AI Offers a New Approach to Tuberculosis Screening
Tuberculosis continues to affect millions of people worldwide, with Indonesia ranking among the countries carrying one of the highest disease burdens. The challenge is particularly severe inside correctional institutions, where overcrowding, poor ventilation, and limited healthcare resources create conditions that facilitate disease transmission.
Traditional TB screening generally depends on chest X-rays and laboratory examinations. While these methods remain the clinical standard, they require specialized equipment, trained personnel, and considerable operational costs. Such limitations often delay diagnosis, especially in correctional facilities and resource-limited healthcare settings.
Recognizing these barriers, the research team explored whether simple physiological indicators routinely collected during basic health examinations could be combined with artificial intelligence to identify individuals who may require further tuberculosis testing. Their innovation became the foundation of TB-Guard, a HealthTech startup concept designed to support faster, more affordable, and more accessible TB screening.
Simple Health Data Combined with Machine Learning
The study employed a quantitative observational design involving 70 female inmates at a Class IIB Correctional Institution in Padang, Indonesia.
Researchers collected six routine physiological measurements:
- Body temperature
- Oxygen saturation (SpO₂)
- Pulse rate
- Systolic blood pressure
- Diastolic blood pressure
- Respiratory rate
Chest radiography served as the clinical reference for determining TB risk, while the physiological data were processed using a Gradient Boosting Machine (GBM) algorithm. The model was developed through data cleaning, feature selection, and validation before being incorporated into the TB-Guard prototype as a clinical decision-support system.
Strong Predictive Performance
The AI model demonstrated excellent predictive capability during testing.
Key performance indicators included:
- Testing accuracy: 92.86%
- Precision: 83.33%
- Recall: 83.33%
- F1-score: 83.33%
- AUC: 0.947
The confusion matrix also showed that the model correctly identified nearly all high-risk individuals while producing no false-positive classifications. According to the researchers, the high AUC indicates that the system has excellent ability to distinguish between individuals at risk of tuberculosis and those who are not.
Feature importance analysis further revealed that several routine physiological measurements contributed substantially to prediction accuracy, suggesting that these vital signs may function as practical biomarkers for preliminary tuberculosis screening.
Beyond Artificial Intelligence: Building a HealthTech Startup
Unlike many machine learning studies that focus solely on algorithm performance, this research extends its contribution into digital entrepreneurship.
TB-Guard was designed not merely as a predictive model but as a scalable HealthTech startup capable of supporting healthcare professionals during early tuberculosis screening. Rather than replacing medical diagnosis, the platform helps prioritize individuals who should undergo confirmatory examinations such as chest radiography or laboratory testing.
By relying on physiological data already collected during routine health assessments, TB-Guard could reduce unnecessary diagnostic procedures while improving the allocation of healthcare resources. The researchers also suggest that the platform could be adapted beyond correctional institutions for use in:
- Primary healthcare centers
- Community clinics
- Rural healthcare facilities
- Population-based screening programs
- Other resource-constrained healthcare environments
This broader applicability increases both the public health value and commercialization potential of the technology.
Potential Benefits for Public Health and Digital Innovation
The findings demonstrate how artificial intelligence can simultaneously address healthcare challenges and stimulate technology-based entrepreneurship.
Earlier identification of individuals at high risk of tuberculosis could improve case detection, accelerate treatment initiation, and strengthen Indonesia's national TB control strategy. From an economic perspective, a scalable HealthTech platform may also reduce healthcare costs associated with delayed diagnosis while supporting the country's ongoing digital health transformation.
The researchers acknowledge that the current prototype still requires further validation before widespread implementation. Future studies should involve larger and more diverse populations, incorporate additional clinical variables, compare GBM with other machine learning algorithms, and evaluate regulatory readiness, usability, and business feasibility before commercialization.
Researchers Emphasize Practical Healthcare Innovation
The authors from Universitas Baiturrahmah emphasize that the significance of TB-Guard extends beyond predictive accuracy. Their work demonstrates how academic research can be translated into a practical digital healthcare solution capable of supporting early tuberculosis detection while creating opportunities for sustainable HealthTech entrepreneurship.
Their findings indicate that combining routinely measured physiological parameters with artificial intelligence offers a rapid, objective, and cost-effective screening approach that could complement existing tuberculosis control programs, particularly in settings where diagnostic resources remain limited.
Author Profile
Fauzyah Aprillia is a researcher and lecturer at the Faculty of Vocational Studies, Universitas Baiturrahmah, specializing in public health, digital health innovation, predictive analytics, and artificial intelligence applications in healthcare. This study was conducted in collaboration with Oktavia Puspita Sari and Noviardi Prima Putra, also from the Faculty of Vocational Studies, Universitas Baiturrahmah, together with Fauziah Putri Ramadani, a Radiology student at Universitas Baiturrahmah. The multidisciplinary team focuses on developing evidence-based digital health technologies that improve disease prevention and early detection.
Source
Article Title: Development of the TB-Guard HealthTech Startup: A Gradient Boosting Machine-Based Tuberculosis Risk Prediction Model for Correctional Inmate Screening
Journal: Indonesian Journal of Entrepreneurship & Startups (IJES), Volume 4, Number 2 (2026)
DOI: https://doi.org/10.55927/ijes.v4i2.17071
Official Journal: https://journal.formosapublisher.org/index.php/ijes
0 Komentar