Automated Cardiothoracic Ratio Estimation Using CPU-Based Deep Learning on Chest X-Rays: A Novel Approach in Sub-Saharan Africa

Background/Objectives: Manual cardiothoracic ratio (CTR) measurement from chest X-rays remains widely used but is time-consuming and error-prone, particularly in low-resource clinical settings. This study presents a CPU-based deep learning model for automated CTR estimation designed for use in Sub-Saharan Africa. Methods: A U-Net segmentation model was trained on 3,000 anonymized chest radiographs to extract heart and thoracic contours. The trained model was deployed via a lightweight desktop application optimized for inference on standard CPUs. Results: The system achieved a mean absolute error (MAE) of 0.019 and an R2 value of 0.91 when compared to expert manual CTR measurements. Usability testing with local radiologists and radiographers revealed strong acceptance, highlighting the tool’s diagnostic and educational value. Conclusions: The findings suggest that accurate and efficient AI-based CTR estimation can be performed without GPU support, offering an accessible and cost-effective diagnostic aid for low- and middle-income healthcare systems.