| CPC G16H 30/40 (2018.01) [G06T 7/0012 (2013.01); G16H 50/20 (2018.01); G16H 50/30 (2018.01); G06T 2207/20081 (2013.01); G06T 2207/30004 (2013.01); G06T 2207/30168 (2013.01)] | 14 Claims |
|
1. A computer aided diagnosis CADx system for analysing medical scan images generated using an imaging modality and determining a disease risk score for the medical scan image comprising:
a processor comprising an input circuit for receiving an input comprising at least one medical scan image;
a dynamic protocol adaptation circuit for receiving and analysing scan protocol information, wherein the scan protocol information comprises at least one of: one or more acquisition parameters, one or more reconstruction parameters, one or more features resulting from the combination of acquisition and reconstruction parameters, for the at least one medical scan image and providing a dynamic output from the protocol adaptation circuit as an input to a disease prediction model based on the received scan protocol information; wherein at least one of the disease prediction model and the dynamic protocol adaptation circuit consist of a machine learning model, trained using training data comprising medical input data, wherein the medical input data comprises at least one of: one or more of medical scan images, one or more clinical parameters, one or more biomarkers and ground truth labels associated with the medical input data and scan protocol information; wherein the training data is provided to the disease prediction model and the dynamic protocol adaptation circuit so that the disease prediction model and dynamic protocol adaptation circuit are trained to recognise input medical scan images and scan protocol information, wherein during the training with the machine learning model, one or more of the scan protocol parameters from the training data is dropped out before the training data is provided to the disease prediction model and/or the dynamic protocol adaptation circuit, the dropped scan protocol parameters are replaced with a missing protocol parameter value, and training is performed with the reduced training set and the missing protocol parameter value, and the disease prediction model and/or dynamic protocol adaptation circuit are trained to handle input medical scan images that have missing scan protocol information; wherein, when the CADx system encounters an input medical scan image with missing scan protocol information, the CADx system will substitute the missing protocol information with the missing protocol parameter value so the input medical scan image can be analysed for a disease risk score and
said processor comprising an output circuit to produce an output that is the disease risk score calculated from the disease prediction model for the at least one medical scan image.
|