Title: Learning with Limited Supervision in Medical Imaging

Presenter: Giacomo Boracchi
Politecnico di Milano, DEIB - Dipartimento di Elettronica, Informazione e Bioingegneria
Milano, Italy
https://boracchi.faculty.polimi.it/

Abstract:
Deep neural networks have a large potential in biomedical imaging applications. In contrast with the natural imaging domain, where there are many large and publicly available datasets rich in annotations, labels are rare and costly to gather in the medical domain. This implicit data-scarcity problem is further exacerbated by many factors that make images very particular (like the type of treatment applied to a tissue sample or a pathological condition in a patient), which limits the effectiveness of transfer learning. Not surprisingly, a vivid research line goes toward designing data augmentation or data generation schemes to circumvent data scarcity. 
In this talk, I will overview our ongoing research activities to successfully train segmentation networks by generating realistic annotated samples or training with sparse annotations, which are much faster to obtain. Applications are developed in collaboration with a medical research institute (Mario Negri) and a company (Ikonisys).

References:
An expert-driven data generation pipeline for histological images
Roberto Basla, Loris Giulivi, Luca Magri, Giacomo Boracchi,
International Symposium on Biomedical Imaging, ISBI 2024



Short Bio

Giacomo Boracchi is Associate Professor of Computer Engineering at Dipartimento di Elettronica, Informazione e Bioingegneria of the Politecnico di Milano (DEIB). Giacomo has received a Ph.D. in information technology (DEIB, 2008), and an MSc degree in Mathematics (Universitá Statale di Milano, 2004). His primary research interests concern image processing and machine learning, particularly image restoration and analysis, change/anomaly detection, domain adaptation, and learning in nonstationary environments. He has been/currently is the advisor of 12 Ph.D. students, regularly teaching deep learning and image processing courses and giving tutorials at major conferences (ICASSP, ICIP, ICPR, IJCNN).
 
He is currently leading industrial research projects with Huawei, STMicroelectronics, Gilardoni Raggi X, and Cisco, among others. He is the author of more than 100 papers in international conferences and journals, and in 2018 - 2024 he has served as an Associate Editor for the IEEE Transactions on Image Processing. In 2015 he received the IBM Faculty Award; in 2016 the IEEE Transactions on Neural Networks and Learning Systems Outstanding Paper Award; in 2017, the Nokia Visiting Professor Scholarship; in 2021 and 2024, an nVidia Applied Research Grant.