The Chest X-ray Imaging Dataset for Multiple Cardio-respiratory Diseases in Ethiopia (Afro Chest X-ray for short) is a project funded by the LacunaFund whose aim is to close the gap in health disparities by fostering interdisciplinary collaborations that create, expand, or aggregate labeled training and evaluation datasets.
Cardio-respiratory diseases (cardiovascular and respiratory diseases) are recognized as serious, worldwide public health concerns that have remained among the leading causes of death globally. There are not many publicly available datasets from Africa making it difficult to determine whether tools and techniques developed in other geographies are as effective in our context. In this project, we propose to create a labeled chest X-ray dataset for multiple cardio respiratory diseases in Ethiopia. We will publish the dataset as open source. We believe this dataset will stimulate researchers and practitioners in Africa and beyond to push the limits of current methods to adapt them to the African context and build assistive technologies that could empower the scarce radiologists.
We envision this dataset to have an impact primarily in research and applications of the medical imaging domain. It will also be essential for researchers in natural language processes and entrepreneurs in medical imaging. Below we highlight some relevant research directions:
Bias in Machine learning: This dataset will be useful to study the effect of selection bias in building machine learning models for X-ray images. There are massive datasets such as CheXpert gathered from different populations. The typical way to leverage models trained on big dadaist's in one population and apply them to another is through transfer learning. Our project will enable researchers to study the implications of selection bias in machine learning. For instance, how models trained on a big dataset from a different population and adapted to our population at test time compares with models built on a relatively smaller dataset on the same population and what effects they have across demography.
Interpretable Machine Learning: This dataset will allow researchers to study interpretive models for image classification. Computer vision models are interpretable in this case when they classify X-ray images on the basis of features that a radiologist can directly understand.
Medical Information Extraction: Large parts of X-ray reports are in written text form. Most of them are written in unstructured form. The usage of standard electronic health records will be convenient to process medical data. Our dataset can serve as a foundation to build and evaluate such natural language processing tools.
Real-world Products: As shortage of radiologists is a common phenomena in medically under-served regions such as Ethiopia, we believe entrepreneurs who want to solve this problem can leverage this dataset as a starting point to provide services that can improve health care services in rural areas.
We are a group of researchers and medical professionals from Addis Ababa University, Tikur Anbessa Specialized Hospital, Ayder Referral Hospital, Menelik II Comprehensive Specialized Hospital, Kotebe general hospital and Lesan AI, who are joining hands to build a diverse dataset for cardio-respiratory diseases in Ethiopia.