Accurate, Interpretable, and Privacy-preserving AI for Healthcare | Dr Xiaoxiao Li, Princeton

Описание: #GraphNeuralNetworks #FederatedLearning #ExplainableAI
We were delighted to have Dr Xiaoxiao Li from Princeton University as an invited speaker to the BASIRA 2021 #AI4Healthcare Seminars. She gave an impressive talk based on her recent #MICCAI2020 #IPMI2017 #ICML2021 #ICLR2021 publications, which cover cutting-edge research in artificial intelligence and healthcare including geometric deep learning for diagnosis, explainable AI and federated learning.

The recorded talk entitled “Towards Accurate, Interpretable, and Privacy-preserving AI for Healthcare” is also followed by a Q&A session.

Table of content:
0:00:00 Speaker introduction by Dr Islem Rekik
0:03:00 Graph Neural Networks for diagnosis
0:28:15 Explainable AI (XAI)
0:44:23 Privacy AI and federate learning
0:59:26 Q&A session
1:26:47 Photo taking :-)


Thank you for joining us and keep up the great work!

#XAI #DeepLearning #Diagnosis #DataPrivacy #Healthcare #Prognosis #Biomarkers #Neuroscience #NetworkNeuroscience



*Abstract*

Recent progress in artificial intelligence (AI) has advanced our ability to analyze biomedical data. However, significant obstacles, such as a lack of model transparency and insufficient training samples, have hindered applying AI to practical healthcare applications. To fill the gaps between AI and healthcare, improving the trustworthiness of AI is desired. In this talk, I will present the progress of developing the next generation of trustworthy AI systems for healthcare applications. Initially, we design accurate deep learning models for disease classification and investigate how to perform computational biomarker discovery from a model interpretation perspective [1,2]. Later, we improve the efficiency [3] and quantify [4] the uncertainly of Shapley value-based feature importance interpretation methods. Our further work explores multi-site learning under data non-IID assumption and proposes a privacy-preserving federated learning framework with built-in novel domain adaptation methods [5]. We validate our methods for characterizing Autism spectrum disorder (ASD) using functional magnetic resonance imaging (fMRI). Our results demonstrate that it is promising to utilize advanced deep learning models, novel model explanation methods, and federated learning to boost neuroimage analysis performance. Our approaches bring new hope for accelerating deep learning applications in estimating image-derived biomarkers and the healthcare field in general.

*Short bio*

Dr. Xiaoxiao Li is a postdoctoral research fellow in the Computer Science Department at Princeton University. She obtained her Ph.D. degree in Biomedical Engineering from Yale University in 2020. During her Ph.D., she was awarded the Advanced Graduate Leadership Fellowship. She received the honored Bachelor‘s Degree at Chu Kochen Honors College, Zhejiang University in 2015. Her research interest lies in the interdisciplinary field of artificial intelligence and biomedical analysis, aiming to improve the AI systems’ trustworthiness for healthcare applications. She has published nearly 30 papers at machine learning and medical imaging conferences and journals, such as MICCAI, IPMI, ICML, ICLR and Medical Image Analysis. Her work has received OHBM Merit Abstract Award, MIML Best Paper Award, and DART Best Paper Award.