Multi-modal AI approach for Early Tuberculosis Detection by combining symptom, imaging, and clinical data

Area of Science:

• Medical Diagnostics
• Artificial Intelligence in Healthcare
• Public Health

Background:

• Tuberculosis (TB) remains a significant global health challenge, causing millions of deaths annually, particularly in low- and middle-income countries.
• Traditional TB diagnostic methods like sputum microscopy and chest X-rays have limitations including low sensitivity, delayed results, and subjective interpretation.
• Existing artificial intelligence (AI) models often rely on single data types, limiting their ability to capture the complexity of TB diagnosis.

Purpose of the Study:

• To develop and evaluate a novel multi-modal AI system for earlier and more accurate tuberculosis detection.
• To integrate diverse patient data including symptoms, chest X-ray images, and clinical factors for improved diagnostic performance.
• To overcome the limitations of uni-modal AI approaches in diagnosing complex diseases like TB.

Main Methods:

• A multi-modal AI system was developed, integrating data from structured symptom surveys, digital chest X-rays, and electronic health records.
• Natural Language Processing (NLP) was employed to analyze symptom data.
• Convolutional Neural Networks (CNNs) were utilized for the analysis of chest X-ray images.

Main Results:

• The multi-modal AI model achieved high diagnostic performance, with 92.4% accuracy, 91.1% sensitivity, 93.2% specificity, and an AUC of 0.95.
• The integrated approach significantly outperformed uni-modal models in TB diagnosis.
• Combining imaging, clinical, and symptom data enhanced diagnostic reliability and reduced the rate of false positives.

Conclusions:

• Multi-modal AI integrating imaging, clinical, and symptom data offers improved accuracy and reliability for early TB diagnosis.
• This technology has the potential to facilitate early disease management, reduce global TB morbidity and mortality.
• Future research will focus on expanding datasets and real-world clinical validation for scalable and adaptable TB testing solutions.