Jintai Chen

Jintai Chen

Postdoc at CS, UIUC; PhD from CS, ZJU. Passionate about AI for Healthcare.

Hey, I am Jintai Chen, a postdoctoral researcher at the University of Illinois at Urbana-Champaign, Illinois, USA.

Before that, I graduated with a Ph.D. degree from the College of Computer Science and Technology, Zhejiang University. I am broadly fascinated by the intersection of AI and healthcare/bioscience, especially the field of AI for healthcare (AI4H), with a specific emphasis on designing machine learning algorithms informed by medical knowledge and utilizing machine learning for uncovering insights in medical data. My research extensively delves into various facets including AI for medical imaging analysis (AI4MIA), AI for electrocardiograms (AI4ECG), AI for EHR table analysis (AI4Table), and AI for clinical trials (AI4CT). My methodological interests revolve around deep tabular models, generative AI for digital twin creation, and exploring deep learning approaches tailored for scenarios with limited data availability.

I am an amateur snooker player and a big fan of Ronnie O’Sullivan, and I also enjoy painting in my leisure time.

🔥 News

📄 Selected Publications

(*: Equal contribution; $\dagger$: Corresponding author(s))

Nature Communications
sym

TL;DR: Congenital heart disease is the most common category among congenital abnormalities, with an incidence rate approaching 1$\%$. Previously, ECGs were considered to have limited effectiveness in diagnosing congenital heart disease. AI demonstrated the value of ECGs in diagnosing congenital heart disease, which surpasses our previous cognition. While techniques like echocardiography and cardiac MRI are currently utilized for precise diagnosis, the cost-effectiveness and non-invasiveness of ECGs continue to harbor substantial potential for precise large-scale population screening and benefiting low-resourced regions.

Congenital Heart Disease Detection by Pediatric Electrocardiogram Based Deep Learning Integrated with Human Concepts [AI4H, AI4ECG, AI4Table] [Code], Jintai Chen$^*$, Shuai Huang$^*$, Ying Zhang$^*$, Qing Chang$^*$, Yixiao Zhang, Dantong Li, Jia Qiu, Lianting Hu, Xiaoting Peng, Yunmei Du, Yunfei Gao, Danny Chen, Abdelouahab Bellou$^\dagger$, Jian Wu$^\dagger$, Huiying Liang$^\dagger$, Nature Communications, 2024

IJCAI 2021
sym
  • TL;DR: Traditional ECG devices can only offer electrocardiograms from a limited number of angles, constrained by electrode positioning. Our Electrocardio Panorama System breaks this barrier, allowing users to effortlessly observe ECG signals from any angle in real-time, based on their queries.
  • Academic Impact: The benefits of our work are manifold: (i) panoramic observations of ECG signals; (ii) a unified representation of ECG signals captured by different ECG devices; (iii) Waveform-aligned Mixup for synthesizing new ECG cases (e.g., for data augmentation); (iv) reconstruction of corrupted ECG views; and (v) exploration of ECG theory.
  • New Data Annotations: We provided ECG wave segmentation annotations for Tianchi ECG dataset and PTB dataset.

Electrocardio panorama: Synthesizing new ECG views with self-supervision [AI4H, AI4ECG] [Code and Data], Jintai Chen$^*$, Xiangshang Zheng$^*$, Hongyun Yu$^*$, Danny Z. Chen, Jian Wu$^\dagger$, International Joint Conference on Artificial Intelligence (IJCAI), 2021

ICML 2021
sym

TL;DR: Neurons in the nervous system transmit signals by releasing different neurotransmitters that match different receptors. Motivated by the concepts of competitive neural networks, prototype learning, hierarchical clustering algorithms, and capsule neural networks, we introduce a novel neural network architecture. This neural network is constructed by neurons capable of generating “transmitters” to send semantic information to other neurons and possessing receptors to receive specific types of “transmitters” from other neurons. By generating “transmitters” to convey semantic information and binding them to specific receptors in the subsequent layer, our approach achieves transparent semantic feature parsing, part-to-whole semantic integration, unsupervised semantics understanding, and object relationship digging.

