Abstract : The minisymposium will explore the various ways in which artificial intelligence is being used in the field of healthcare and medicine, with a focus on the use of privacy-preserving machine learning. The symposium will feature presentations from experts in the field, who will discuss the latest developments and trends in AI for healthcare and medicine, including the use of federated learning and data collaboration for training machine learning models on decentralized data. The symposium will provide attendees with a comprehensive overview of the current state of AI in healthcare and medicine, and will offer insights into the potential future developments in this rapidly evolving
Organizer(s) : Tetsuya Sakurai, Akira Imakura, Weichung Wang, Li-Chen Fu
[02894] Data Collaboration Cox Proportional Hazards Model for Privacy-preserving Survival Analysis
Format : Talk at Waseda University
Author(s) :
Akira Imakura (University of Tsukuba)
Ryoya Tsunoda (University of Tsukuba Hospital)
Rina Kagawa (University of Tsukuba)
Kunihiro Yamagata (University of Tsukuba)
Tetsuya Sakurai (University of Tsukuba)
Abstract : In recent years, privacy-preserving machine learning for datasets held by multiple organizations in a distributed manner has been attracted attention. In this study, we focus on privacy-preserving survival analysis for datasets held by multiple medical institutions and propose a data collaboration technique that shares dimensionality-reduced intermediate representations instead of raw data. The proposed DC-COX can calculate the contribution of each feature to survival time and the corresponding p-value. Numerical experiments verify the effectiveness of DC-COX.
[05307] AI-Enhanced Medical Imaging Analysis: Advancing Precision Treatment for NSCLC Brain Metastases
Format : Talk at Waseda University
Author(s) :
Cheyu Hsu (National Taiwan University Hospital)
Abstract : In this talk, we discuss the innovative use of radiomics and deep learning in AI-enhanced medical imaging analysis for managing NSCLC brain metastases. Through automated segmentation, we streamline diagnosis and treatment planning. The presentation delves into predicting local recurrence after radiosurgery, detecting EGFR mutations, and evaluating distant metastases or brain metastases velocity in radiosurgery-treated patients. Our focus on AI-driven methodologies fosters tailored, precision treatment strategies, ultimately enhancing patient outcomes for those with NSCLC brain metastases.
[05001] Explainability and Fairness of Distributed Data Analysis
Format : Talk at Waseda University
Author(s) :
Anna Bogdanova (University of Tsukuba)
Tetsuya Sakurai (University of Tsukuba)
Akira Imakura (University of Tsukuba)
Abstract : Ensuring fairness and transparency in machine learning models is critical for their ethical application in the medical field. With the increasing use of distributed machine learning to protect patient privacy, there is a growing need to address the challenges of explainability and fairness in medical data analysis. Machine learning models trained on horizontally or vertically partitioned medical data may present difficulties for explainability, as different participants may have a biased view of the background data or a partial view of the feature space, leading to inconsistencies in the explanations obtained. To address these issues, this paper proposes an Explainable Data Collaboration Framework that combines a model-agnostic additive feature attribution algorithm (KernelSHAP) with a privacy-preserving distributed machine learning method called Data Collaboration. The framework offers three algorithms for various scenarios of explainability in medical data collaboration, which were tested on open-access medical datasets. In addition, we show that our proposed framework can be combined with fairness-sensitive data representation techniques to eliminate data biases at the local level.
[05362] Mitigating Non-IID Data Challenges in Federated Learning for Healthcare Applications
Format : Talk at Waseda University
Author(s) :
Fan Zhang (University of Cambridge)
Abstract : Federated Learning has emerged as a promising technique for healthcare applications, enabling collaboration among different healthcare institutions without sharing sensitive data. Data in each healthcare institution usually has a unique distribution, leading to non-IID (independent and identically distributed) data that can impact model performance and convergence in Federated Learning. In this talk, we will present the findings of non-IID challenges in Federated Learning and recommendations for the strategies we evaluated to mitigate these challenges.
[03888] Causal inference and machine learning on distributed data
Format : Talk at Waseda University
Author(s) :
Yuji Kawamata (Center for Artificial Intelligence Research, University of Tsukuba)
Ryoki Motai (Graduate School of Science and Technology, University of Tsukuba)
Yukihiko Okada (Center for Artificial Intelligence Research, University of Tsukuba)
Akira Imakura (Center for Artificial Intelligence Research, University of Tsukuba)
Tetsuya Sakurai (Center for Artificial Intelligence Research, University of Tsukuba)
Abstract : Utilizing distributed data allows for more reliable estimation of conditional average treatment effects. However, it is difficult to share data owing to privacy concerns. To address this issue, we proposed Data Collaboration Double Machine Learning (DC-DML), which can address horizontally and vertically distributed data and provide point and interval estimation. In experiments using synthetic data, we found that DC-DML could lead to more accurate estimation results than when using distributed data individually.
[05313] Precision Preventive Medicine in Sub-Healthy Population
Format : Talk at Waseda University
Author(s) :
Han-Mo Chiu (National Taiwan University Hospital)
Hung-Ju Lin (National Taiwan University Hospital)
Abstract : Preventing the onset or progression of non-communicable diseases in sub-health population tremendously impact the population health and the related cost and both primary and secondary prevention play pivotal roles in this aspect. The State-of-the-art digital health and artificial intelligence technologies have been applied widely in the healthcare sector, and are anticipated to play a more proactive role in preventive medicine in terms of risk stratification, adopting clinical, genomic, or metagenomic information, and leveraging lifestyle modification.
[03660] Medical AI, Biosensors and Privacy
Format : Talk at Waseda University
Author(s) :
Takeshi Kimura (University of Tsukuba)
Abstract : As Medical AI is expected to improve health care, there is also a concern regarding collecting personal data via AI devices, including biosensors. Advanced biosensors could read and collect inner physiological, emotional, and sensitive conditions. However, patients cannot control their private information collected with biosensors and, once they become data belonging to a hospital or data collection company, cannot have access to their private information. The associated ethical issues are examined.
[05312] HeaortaNet: AI for Quantifying Heart Structures on Non-Contrast CT Images
Format : Talk at Waseda University
Author(s) :
Wen-Jeng Lee (Department of Medical Imaging, National Taiwan University Hospital)
Abstract : HeaortaNet is an AI model developed by TW-CVAI for the segmentation of pericardium/aorta and calcium/fat quantification on non-contrast chest CT images. This talk introduces the technology, benefits, and real-world applications of CT data from Taiwan's National Health Insurance Administration. Our research aims to enhance patient care by providing an effective tool for identifying and measuring heart disease, ultimately leading to better treatment and outcomes.
