Positioning system for radiation therapy

Abstract

A system (1) for positioning a patient for radiation therapy according to patient-specific data including a first three-dimensional image (e.g., a pCT image) of the patient including tissue label data and dose specification data is provided, the system (1) including an imaging device (2), a positioning device (3) and an optimization controller (4). The imaging device (2) is configured to provide a second three-dimensional image (e.g., an rCT image) of the patient including a designated portion of the patient to be treated. The positioning device (3) is provided to hold the patient in a variable position and / or orientation within a beam of radiation for precise treatment of the designated portion of the patient. The optimization controller (4) includes a dose-based control module configured to provide registration control data (ΔP) to guide the positioning device (3) such that an actual applied treatment dose optimally matches the planned treatment dose.

Description

[Technical field]

[0001] Field The present invention relates to a positioning system for radiotherapy.

[0002] The present invention further relates to a training system.

[0003] The present invention further relates to a training method. [Background technology]

[0004] background Radiation therapy (also called radiation therapy) is a cancer treatment that uses high doses of ionizing radiation to kill cancer cells and shrink tumors. In preparation for treatment, radiation oncologists working with medical physicists and radiation therapists design treatment plans that aim to result in effective irradiation of malignant areas while minimizing exposure to healthy tissue.

[0005] Radiation oncologists base their treatment plans on the location of malignant regions from 3D / 4D CT images of the patient acquired prior to treatment. In this context, it is noted that throughout this specification the term "image" is used to denote a three-dimensional image data set or scan acquired from a three-dimensional imaging method. Non-limiting examples of three-dimensional imaging methods are CT, CBCT, PET, MRI and synthetic CT. During the course of treatment, which typically includes multiple treatment sessions, the actual situation may differ from the situation observed in the 3D / 4D images. This may be due to various causes such as the patient's position, changes in the shape of the tumor, the patient's weight gain and loss, and internal motion resulting from breathing and the patient's heartbeat.

[0006] Haehnle et al, 2017 Phys.Med.Biol.62 165 describes a method for interactive multi-objective dose-guided patient positioning. The method proposed herein aims to reposition the patient such that an optimal dose distribution (i.e., a dose distribution that best matches the objective of effective irradiation of malignant regions) is achieved while sparing healthy regions (especially organs at risk (OARs)). The method proposed herein includes a first step in which a 3D cone-beam computed tomography (CBCT) scan is acquired with the patient in the treatment position. In the first step, a strict alignment is calculated in which the patient's anatomy in the original image best matches the anatomy in the CBCT scan. In a second step, the dose distribution is calculated, which would be realized when applying the original treatment plan to the patient with this rigid realignment and for alignments different from this alignment in the space C of accessible shifts. Furthermore, the dose distributions for intermediate alignments are calculated by interpolation. It is a drawback of the known method that the dose calculations for various alignments impose a large computational load and involve a large calculation time. Summary of the Invention [Means for solving the problem]

[0007] overview It is an object of the present invention to provide an improved positioning system which requires less computer resources.

[0008] According to this, an improved positioning system as defined in claim 1 is provided.

[0009] The improved system claimed herein positions a patient in a beam of radiation for treatment according to patient-specific data, including a first image of the patient. The first image is typically a 3D / 4D image acquired by a three-dimensional imaging method as described above (e.g., acquired by a CT and / or less frequently an MRI or PET / CT scanner). The patient-specific data is provided by a specialist (e.g., an oncologist) for example by using a planning tool. During the treatment planning process, the target and organs at risk are segmented, an appropriate number of beam orientations or arc lengths are established for the treatment isocenter, and based on the physician-prescribed dose to the tumor and the objectives of the organs at risk, the plan is optimized and calculated to determine the best dose distribution for the specific patient. The treatment plan is typically delivered over several weeks (5 days per week) of treatment.

[0010] The improved system further includes an imaging device configured to provide a second image (scan) of the patient including the part of the body to be treated. And the imaging device for providing the second image is typically a 3D imaging device such as a CT scanner or an MRI scanner. The second image provides more current information about the location and size of tissues and structures within the patient's body (preferably immediately prior to the treatment session).

