Skip to content
Zheng Rong JIA

Zheng Rong JIA

Medical AI & Predictive Modeling · Independent Researcher

Education

B.Sc. Software Engineering

Macau University of Science and Technology

Graduated Aug 2025

Selected Publications
  • Deep Learning for Stroke Mortality Prediction in eICU: A Dual-Tower Transformer Framework

    Zheng Rong JIA*, Kwong-Cheong Wong*

    CCAI 2026 (EI-indexed)

Latest Updates
  • 2026 Finalizing the open-source release of the DT-Transformer framework for reproducible clinical AI research.
  • 2026 Paper Accepted at CCAI 2026: "Deep Learning for Stroke Mortality Prediction in eICU: A Dual-Tower Transformer Framework" (Upcoming: May 2026).
  • 2025 Earned B.Sc. in Software Engineering from Macau University of Science and Technology.
Research Interests
Clinical Predictive Modeling

Transformer-based temporal modeling on longitudinal EHR data; early warning systems for ICU critical events with missing-data-aware architectures, building on prior work in stroke risk prediction (CCAI 2024).

Medical Image Analysis

Vision Transformer and hybrid CNN-ViT architectures for brain MRI segmentation; semi-supervised learning strategies to address annotation scarcity in radiology AI, with focus on multi-class lesion delineation.

Explainable AI for Healthcare

Faithfulness evaluation of attention-based saliency methods for deep radiology models; human-in-the-loop clinical validation frameworks to bridge the gap between model accuracy and clinician trust in high-stakes diagnosis.

Experience & Projects

Lead Researcher — Medical AI

Aug 2025 – Present

Independent · Clinical AI · Stroke Mortality Prediction · eICU

  • Designed a Dual-Tower Transformer achieving AUPRC of 0.6171 — a 14.4% improvement over neural baselines — on the multicenter eICU database (200k+ records)
  • Integrated an Adaptive Runtime Safeguard for out-of-distribution detection at inference time, ensuring clinical deployment safety
  • Paper accepted at CCAI 2026 (EI-indexed)

Time-Series Forecasting — Undergraduate Thesis

2023 – 2024

Python · PyTorch · LSTM · Pandas · NumPy

Built an LSTM-based sequential forecasting model with a full preprocessing pipeline (normalization, sliding-window construction, data-leakage-free splits), achieving 1.42% MAPE on the held-out test set.

MLE Intern — China Southern Power Grid AI

Jul – Aug 2024 · Guangzhou

Fine-tuned YOLOv8 for inspection robots; built and maintained a 60 GB multi-province industrial image dataset. Evaluation score: 100/100.