Systems and methods for optimizing treatment planning

A radiotherapy system comprising:

• a memory storing computer executable instructions;
• at least one processor device communicatively coupled to the memory, wherein the at least one processor device is configured to execute the computer executable instructions to:
• receive training data reflecting radiotherapy treatment data, wherein the training data includes:
• a training feature vector including data generated from a plurality of training medical images, the training feature vector including image data of a first region of each training medical image and image data of a second region of each training medical image, and
• a training target vector including at least one of a treatment plan outcome or a radiotherapy session parameter;
• determine a training model, wherein determining the training model comprises:
• obtaining, from the training feature vector, signed distances of the training medical images, each signed distance reflecting a minimum distance between a non-target tissue of a subject in the first region of a given training medical image and a boundary of a target volume of the subject in the second region of the given training medical image,
• obtaining, from the training target vector, radiotherapy dose data reflecting radiation doses delivered to the non-target tissues in the first regions of the training medical images, and
• for each signed distance of the training feature vector, determining conditional probabilities of delivering the radiation doses corresponding to the radiotherapy dose data to a non-target tissue having a minimum distance from a boundary of a target volume that is equal to the signed distance;
• receive testing data associated with a patient, wherein the testing data includes a descriptive feature vector that comprises a signed distance of the patient reflecting a minimum distance between a non-target tissue of the patient in a first region of a testing medical image and a boundary of a target volume of the patient in a second region of the testing medical image;
• determine a therapy model, wherein determining the therapy model comprises:
• identifying a signed distance in the training feature vector that is equal to the signed distance of the patient, and
• determining a distribution of the conditional probabilities of the identified signed distance and determining the therapy model based on the distribution;
• and
• generate a treatment plan for the testing data based on the therapy model, the treatment plan indicating a radiation dose of the target volume of the patient;
• and
• a radiotherapy device configured to use the generated treatment plan to deliver radiation to the patient.

One or more computer-readable storage devices having computer-executable instructions stored thereon causing a processor device to perform a method for generating a treatment plan for a patient, the method comprising:

• receiving training data reflecting therapy treatment data, wherein the training data includes:
• a training feature vector including data generated from a plurality of training medical images, the training feature vector including image data of a first region of each training medical image and image data of a second region of each training medical image, and
• a training target vector including at least one of a treatment plan outcome or a radiotherapy session parameter;
• selecting a specific feature type from the received training data, wherein the specific feature type includes signed distance data reflecting a minimum distance between a non-target tissue of a subject in the first region of a given training medical image and a boundary of a target volume of the subject in the second region of the given training medical image;
• selecting a specific target type from the received training data, wherein the specific target type includes radiotherapy dose data reflecting radiation doses delivered to the non-target tissues in the first regions of the training medical images and wherein the specific feature type is predictive of the specific target type;
• aggregating feature vectors of the specific feature type and target vectors of the specific target type across a plurality of the received training data, the aggregated feature vectors including the signed distances of the training medical images and the aggregated target vectors including the radiation doses delivered to the non-target tissues in the first regions of the training medical images;
• for each signed distance of the aggregated feature vectors, determining conditional probabilities of delivering the radiation doses corresponding to the aggregated target vectors to a non-target tissue arranged at a location corresponding to the signed distance;
• receiving testing data associated with the patient, wherein the testing data includes a descriptive feature vector that comprises signed distance data of the patient reflecting a minimum distance between a non-target tissue of the patient in a first region of a testing medical image and a boundary of a target volume of the patient in a second region of the testing medical image;
• identifying one of the signed distances of the aggregated feature vectors that is equal to the signed distance of the patient;
• determining a probability density function based on the conditional probabilities of the identified signed distance;
• extracting parameters of the probability density function;
• minimizing a convex optimization function based on the parameters of the probability density function to determine one or more parameters of a treatment plan, the treatment plan indicating a radiation dose of the target volume of the patient;
• and
• delivering radiation to the patient according to the one or more parameters of the treatment plan.