Automatic generation of patient-specific radiation therapy planning parameters
An apparatus and method for automatically generating radiation treatment planning parameters are disclosed. In accordance with the illustrative embodiment, a database is constructed that stores: (i) patient data and past treatment plans by expert human planners for these patients, and (ii) optimal treatment plans that are generated using multi-objective optimization and Pareto front search and that represent the best tradeoff opportunities of the patient case, and a predictive model (e.g., a neural network, a decision tree, a support vector machine [SVM], etc.) is then trained via a learning algorithm on a plurality of input/output mappings derived from the contents of the database. During training, the predictive model is trained to identify and infer patterns in the treatment plan data through a process of generalization. Once trained, the predictive model can then be used to automatically generate radiation treatment planning parameters for new patients.
1. A method comprising:
receiving by a data-processing system:
(a) first data based on a geometric characterization of one or more organs at risk proximate to a target volume of a patient P, wherein the geometric characterization associates each of a plurality of distances from the target volume with a respective percentage of the volume of the one or more organs at risk;
(b) second data comprising a size of the target volume and the respective sizes and shapes of the one or more organs at risk; and
generating, by said data-processing system using a predictive model, one or more radiation treatment planning parameters for said patient P based on the first data and the second data.
2. The method of claim 1 , wherein said one or more radiation treatment planning parameters are represented by at least one of a dose distribution and a dose volume histogram.
3. The method of claim 1 further comprising receiving by a data processing system third data based on a geometric characterization of said target volume with respect to said one or more organs at risk.
4. The method of claim 1 wherein the first data and the second data are derived via a feature selection algorithm that is applied to a larger set of data that represent geometric characterizations of one or more organs at risk proximate to a target volume.
5. The method of claim 4 wherein said feature selection algorithm comprises a principal component analysis, and wherein said principal component analysis is based on an N-by-N covariance matrix, and wherein N is a positive integer greater than one.
6. The method of claim 5 wherein the first data and the second data are based on the M largest eigenvalues of said covariance matrix, and wherein M is a positive integer between 1 and N inclusive.
7. The method of claim 1 and further wherein said predictive model has been trained on a set of input-output pairs, and wherein the output of each input-output pair is based on a dose distribution and a dose volume histogram for a respective patient, and wherein the input of each input-output pair comprises one or more data that are based on the geometric characterization for said respective patient.
8. A method comprising training by a data-processing system a predictive model on a plurality of input-output mappings, wherein the output of each input-output mapping is based on a dose distribution and a dose volume histogram for a respective patient, and wherein the input of each input-output mapping comprises one or more data that are based on a geometric characterization of one or more organs at risk proximate to a target volume of the respective patient, wherein said geometric characterization associates each of a plurality of distances from said target volume with a respective percentage of the volume of said one or more organs at risk.
9. The method of claim 8 further comprising generating a computer-executable program based on the trained predictive model.
10. The method of claim 8 wherein said one or more data also comprises the size of one or more partial volumes.
11. The method of claim 8 wherein said geometric characterization associates a three-dimensional point in said target and one or more organs at risk with a set of values that represent the unique relationship between the point and the target and the organs at risk.
12. The method of claim 8 wherein the output of each input-output mapping is derived via a feature selection algorithm that is applied to dose volume histograms for a plurality of patients.
13. The method of claim 12 wherein said feature selection algorithm comprises a principal component analysis, and wherein said principal component analysis is based on an Z by Z covariance matrix, and wherein Z is the plurality of points from a dose volume histogram for a respective patient, and wherein Z is a positive integer greater than one.
14. The method of claim 13 wherein the output of an input-output mapping comprises the Q largest eigenvalues of said covariance matrix, and wherein Q is a positive integer between 1 and Z inclusive.
15. A method comprising storing in a database, by a data-processing system:
(a) one or more planning parameters of a first radiation treatment plan, wherein said first radiation treatment plan is for a first patient and is generated by one of:
(i) an expert human planner, and
(ii) a data-processing system using one or both of multi-objective optimization and pareto front search;
(b) a first geometric characterization of a first non-empty set of organs at risk proximate to a first target volume of said first patient, wherein the first geometric characterization is a function of volume for the organs at risk;
(c) one or more planning parameters of a second radiation treatment plan, wherein said second radiation treatment plan is for a second patient and is generated by one of:
(i) an expert human planner, and
(ii) a data-processing system using one or both of multi-objective optimization and pareto front search; and
(d) a second geometric characterization of a second non-empty set of organs at risk proximate to a second target volume of said second patient, wherein the second geometric characterization is a function of volume for the organs at risk.
16. The method of claim 15 further comprising:
generating by said data-processing system:
(i) a first input-output mapping based on said first geometric characterization and on said one or more planning parameters of said first radiation treatment plan, and
(ii) a second input-output mapping based on said second geometric characterization and on said one or more planning parameters of said second radiation treatment plan; and
training by said data-processing system a predictive model on said first input-output mapping and said second input-output mapping.
17. A method comprising:
receiving by a data-processing system one or more data, wherein at least one of the data is based on a geometric characterization of one or more organs at risk proximate to a target volume of a patient P, and wherein the geometric characterization associates each of a plurality of distances from said target volume with a respective percentage of the volume of said one or more organs at risk; and
generating by the data-processing system one or more radiation treatment planning parameters for the patient P based on the one or more data.
18. A method comprising:
receiving first data and second data at a data-processing system, wherein:
(a) the first data is based on a geometric characterization of one or more organs at risk proximate to a target volume of a patient P; and
(b) the second data is based on the size of the target volume and the respective sizes and shapes of the one or more organs at risk; and
generating by the data-processing system one or more radiation treatment planning parameters for the patient P based on the first data and the second data.