Taiwan – May 2025 – HealthConn Biomedical (6665-TW), a leading force in the health services ecosystem, announced that in collaboration with research partners, it has developed an AI-powered model capable of predicting rifampicin resistance in Mycobacterium tuberculosis complex (MTBC) using mass spectrometry data and machine learning techniques. The study was selected for presentation at the European Society of Clinical Microbiology and Infectious Diseases (ESCMID) Global 2025 Congress, showcasing Taiwan’s growing international influence in AI-driven antimicrobial resistance (AMR) research.
The joint study, titled “Predicting rifampicin resistance in Mycobacterium tuberculosis complex with a machine learning-based MALDI-TOF MS approach”, was co-developed with national research institutions and clinical partners, including Reveres Medical Laboratory. It was featured in the dedicated session AI tools empowering AMR prediction in WGS and MALDI-TOF, one of the few deep-dive discussions at the conference focused on AI-powered AMR prediction using mass spectrometry. The research highlights the potential to provide clinicians with rapid, actionable insights for tuberculosis treatment.
“By integrating AI and mass spectrometry, we aim to equip clinicians with reliable resistance predictions at an early stage of treatment,” said Huang Ying-Shih, Chairman of HealthConn Biomedical. “This will accelerate decision-making, reduce drug resistance risks, and ultimately improve patient outcomes. Being invited to present at ESCMID marks a significant milestone for Taiwan’s AI medical technology on the global stage, and we look forward to advancing toward clinical implementation.”
AI + Mass Spectrometry: Reducing Diagnostic Time by Over Three Weeks
According to Taiwan CDC, 6,222 new TB cases were confirmed in 2024, underscoring the continuing public health burden of tuberculosis and the challenges posed by drug-resistant strains. Traditional drug susceptibility testing (DST) still requires at least 28 days, delaying treatment decisions.
The HealthConn-led team developed an AI prediction model using MALDI-TOF MS data combined with machine learning algorithms. Once MTBC cultures are prepared and mass spectrometry completed, the AI model can generate predictions within one minute, providing results more than three weeks earlier than conventional DST. The study achieved an AUC of 98.45% and an accuracy rate of 91.96%, demonstrating strong clinical potential.
The model was trained using 188 isolates with cross-validation and optimized via the Extreme Boosting algorithm. It leverages Bruker MS systems with full signal retention, enabling detailed learning of subtle resistance variations. Signal normalization across operators and batches further ensured robustness for future clinical deployment. The team also plans to evaluate regulatory pathways for Software as a Medical Device (SaMD), moving toward real-world implementation.
Contributing to the WHO 2035 End TB Goal
This technology aligns with the World Health Organization’s End TB Strategy for 2035, offering an innovative tool for AMR management. HealthConn’s invitation to present at ESCMID Global 2025 highlights Taiwan’s maturity and research capacity in smart healthcare and resistance diagnostics, laying a strong foundation for international collaboration and clinical adoption.