Tuberculosis (TB) remains one of the leading infectious causes of death globally, particularly in low- and middle-income countries such as Kenya, where early detection is still a major challenge. Existing TB diagnostic methods rely heavily on sputum-based testing and symptom screening, which are difficult to scale in decentralized and resource-limited settings. Many TB cases also remain undetected due to subclinical or mild symptoms. TB-SCAN Africa was developed as a cost-disruptive, AI-enabled, non-sputum screening and triage platform designed to improve early TB detection using multimodal artificial intelligence and low-cost reusable technologies suitable for community and primary healthcare settings.
The project aims to develop a multimodal AI framework capable of detecting tuberculosis using respiratory acoustics, physiological signals, structured clinical data, and chest radiographs. The goal is to create an affordable, scalable, and explainable screening system that improves sensitivity for early and asymptomatic TB detection while reducing dependence on laboratory-based sputum diagnostics. The project also seeks to support rapid same-visit TB risk assessment in low-resource environments.
Python: TensorFlow, scikit-learn, FastAPI, NumPy, pandas, OpenCV
AI & Deep Learning Models: DenseNet121, multimodal fusion models, explainable AI systems
Medical Imaging: Chest X-ray analysis with Grad-CAM explainability
Acoustic Analysis: Respiratory sound and cough analysis using AI
Mobile & Deployment Tools: Flutter, TensorFlow Lite, offline-first AI deployment
Data Visualization & Analysis: Matplotlib, Seaborn, ROC/AUC evaluation
Cloud & Edge AI Integration: Offline mobile inference and low-resource AI deployment
The system integrates respiratory acoustic recordings, physiological measurements, structured clinical data, and chest radiographs collected from community and primary healthcare environments.
Machine learning and deep learning models are developed for each modality, including acoustic analysis models, clinical prediction models, and chest X-ray classification systems using convolutional neural networks.
Explainability techniques such as Grad-CAM are incorporated to improve transparency and interpretability of AI-based TB predictions.
Outputs from individual AI models are combined into a unified predictive framework that generates a TB risk score and triage recommendation.
The platform undergoes iterative testing in community and clinical environments to evaluate usability, operational performance, sensitivity, and deployment feasibility.
Multimodal AI significantly improves sensitivity for TB screening compared to single-modality approaches.
AI-based chest radiograph analysis demonstrates strong capability for early TB detection.
Acoustic and physiological biomarkers show potential for non-invasive TB screening.
Offline-first deployment improves usability in low-resource and decentralized healthcare settings.
Explainable AI methods enhance interpretability and trust in TB risk predictions.
Reusable low-cost screening tools substantially reduce the per-screening cost.
Expand deployment of AI-enabled non-sputum TB screening systems in community healthcare settings.
Strengthen training programs for community health workers on AI-assisted screening workflows.
Integrate multimodal AI systems into national TB surveillance and triage programs.
Scale low-cost reusable diagnostic platforms to improve healthcare accessibility in LMICs.
Extend the framework with additional physiological and biosensor-based biomarkers for improved predictive performance.
TB-SCAN Africa demonstrates the potential of multimodal artificial intelligence to transform tuberculosis screening through affordable, scalable, and explainable non-sputum diagnostics. By integrating acoustic, physiological, clinical, and imaging data into a unified AI framework, the project supports early TB detection, improved triage, and enhanced healthcare accessibility in resource-limited settings. The work bridges AI, medical imaging, computational modeling, and global health to advance next-generation digital health solutions for TB control in Africa and other low-resource regions.