An Automated Classifier of Harmful Brain Activities for Clinical Usage Based on a Vision-Inspired Pre-trained Framework

Yulin Sun, Xiaopeng Si, Runnan He, Xiao Hu, Peter Smielewski, Wenlong Wang, Xiaoguang Tong, Wei Yue, Meijun Pang, Kuo Zhang, Xizi Song, Dong Ming, Xiuyun Liu

Project

• ​Submitted to npj Digital Medicine.

Deep learning-based classification and segmentation of interictal epileptiform discharges using multichannel electroencephalography

Yulin Sun, Min Guan, Xun Chen, Fengling Feng, Runnan He, Lian Huang, Xiaoguang Tong, Huan Zhou, Xiuyun Liu, Dong Ming

Project

• This study was undertaken to develop a deep learning framework that can classify and segment interictal epileptiform discharges (IEDs) in multichannel electroencephalographic (EEG) recordings with high accuracy, preserving both spatial information and interchannel interactions.

Multi-task transformer network for subject-independent iEEG seizure detection

Yulin Sun, Longlong Cheng, Xiaopeng Si, Runnan He, Tania Pereira, Meijun Pang, Kuo Zhang, Xin Song, Dong Ming, Xiuyun Liu

Project

• Proposed a subject-independent intracranial EEG (iEEG) seizure detection model integrating channel-wise mixup and Transformer, enhanced generalization via multi-task learning, achieving AUCs of 0.97 and 0.95 on two public datasets, significantly outperforming existing methods

Continuous seizure detection based on transformer and long-term iEEG

Yulin Sun, Weipeng Jin, Xiaopeng Si, Xingjian Zhang, Jiale Cao, Le Wang, Shaoya Yin, Dong Ming

Project

• Developed an end-to-end seizure detection model using Transformer and convolutional layers to analyze continuous intracranial EEG (iEEG), achieving 97.5% and 98.1% sensitivity on two datasets with low false detection rates, while enhancing explainability via channel-specific attention mechanisms

​Limb Motion Identification Technology Based on Electrocorticographic (ECoG) Signals​​ (基于皮层脑电信号的肢体运动辨识技术赛项)

Yulin Sun, Xin Song, Leshui Dong, Xinyang Liu, Wenlong Wang

Project

• In this competition task, the pig undergoes passive motion on a treadmill. A 32-channel semi-invasive EEG acquisition system is used to collect electrocorticographic (ECoG) signals from the pig’s left motor cortex. Through soft timing synchronization, a motion capture system simultaneously records joint motion signals from 8 channels on the pig’s right forelimb and hindlimb. Participants are required to decode the angles of these 8 joints based on the EEG signals.
• National Champion (Grand Prize)​​, 2025.
• National Runner-up (First Prize), 2024.

Metaverse-Oriented Off-Body Generation and Detection Technology for EEG Signals​​ (面向元宇宙的脑电信号离体生成和检测技术赛项)

Yulin Sun, Xin Song, Xinyang Liu, Leshui Dong, Wenlong Wang

Project

• Based on electroencephalogram (EEG) signals, individual identification technology has made significant progress and has become a cutting-edge research hotspot. This competition task involves designing a security system capable of discriminating the authenticity of resting-state EEG data and identifying the corresponding subject. Participants are required to generate simulated data online to perform real-time hacking attacks on the competition system, ensuring that the system recognizes them as the correct subject while not detecting the data as simulated.
• National Runner-up (First Prize), 2025.
• National Runner-up (First Prize), 2024.

HMS - Harmful Brain Activity Classification

Yulin Sun

Code

• The goal of this competition is to detect and classify seizures and other types of harmful brain activity. Participants develop models trained on electroencephalography (EEG) signals recorded from critically ill hospital patients. This work may help significantly improve electroencephalography pattern classification accuracy, unlocking transformative benefits for neurocritical care, epilepsy, and drug development.
• Ranked 42nd (top 1.5%) among 2767 teams from 113 countries. silver medalist.

Emotional Brain Computer Interface​​ (情绪脑机接口)

Yulin Sun, Dong Huang, Zhuobin Yang, Shudi Huang, He Huang

Project

• Affective computing technologies, also known as emotional brain-computer interfaces, which enable machines to understand human emotions, are emerging as research hotspots in fields such as human-computer interaction, mental health, and artificial intelligence. This competition provides participants with a set of EEG data from 80 subjects whose emotional states are known. Participants are required to develop an EEG computational model capable of cross-individual emotion recognition. This model must perform real-time emotion recognition on EEG data from another set of subjects, and the competition results will be determined based on the accuracy of this emotion recognition
• National Champion (First Prize), 2022​​.

Parkinson’s Freezing of Gait Prediction

Yulin Sun

• The goal of this competition is to detect freezing of gait (FOG), a debilitating symptom that afflicts many people with Parkinson’s disease. Participants develop machine learning models trained on data collected from a wearable 3D lower-back sensor. By doing so, their work will help researchers gain a better understanding of when and why FOG episodes occur.
• Ranked 51st (top 4%) among 1361 teams from 83 countries. silver medalist.

Ubiquant Market Prediction

Yulin Sun

• In this competition, participants build models to forecast investment return rates. They train and test their algorithms on historical price data, aiming to solve this real-world data science problem with the highest possible accuracy.
• Ranked 23rd (top 1%) among 2893 teams from 101 countries. silver medalist.