<![CDATA[Introduction Thalassemia is a prevalent inherited hemoglobinopathies in Indonesia, with an estimated carrier rate of 3-10% of the population. Under-resourced areas struggle with limited diagnostic services like genetic testing or hemoglobin electrophoresis. Prevention and clinical management by early screening and detection of potential thalassemia carriers are critical. This opens the opportunity to leverage computer vision and on-device machine learning to develop an iOS-based application, providing a digital anamnesis tool that is affordable and accessible for early thalassemia risk assessment. Methods Large, diverse datasets were collected to compile the clinical reports and to conduct the AI training. The system architecture of Thalassemia NusaCare AI consists of three integrated computational modules: Vision-Based Analysis using MobileNetV3 for facial and blood image detection, CBC laboratory data Interpretation using OCR and decision-tree algorithms. Digital Anamnesis uses adaptive, federated learning for accurate, real time Thalassemia risk prediction. Results Preliminary testing using a curated dataset (n=200 images, 120 lab entries, 80 questionnaire records) leads to a mean classification accuracy of 91.3% for detecting thalassemia major, minor, and non-thalassemic anemia, demonstrating high operational efficiency. Hybrid ensemble models result in an F1-score of 0.88 and enhanced sensitivity by 12% relative to single-input models. User experience testing with early adopters also suggested strong usability and intuitiveness (SUS = 89.2). Conclusion Thalassemia NusaCare AI integrates edge AI and inclusive design to deliver adaptive diagnostics in low-connectivity areas. Combining visual, numerical, and behavioral data, enabling on-device screening that aligns with Indonesia’s “Thalassemia-Free 2045” through federated learning and clinical collaboration. ]]>
