2026 Projects
2026 Project: Climate and Environmental Health
LEINA: Low-Compute Vision–Language Model for Equitable Tuberculosis Detection in Indonesian Chest X-Rays
Project Team
- Vanya Valindria (lead), Monash University Indonesia
- Reyhan Eddy Yunus, RSCM Hospital / Universitas Indonesia
- Prasandhya Astagiri Yusuf, Head of Medical Big Data Center (BDC), IMERI FK UI
Location
Indonesia
Executive Summary
Tuberculosis (TB) remains one of Indonesia’s most urgent health challenges, with over 800,000 new cases annually. Diagnostic inequity persists across Indonesia’s primary healthcare network (Puskesmas community health centers) where radiologists and modern imaging infrastructure are limited. Climate-related factors—air pollution, urban overcrowding, and environmental stress - further exacerbate respiratory disease severity and significantly alter chest imaging appearances.
Most existing AI models for chest X-ray–based TB detection are developed using clean datasets from high-income countries. These models often underperform when applied to Indonesian patients due to environmental, demographic, and imaging-quality differences. Despite a population exceeding 270 million, Indonesia has no widely deployed, locally trained AI models for chest X-ray screening - creating a substantial equity gap.
LEINA aims to close this gap through a low-compute, data-efficient Vision–Language Model (VLM) trained using paired Indonesian chest X-rays and radiology reports. Designed for climate-impacted and resource-constrained healthcare environments, LEINA will:
Leverage real-world data reflecting pollution-related variations in lung imaging.
Align visual features with clinical text for better interpretability and generalizability; and
Enable deployment without GPU requirements, as a lightweight web-based screening tool suitable for Puskesmas and district hospitals.
Expected Outcomes
- Curated Indonesian chest X-ray dataset with paired annotations / radiology report
- LEINA model performance benchmarked against global datasets to quantify geographic bias
- Prototype web-based screening application enabling equitable TB triage in low-resource settings, ready to integrate with Indonesia’s national TB initiatives
As one of the first MICCAI-aligned medical imaging AI efforts from Indonesia, LEINA will advance climate-conscious, locally relevant, and affordable AI innovation - supporting health equity across Indonesia and the broader Southeast Asian region.