A receptor skeleton for capsule neural networks, Jintai Chen, Hongyun Yu, Chengde Qian, Danny Z. Chen, Jian Wu$^\dagger$, International Conference on Machine Learning (ICML), 2021

ICLR 2024 SpotLight
sym

TL;DR: Tabular data exhibits diversity in both feature and target definitions. How can we achieve transferability across such heterogeneity? We propose an approach to empower language models as a robust deep tabular prediction model. By training the language model to comprehend precise numeric values, our approach gains the capability to leverage tabular data from other domains to enhance predictions on EHR tables, where data availability is often limited.

Making Pre-trained Language Models Great on Tabular Prediction [AI4Table] [Code and Data], Jiahuan Yan, Bo Zheng, Hongxia Xu, Yiheng Zhu, Danny Chen, Jimeng Sun, Jian Wu$^\dagger$, Jintai Chen$^\dagger$, ICLR (SpotLight), 2024

MICCAI 2020 Oral
sym

TL;DR: This study transforms unstructured hand radiography images into a structured semantics represented as a table /graph, utilizing clinical prior information (the TW3 approach used in clinical practice). We then use a GNN to process such structured data, leading to impressive and interpretable bone age assessments. It’s noteworthy that many medical images are semi-structured data, and this paper introduces a potentially interpretable and efficient approach for processing such semi-structure.

Doctor imitator: Hand-radiography-based bone age assessment by imitating scoring methods [AI4H, AI4MIA], Jintai Chen, Bohan Yu, Biwen Lei, Ruiwei Feng, Danny Z. Chen, and Jian Wu$^\dagger$, International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI, Oral), 2020

🎖 Awards

  • 2023.04, Excellent Doctoral Graduates of Zhejiang Province, China (Top 1%)
  • 2023.04, Excellent Doctoral Graduates of Zhejiang University (Top 1%)
  • 2022.10, Huawei Fundamental Research Scholarship (Top 3%)
  • 2021.10, Tencent Doctoral Scholarship (Top 1%)
  • 2021.10, National Scholarship of China (Top 1%)
  • 2020.10, Outstanding Doctoral Student Scholarship (Top 3%)
  • 2019.10, Doctoral Freshman Scholarship (Top 3%)
  • 2016.10, Chinese Bank Scholarship (Undergraduate) (Top 1%)
  • 2015.10, National Scholarship (Undergraduate) (Top 1%)

💬 Talks

  • 2022.11, How to Excel in AI-for-healthcare Researches, @ Shanghai University
  • 2022.10, ECG Signal Processing and Synthesis for Computer-Aided Heart Disease Diagnosis, @ Carnegie Mellon University
  • 2022.10, ECG Synthesis for New View and New Data, @ Shanghai AI Lab
  • 2022.09, Part-Hierarchy Learning, @ ByteDance
  • 2022.06, Supervised Tabular Learning, @ UberAI
  • 2021.10, Domain Mixup for Distant Transfer Learning, @ Shanghai Jiaotong University

🏫 Teaching

  • Fall 2023, Frontiers of Medical Artificial Intelligence (lecture slice preparation, teaching assistant)

🔎 Reviews:

  • Review for Conferences: ICLR, ICML, AAAI, IJCAI, CVPR, MICCAI, ISBI, NeurIPS, ICCV, ECCV, EMNLP.
  • Review for Journals: TPAMI, TNNLS, TCBB, JBHI, Frontiers in Public Health, JBSM, TCDS, Frontiers in Genetics, Scienstific Report

🎒 Visiting

  • 2021.06 - 2021.09, Medical Big Data Center, Guangdong Academy of Medical Sciences, had the honor of working with Prof. Huiying Liang, Shuai Huang, and Dantong Li.
This homepage is visited times