Method for controlling scanner by estimating patient internal anatomical structures from surface data using body-surface and organ-surface latent variables

1. A method for controlling a scanner, comprising

sensing an outer surface of a body of a subject, the sensing providing body surface data comprising a skin surface of the subject;

using a machine-trained predictor to predict a surface of an internal organ of the subject based on the body surface data, the machine-trained predictor including a regressor that receives as input to the regressor a representation of the skin surface and outputs, in response to the input and by the regressor, a representation of the surface predicted as a mesh or volume mask output by the machine-trained predictor, wherein the predicting includes:

encoding the body surface data into body surface principal coordinates;

predicting a set of organ principal coordinates representing the surface of the internal organ; and

decoding the set of organ principal coordinates to predict the surface of the internal organ, wherein the encoding is performed by a first autoencoder, predicting the set of organ principal coordinates is performed by Hall the regressor, and the decoding is performed by a second autoencoder, the method further comprising, before the sensing:

training the first autoencoder using a first plurality of body surfaces;

training the second autoencoder using a first plurality of organ surfaces; and

training the regressor using a plurality of body-surface, organ-surface sets, each body-surface, organ-surface set including a plurality of body-surface latent variable values and a plurality of organ-surface latent variable values corresponding to the body-surface latent variable values; and

controlling the scanner based on the predicted surface of the internal organ, the scanner controlled for scan range or coil placement.

2. The method of claim 1 , wherein using the machine-trained predictor to predict the surface includes:

reducing dimensionality of the body surface data;

predicting a compact representation of the surface of the internal organ based on the reduced dimensionality body surface data; and

expanding the compact representation.

3. The method of claim 2 , wherein the regressor is a sparse linear regressor.

4. The method of claim 2 , wherein the regressor is a least absolute shrinkage and selection operator (LASSO) regressor.

5. The method of claim 2 , wherein the regressor is a deep neural network.

6. The method of claim 1 , wherein the scanner is a magnetic resonance (MR) scanner, and the controlling includes selecting the coil placements of a plurality of coils of the MR scanner.

7. The method of claim 1 , wherein the scanner is a computed tomography (CT) scanner, and the controlling includes selecting the scan range of the CT scanner.

8. A method for controlling a scanner, comprising

sensing an outer surface of a body of a subject, the sensing providing body surface data comprising a skin surface of the subject;

using a machine-trained predictor to predict a surface of an internal organ of the subject based on the body surface data, wherein using the machine-trained predictor to predict the surface includes: reducing dimensionality of the body surface data by a machine-trained encoder of the machine-trained predictor, predicting a compact representation of the surface of the internal organ as output of a machine-trained regressor in response to input to the machine-trained regressor of the reduced dimensionality body surface data, and expanding the compact representation by a machine-trained decoder of the machine-trained predictor, wherein the predicting includes:

encoding the body surface data into body surface principal coordinates;

predicting a set of organ principal coordinates representing the surface of the internal organ; and

decoding the set of organ principal coordinates to predict the surface of the internal organ, wherein the encoding is performed by a first autoencoder as the machine-trained encoder, predicting the set of organ principal coordinates is performed by the machine-trained regressor, and the decoding is performed by a second autoencoder, the method further comprising, before the sensing:

training the first autoencoder using a first plurality of body surfaces;

training the second autoencoder using a first plurality of organ surfaces; and

training the regressor using a plurality of body-surface, organ-surface sets, each body-surface, organ-surface set including a plurality of body-surface latent variable values and a plurality of organ-surface latent variable values corresponding to the body-surface latent variable values; and

controlling the scanner based on the predicted surface of the internal organ.

9. A method for controlling a scanner, comprising

sensing an outer surface of a body of a subject, the sensing providing body surface data comprising a skin surface of the subject;

using a machine-trained model to predict a surface of an internal organ of the subject based on the body surface data, wherein the predicting includes encoding the body surface data into body surface principal coordinates by a first machine-trained autoencoder, predicting a set of organ principal coordinates representing the surface of the internal organ as an output by a machine-trained correlator in response to input of the body surface principal coordinates to the machine-trained correlator, and decoding the set of organ principal coordinates to predict the surface of the internal organ by a second machine-trained autoencoder, the first autoencoder having been trained using a first plurality of body surfaces, the second autoencoder having been trained using a first plurality of organ surfaces, and the machine-trained correlator having been trained using a plurality of body-surface, organ-surface sets, each body-surface, organ-surface set including a plurality of body-surface latent variable values and a plurality of organ-surface latent variable values corresponding to the body-surface latent variable values; and

controlling the scanner based on the predicted surface of the internal organ.

10. The method of claim 9 wherein using the machine-trained model to predict the surface includes: reducing dimensionality of the body surface data, predicting a compact representation of the surface of the internal organ based on the reduced dimensionality body surface data, and expanding the compact representation.