Physicians requested that medical images containing certain artifacts or visual similarities to other images be grouped together and made searchable.
Enable visual search, classification and sorting of image artifacts and similar images.
Images of hearts contain certain visual elements that are diagnostic to a variety of conditions. A visual search tool that was able to locate and compare similar visual elements and overall images was needed.
We trained a group of deep learning neural networks with a large body of imaging data. The combined outputs were then used as input for another contextually trained neural network to identify and automatically label related anatomical-based visual elements along with clinically relevant image correlations. The result is a natural language enabled visual search tool that in real time can identify a wide variety of anatomical irregularities in newly obtained imaging data.