Decision support system for medical therapy planning
For decision support in a medical therapy, machine learning provides a machine-learned generator for generating a prediction of outcome for therapy personalized to a patient. Deep learning may result in features more predictive of outcome than handcrafted features. More comprehensive learning may be provided by using multi-task learning where one of the tasks (e.g., segmentation, non-image data, and/or feature extraction) is unsupervised and/or draws on a greater number of training samples than available for outcome prediction alone.
1. A method for decision support in a medical therapy system, the method comprising:
acquiring a medical scan of a patient;
generating a prediction of outcome of therapy for the patient, the outcome generated by a machine-learned multi-task generator having been trained based on a segmentation loss and an outcome loss; and
displaying an image of the outcome.
2. The method of claim 1 wherein acquiring comprises scanning the patient with a computed tomography scanner.
3. The method of claim 1 wherein acquiring comprises acquiring voxel data representing a three-dimensional distribution of locations in a volume of the patient, and wherein generating comprises generating based on input of the voxel data.
4. The method of claim 1 wherein generating comprises generating with the machine-learned multi-task generator comprises a convolutional neural network.
5. The method of claim 4 wherein generating comprises generating with the convolutional neural network comprising an encoder network trained as part of an encoder and decoder network, and the convolutional neural network comprising a neural network configured to receive bottleneck features of the encoder network, the neural network generating the outcome.
6. The method of claim 1 wherein generating comprises generating with the machine-learned multi-task generator having been trained with deep learning to create segmentations compared to ground truth segmentation for the segmentation loss and to compare prediction results to ground truth results for the outcome loss.
7. The method of claim 1 wherein generating comprises generating with the machine-learned multi-task generator having been trained with a greater number of training data samples for a segmentation loss than for the outcome loss.
8. The method of claim 1 wherein generating comprises generating in response to input of non-image data for the patient to the machine-learned multi-task generator, the machine-learned multi-task generator comprising an autoencoder.
9. The method of claim 1 wherein generating comprises generating with the machine-learned multi-task generator having been selected from a group of trained networks, each of the trained networks corresponding to different combinations of types of input data.
10. A method for decision support in a medical therapy system, the method comprising:
acquiring first non-image data representing characteristics of a patient;
generating a prediction of outcome of therapy for the patient in response to input of the first non-image data, the outcome generated by a machine-learned network having been trained as an autoencoder for generating second non-image data representing patients from third non-image data representing other patients, the second non-image data of a same type as the third non-image data; and
displaying an image of the outcome.
11. The method of claim 10 wherein acquiring the first non-image data comprises acquiring genomic, clinical, measurement, and/or family history data of the patient, and wherein the third non-image data comprise genomic, clinical, measurement, and/or family history data of the other patients.
12. The method of claim 10 wherein generating comprises generating with the autoencoder comprising an encoder network trained as part of an encoder and decoder network, and a fully connected network configured to receive bottleneck features of the encoder network, the fully connected network generating the outcome.
13. The method of claim 10 wherein generating comprises generating with the machine-learned network having been trained with multi-task deep learning to create segmentations compared to ground truth segmentation for a segmentation loss and to create outcome predictions compared to ground truth results for an outcome loss.
14. The method of claim 10 wherein generating comprises generating with the machine-learned network having been selected from a group of trained networks, each of the trained networks corresponding to different combinations of types of input data.
15. A method for machine training decision support in a medical therapy system, the method comprising:
defining a machine-training architecture with a plurality of data input options, each data input option corresponding to a different type of data;
machine training the machine-training architecture with different combinations of one or more of the data input options, the machine training with the different combinations resulting in different machine-trained generators configured to generate an output prediction for therapy;
determining performance of each of the machine-trained generators;
selecting one of the machine-trained generators based on the performance; and
storing the selected machine-trained generator.
16. The method of claim 15 wherein machine training comprises machine training as a multi-task operation with a loss function comprising a weighted combination of an image feature loss and an outcome loss.
17. The method of claim 15 wherein defining comprises defining the machine-training architecture as an autoencoder where the data input options comprise non-image data and as a neural network configured to receive bottleneck features from the autoencoder, the neural network trained to generate the prediction.
18. The method of claim 15 wherein defining comprises defining with the different types of data comprising image data cropped to tumors, a therapy dose map, image data in which the tumors are less than half a represented region, demographic data, and clinical data.