US 12,476,005 B2
Machine learning techniques for generating historically dynamic explanation data objects
Gregory Buckley, Dublin (IE); David S. Monaghan, Dublin (IE); Johnathon E. Schultz, Cross Hill, SC (US); and Ajay Ajit Maity, Donegal (IE)
Assigned to Optum, Inc., Minnetonka, MN (US)
Filed by Optum, Inc., Minnetonka, MN (US)
Filed on Sep. 23, 2021, as Appl. No. 17/483,319.
Prior Publication US 2023/0090591 A1, Mar. 23, 2023
Int. Cl. G16H 50/20 (2018.01); G06N 5/02 (2023.01); G16H 30/20 (2018.01)
CPC G16H 50/20 (2018.01) [G06N 5/027 (2013.01); G16H 30/20 (2018.01)] 20 Claims
OG exemplary drawing
 
1. A computer-implemented method comprising:
generating, by one or more processors, a convolutional embedding vector of a dental image and one or more tooth-specific bounding box sub-images of the dental image;
generating, by the one or more processors, based at least in part on the dental image and the one or more tooth-specific bounding box sub-images of the dental image, and by using an encoder machine learning framework, a current diagnosis code vector for the dental image;
generating, by the one or more processors, based at least in part on historical data associated with an individual identifier associated with the dental image, and by using the encoder machine learning framework, a historical diagnosis code vector for the dental image;
generating, by the one or more processors, using the encoder machine learning framework, and based at least in part on the historical diagnosis code vector and the current diagnosis code vector, a new diagnosis code vector for the dental image configured to describe a set of diagnosis codes depicted by the dental image independent of the historical data associated with the individual identifier;
generating, by the one or more processors and based at least in part on the convolutional embedding vector and the new diagnosis code vector for the dental image, a dental image embedding for the dental image;
generating, by the one or more processors, based at least in part on the dental image embedding, and by using a decoder machine learning framework, a dynamic explanation data object, wherein the decoder machine learning framework and the encoder machine learning framework are trained, in combination, based at least in part on a measure of deviation of the dynamic explanation data object and a ground-truth data object for the dental image;
generating, by the one or more processors and based at least in part on the dynamic explanation data object, an automated caption for the dental image; and
generating, by the one or more processors, user interface data for a prediction output user interface that is configured to display the automated caption and the dental image.