MAIDA - Medical AI Data for All
Bringing you diverse medical image datasets from across the globe
News: We’re hiring a full-time position for a relationship manager. Apply here.
We Share De-identified Datasets from Across the Globe
Medical data is rarely shared across institutions due to data security concerns [1], vendor lock-in, and significant upfront costs for data infrastructure [2]. This slows down medical AI progress, widening technology gaps between developing and high-income countries. We are facilitating de-identified public data sharing among the global community through a streamlined process.
We Focus on Diversity
Existing data lacks diverse representation. Algorithms for clinical applications are disproportionately trained on a few hospitals, with little to no representation at a national or global level [3]. Populations not adequately represented in the training cohort will likely receive biased results. For example, darker skin is underrepresented in widely used dermatology datasets [4], and state-of-the-art dermatology AI models substantially underperform on darker skin tones until they are fine-tuned on diverse data [5]. We create diverse datasets that capture different demographics, clinical practices and imaging equipment used around the world.
We Enable Clinically Useful Artificial Intelligence
Existing public datasets usually provide annotations that oversimplify the tasks clinicians perform, making it difficult to train artificial intelligence (AI) models for clinically useful decisions. Most datasets also lack the information that would otherwise be available to clinicians in their actual workflow. We work with multidisciplinary teams to curate datasets with relevant clinical information and granular, high-quality annotations. In addition to high-quality annotations, our multi-institutional datasets enable the development of AI models that can be validated across diverse clinical settings.
Clinical Settings
Endotracheal Tube Assessment in Adult ICU
Endotracheal (ET) tubes are placed into the windpipe for patients to receive oxygen, medicine, or anesthesia. A misplaced ET tube can cause serious complications such as hypoxia, hypercapnia, and pneumothorax. Still, ET tube misplacement is relatively common, with a rate between 7 and 15.5 % and an even higher rate among women. Currently, ET tubes are placed daily in intensive care units (ICUs) with their placement assessed based on CXRs. Our goal is to facilitate the development of models for automated assessment of ET tube placement, which will reduce the burden on radiologists and front-line clinicians, as well as enable instant evaluation of the ET tube positioning
Endotracheal Tube Assessment in Neonatal ICU
Similar to adult ICU, endotracheal (ET) tubes are placed daily in neonatal ICUs, and physicians are asked to evaluate their placement based on CXRs. Determining the optimal position is challenging in this clinical setting because very little margin of error is allowed for premature babies. Our goal is to facilitate the development of models for automated assessment of ET tube placement in kids.
Pneumonia Detection in the Emergency Department
CXRs are routinely ordered in the emergency department to rule out pneumonia. Early treatment of pneumonia is critical for reducing the risk of more severe complications. Our goal is to facilitate the development ML models for automated detection of pneumonia in the emergency department (ED). These models can improve diagnosis consistency amongst clinicians and radiologists, speed up the ED workflow, and reduce the burden on radiologists by generating the corresponding report.