Justice Ohene Amofa
Applied Bionformatician at Noguchi Memorial Institute for Medical Research
I am an Applied Bioinformatician and Machine Learning Engineer at the Noguchi Memorial Institute for Medical Research, University of Ghana. My work focuses on applying AI, bioinformatics, and computational methods to public health challenges, particularly in infectious disease surveillance, antimicrobial resistance, and genomics.
I have developed machine learning models for tuberculosis prediction and contributed to genomic analysis pipelines for pathogens such as cholera and tuberculosis. I am also actively involved in building open-source bioinformatics workflows and training students and researchers across Africa. My interests span AI for healthcare, large language models, and scalable data systems for scientific research.
Abstract
Machine learning is increasingly used in medical diagnostics, but many models remain difficult to reproduce, interpret, or deploy in real-world systems. This work presents a Python-based deep learning pipeline for tuberculosis detection using chest X-ray images combined with multivariate patient data, including age, comorbidities, socioeconomic indicators, and radiological features. The system is implemented using a modular Python ML stack and integrates transfer learning, classical machine learning models, and explainable AI techniques. Key components include: A deep learning pipeline built on a pre-trained ResNet50 backbone with custom classification layers Classical ML models (SVM, decision trees) for comparative evaluation A dataset of 4,800 chest X-ray images used for training and validation Performance metrics including Accuracy (~99%), Precision (100% TB class), F1-score (~0.9787), and AUC (~0.9989) Explainable AI methods to improve interpretability and clinical trust A Streamlit-based interface for interactive inference and deployment The architecture follows a standard Python ML workflow: Data preprocessing and normalization Feature integration across clinical and imaging modalities Model training, evaluation, and validation Lightweight deployment using Python web tooling Beyond the current system, the project is designed to be modular and extensible for multimodal expansion, including the incorporation of additional imaging modalities such as ocular/eye-based visual biomarkers in future iterations. This reflects a broader direction toward multi-source diagnostic intelligence systems. We also discuss important real-world limitations, particularly domain shift in TB presentation across regions (e.g., Mycobacterium africanum prevalence in West Africa), emphasizing the need for regionally representative datasets and retraining strategies. Overall, the work focuses on: Reproducible ML pipeline design in Python Bridging research-grade models and deployable tools Explainable AI for healthcare trust and usability Practical constraints in low-resource deployment environments Project Links GitHub: https://github.com/iamamofa/TB-detection