conversely, this leads to the problem of interpretability of these models, and this is a key aspect in applications such as medical image analysis. In this talk, we will see several works performed at the BCN Medtech team in Universitat Pompeu Fabra, Barcelona, Spain. Then, we will focus in particular in three recent works. The first is a method for attribute-based regularization of variational autoencoders, for interpretable deep learning in the context of cardiology. The second application is on the study of the evolution of lung cancer, and we will present work on nodule re-identification and also a probabilistic neural network framework for the prediction of the evolution of lung nodules, which also incorporates uncertainty quantification. Finally, the third work focuses on 3D model generation of the spine from 2D images. |
| 講演言語(Language) | 英語 /English |
| 講演者紹介(Introduction of Lecturer) |