Improving tuberculosis diagnostics using deep learning and mobile health technologies among resource-poor communities in Perú

Tuberculosis (TB) an infectious disease and remains a major cause of death globally. The World Health Organization (WHO) estimates that there were 10.4 million new TB cases worldwide in 2015. The majority of the infected populations come from resource-poor and marginalized communities with poor healthcare infrastructure. It is critical to reduce TB diagnosis delay in mitigating disease transmission and minimizing the reproductive rate of the tuberculosis epidemic. To combine machine learning and mobile computing techniques may help to accelerate the TB diagnosis among these communities. The goal of our research is to reduce TB patient wait times for being diagnosed by developing new machine learning techniques and mobile health technologies. In this paper, major technique barriers and proposed system architecture are first introduced. Then two major progresses are reported: (1) To develop an X-ray image database and annotation software dedicated for automated TB screening. The annotation software can help to highlight the TB manifestations, which are very useful for machine learning algorithms; (2) To develop effective and efficient computational models to classify the image into different category of TB manifestations. The model we proposed is a deep convolutional neural networks (CNN)-based models. We have conducted substantial experiments and the results have demonstrated that our approach is promising. We envision our future work includes two research activities. First, we plan to improve the performance of the algorithms with deeper neural networks. Second, we plan to implement our algorithms on mobile device and deploy our system in the city of Carabayllo, a high-burden TB area in Lima, the capital of Perú.