# US 10,864,382 B2 · System and method for radiation therapy treatment planning > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 10,864,382 B2 - Document ID: US10864382B2 - Status: Granted - Kind: B2 - Application number: 15/959,470 - Filed: 2018-04-23 - Published: 2018-10-25 - Granted: 2020-12-15 - Earliest claimed: 2017-04-24 - Assignees: RaySearch Laboratories AB - Applicants: RaySearch Laboratories AB - Inventors: Rasmus Bokrantz - Web page: https://patentgrep.com/patent/US10864382B2 ## CPC - A61N 5/1031: {using a specific method of dose optimization} - A61N 5/1039: {using functional images, e.g. PET or MRI} - G06F 16/90335: {Query processing} - G06N 5/022: Knowledge engineering; Knowledge acquisition - G06N 20/00: Machine learning - G16H 20/40: relating to mechanical, radiation or invasive therapies, e.g. surgery, laser therapy, dialysis or acupuncture - G16H 30/40: for processing medical images, e.g. editing - G16H 50/70: for mining of medical data, e.g. analysing previous cases of other patients - A61N 2005/1041: {using a library of previously administered radiation treatment applied to other patients} - A61N: ELECTROTHERAPY; MAGNETOTHERAPY; RADIATION THERAPY; ULTRASOUND THERAPY (measurement of bioelectric currents A61B; surgical instruments, devices or methods for transferring non-mechanical forms of energy to or from the body A61B18/00; anaesthetic apparatus in general A61M; incandescent lamps H01K; infrared radiators for heating H05B) - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G16H: HEALTHCARE INFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR THE HANDLING OR PROCESSING OF MEDICAL OR HEALTHCARE DATA ## Abstract Radiation therapy treatment planning methods facilitate predicting achievable dose distributions by training a prediction model for clinical DVH curves based on simplified DVH curves obtained using standardized methods. Machine learning applied on pairs of clinical and simplified DVH curves enables predicting actual clinical DVH curves based on simplified DVH curves. ## Claims ### Claim 1 A method for use in radiation therapy treatment planning, comprising: a. obtaining a protocol plan, which has been created by applying a plan generation protocol to a set of patient data for a current patient, said set of patient data comprising at least one planning image; b. calculating at least one protocol dose volume histogram (DVH) curve, each protocol DVH curve relating a dose distribution for the protocol plan to a volume of a structure in a structure set, the dose distribution and the structure set being associated with the at least one planning image; c. obtaining estimated clinical DVH curves by sending the at least one protocol DVH curve as a query to a prediction model, the prediction model being arranged to determine estimated clinical DVH curves on the basis of the at least one protocol DVH curve for the current patient based on a relationship between previous protocol DVH curves and previous clinical DVH curves for a collection of previous patients, wherein the estimated clinical DVH curves are determined based on clinical DVH curves from previous patients that have protocol DVH curves having a distance to the at least one protocol DVH curve of the current patient below a specified threshold value; and d. using at least one of the estimated clinical DVH curves in treatment planning for the current patient. ### Claim 2 (depends on 1) The method according to claim 1, wherein the plan generation protocol is a sequence of tasks to be performed sequentially, the tasks involving at least one of the following tasks: augmenting the structure set with additional structures, either manually or by automatic image segmentation; defining radiation beams; defining settings including resolutions and tolerances; defining an optimization problem and optimizing a dose with respect to the optimization problem; or calculating a clinical dose distribution. ### Claim 3 (depends on 2) The method according to claim 2, wherein defining the optimization problem further comprises updating the optimization problem based on the dose distribution for the protocol plan. ### Claim 4 (depends on 1) A computer program product comprising non-transitory computer-readable code means which, when run in a computer will cause the computer to perform the method according to claim 1. ### Claim 5 (depends on 4) A computer system comprising a processor for executing computer programs and program memory holding a computer program product according to claim 4. ### Claim 6 (depends on 5) A treatment planning system for calculating radiation therapy treatment plans, comprising a computer system according to claim 5. ### Claim 7 A method of training a prediction model for use in radiation therapy treatment planning, the training of the prediction model comprising the following steps: a. obtaining a set of patient data