Challenge
The developing world is facing a severe shortage of radiologists.
Proposed solution
We believe that AI, especially when deployed to work alongside clinicians, is critical to solving this public health crisis. To this end, we have implemented a deep neural network that is able to accurately detect evidence of pneumonia and other abnormalities from chest x-rays.
We offer bespoke deployment of our model in a clinical setting to ensure the unique issues faced by different clients are addressed. Based on the intended purpose, specific adaptions may be required. For example, the tool may need to be calibrated to minimise false negatives, in order for it to be used for triage purposes. Alternatively, the tool may need to be calibrated to provide its best estimates in cases where it is intended to guide primary care physicians with decision making.
What sets our model apart?
Since there is a large publicly available dataset of labelled chest x-rays, several other algorithms tackling this task exists. However, as far as we have been able to detect, no case studies detailing testing or implementation in a live environment appear to be available.
We believe we are uniquely placed to work with partners in a clinical setting to train and calibrate a model for the specific intended purpose, test the model on new data and explore how additional data sources could be incorporated into the model to better diagnose respiratory disease. This would prove particularly valuable in low-resource settings where few radiologists exist.
We are currently seeking a collaborating partner interested in testing our model in a clinical setting.
