US 12,133,993 B2
Real-time anatomic position monitoring in radiotherapy using machine learning regression
Philip P. Novosad, Montreal (CA); and Silvain Beriault, Longueuil (CA)
Assigned to Elekta, Inc., Atlanta, GA (US)
Filed by Elekta, Inc., Atlanta, GA (US)
Filed on Apr. 28, 2021, as Appl. No. 17/302,252.
Prior Publication US 2022/0347493 A1, Nov. 3, 2022
Int. Cl. A61N 5/10 (2006.01); G06T 7/33 (2017.01)
CPC A61N 5/1068 (2013.01) [A61N 5/1049 (2013.01); G06T 7/337 (2017.01); A61N 2005/1055 (2013.01); A61N 2005/1091 (2013.01); G06T 2207/10081 (2013.01); G06T 2207/10088 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/30096 (2013.01)] 23 Claims
OG exemplary drawing
 
1. A computer-implemented method for monitoring anatomic position of a human subject for a radiotherapy treatment session, the method comprising:
obtaining three-dimensional image data corresponding to the subject, the three-dimensional image data including: a reference volume that represents anatomy of the subject in three dimensions, at least one region of interest defined within the three dimensions, and an intermediate three-dimensional reference volume captured prior to the radiotherapy treatment session;
performing a registration of the intermediate three-dimensional reference volume to the reference volume;
obtaining two-dimensional image data corresponding to the subject, the two-dimensional image data captured during the radiotherapy treatment session, and the two-dimensional image data capturing at least a portion of the at least one region of interest;
extracting features from the two-dimensional image data;
providing the extracted features as input to a machine learning regression model, the machine learning regression model trained to estimate a spatial transformation in the three dimensions of the reference volume from the features extracted from the two-dimensional image data, wherein analysis of the extracted features by the machine learning regression model includes use of the registration; and
obtaining, from output of the machine learning regression model, a relative motion estimation of the at least one region of interest, wherein the relative motion estimation indicates motion relative to the reference volume which is estimated from the extracted features.