for a previous patient, the set of patient data comprising at least one planning image; b. obtaining a plan generation protocol; c. obtaining a protocol plan of the previous patient by applying the plan generation protocol to the set of patient data; d. calculating at least one protocol dose volume histogram (DVH) curve for the previous patient, each protocol DVH curve relating a dose distribution for the protocol plan of the previous patient to a volume of a structure in a structure set associated with the at least one planning image; e. obtaining a clinical DVH curve associated with each protocol DVH curve, each clinical DVH curve relating a clinical dose distribution for the previous patient to the volume; f. performing steps a-e for each of a number of previous patients to obtain a number of pairs including a protocol DVH curve and an associated clinical DVH curve; g. selecting a subset of the pairs of DVH curves; and h. training a machine learning model using the pairs of protocol DVH curve and associated clinical DVH curve of the previous patients, the machine learning model being configured to take protocol DVH curves of a current patient as input and return estimated clinical DVH curves as input, the process of training the machine learning model involving adjusting system parameters of the machine learning model toward minimization of a deviation between estimated clinical DVH curves and true clinical DVH curves for the previous patients. ### Claim 8 (depends on 7) The method according to claim 7, wherein the selected subset comprises all DVH pairs. ### Claim 9 (depends on 7) The method according to claim 7, wherein according to a functional relationship between protocol DVH and estimated clinical DVH, the estimated clinical DVH curve is described by a sequence of coordinates pairs (x, y), each coordinate pair (x, y) being calculated according to one of the following: setting y to a fixed value Y and x to a value in a range of attained dose values at a cumulative volume of Y for the clinical DVH curves in the pairs of DVH curves among the selected subset of pairs of DVH curves that are associated with the structure; or setting x to a fixed value X and y to a value in a range of attained cumulative volume values at a dose of X for the clinical DVH curves in the pairs of DVH curves among the selected subset of pairs of DVH curves that are associated with the structure. ### Claim 10 (depends on 9) The method according to claim 9, wherein the value in the range of attained dose values or the range of attained cumulative volume values for the selected subset of pairs of DVH curves is one of the following: a generalized mean value, including the arithmetic mean, quadratic mean, minimum, or maximum; or a percentile value, including the median. ### Claim 11 (depends on 7) The method according to claim 7, wherein the selected subset of pairs of DVH curves is determined as a subset of pairs of DVH curves where a distance between the protocol DVH curve in the pair of DVH curves and a current protocol DVH curve of a current patient is less than or equal to a tolerance value. ### Claim 12 (depends on 7) The method according to claim 7, wherein the selected subset of pairs of DVH curves is determined as a subset of pairs of DVH curves where a distance between the protocol DVH curve in the pair of DVH curves and a desired DVH curve is greater than or equal to a distance between a current protocol DVH curve of the current patient and the desired DVH curve. ### Claim 13 (depends on 12) The method according to claim 12, wherein the desired DVH curve for a structure corresponds to a uniform dose at a prescription dose level if the structure is a target and a uniform dose of zero otherwise. ### Claim 14 (depends on 7) The method according to claim 7, further comprising; the step of simplifying each DVH curve in the selected subset of pairs of DVH curves before defining a functional relationship, by mapping the curve to a space of reduced dimension by application of a feature extraction method, comprising: a linear decomposition method, including principal component analysis or singular value decomposition; a nonlinear manifold learning method; or curve fitting of some form of basis functions, including Gaussian functions or polynomial functions. ### Claim 15 (depends on 7) The method according to claim 7 wherein a functional relationship from protocol DVH to estimated clinical DVH is based on a regression model, the regression model being generated by application of a machine learning algorithm on the selected subset of pairs of DVH curves, or a simplified set of such curve pairs obtained by feature extraction. ### Claim 16 (depends on 7) The method according to claim 7, further comprising refining the prediction model over time by updating the set of patient data for previous patients and re-training the prediction model.