[0011] The improved system further includes a positioning device, such as a treatment couch, for holding the patient in a variable position and / or orientation of the beam of radiation for precise treatment of a specified portion of the patient. The positioning device is controlled by an optimization controller included in the improved system. The optimization controller is configured to provide registration control data (i.e. control data for controlling the position and / or orientation of the positioning device to guide the positioning device based on the first image and the second image). The optimization controller includes a dose-based control module configured to calculate an optimized re-positioning of the patient by performing a dose-guided optimization. For this purpose, the dose-based control module is configured to perform at least the following operations:

[0012] The dose-based control module performs a neuromorphic data processing operation on the first image data acquired from the first image to render the first feature map data. The dose-based control module also performs the same neuromorphic data processing operation on the second image data acquired from the second image to render the second feature map data. The dose-based control module then concatenates the first feature map data and the second feature map data to provide concatenated feature map data. The dose-based control module then performs a neuromorphic data processing operation on the concatenated feature map data to provide a correction vector that indicates the patient's requirement registration for dose-based optimization. The concatenated feature map data is, for example, a feature map that includes both the first feature map data and the second feature map data. Due to the fact that the steps of the method are performed in this order (i.e., the first respective feature map data is extracted from each image, and then the feature map data so obtained are concatenated), it is believed possible that the same neural network architecture can be trained and used with various types of images, such as CBCT / MRI. For this reason, registration between different image types (CT+CBCT or CT+MRI) can be performed by the same neural network.

[0013] As defined in more detail below, the dose-based control module defined for the improved system can be easily trained to perform registration in an efficient manner.

[0014] In one embodiment, the improved system further includes a geometry-based control module and a transformation module. The geometry-based control module is configured to perform a geometrical rigid registration to provide a first estimate of the patient's required registration based on the first image and the second image. This geometrical rigid registration is based on the registration of rigid structures in the patient's body, such as bones, or on the registration of markers that are intentionally introduced when planning the treatment. This geometrical rigid registration would be sufficient in a hypothetical case where no morphological changes occur in the patient's body. However, in practice, such morphological changes do occur due to, for example, changes in body weight and changes in the size of the tumor to be treated. Therefore, the geometry-based control module is provided in addition to (and not as a replacement for) the dose-based control module. However, the registration provided by the geometry-based control module is a first approximation that serves as a starting point for the dose-based control module. To compensate for the corrections already provided by the first estimate, a transformation module is provided to transform either the first image or the second image to compensate for the difference in registration dictated by the first estimate. The dose-based control module in this embodiment of the improved system is configured to determine an estimate of the required additional registration of the patient based on the first image and the second image, one of which is transformed. For example, the transformation module transforms the second image according to a first approximation as if the patient's registration had already been corrected according to the first approximation, and the dose-based control module determines an estimate of the required additional registration correction based on the first image and the transformed second image. Alternatively, the transformation module transforms the first image according to the inverse of the transformation defined by the first approximation, and the dose-based control module determines an estimate of the required additional registration correction based on the inverse transformed first image and the second image. It is an advantage of the embodiment of the improved system that the dose-based control module further includes the above-mentioned geometry-based control module and transformation module, which can be more easily trained. The geometry-based control module and transformation module can be implemented in a straightforward manner requiring only modest computational effort.

[0015] In some embodiments of the improved system, the dose-based control module includes first and second mutually identical convolutional neuromorphic processing branches, a neuromorphic concatenation stage and a fully-connected neuromorphic stage.

[0016] In these embodiments, the first convolutional neuromorphic processing branch is configured to receive first image data obtained from a first image and to provide first feature map data based on the received first image data.

[0017] The second convolutional neuromorphic processing branch is configured to receive second image data obtained from the second image, and to provide second feature map data based on the received second image data. It should be noted that instead, a single convolutional neuromorphic processing unit may be used in a time-shared manner for the computation of the first feature map data and the second feature map data from the first and second image data.

[0018] The neuromorphic concatenation stage is configured to receive the first feature map data and the second feature map data, and to provide concatenated feature map data based on the first feature map data and the second feature map data.

[0019] The fully-connected neuromorphic stage is configured to receive the concatenated feature map data and to provide a correction vector that indicates a correction in the registration.

[0020] In some exemplary embodiments, the mutually identical convolutional neuromorphic processing branches each include multiple stages having a convolutional neuromorphic layer. Typically, the mutually identical convolutional neuromorphic processing branches or a single convolutional neuromorphic processing branch applied in a time-shared manner includes up to 10 stages (e.g., 5 stages).

[0021] In addition to convolutional neuromorphic layers, these stages may include one or more of a batch normalization layer, an activation layer (e.g., a ReLU layer (e.g., a leaky ReLU activation layer)), a pooling layer (e.g., a max pooling layer), etc.

[0022] An embodiment of the improved system further includes a third control module configured to calculate a correction vector for dose guided optimization by applying a gradient descent algorithm, and in this embodiment, the dose based control module is optionally combined with the geometry based control module.

[0023] In one example of this embodiment, the third control module includes the following components: a) a transformation vector modifier configured to generate a set of modified transformation vectors for each transformation vector at its input; b) an image transformation module configured to provide a set of modified second images by transforming the second image according to each modified transformation vector of the set of modified transformation vectors. c) a dose simulation module configured to provide an estimated spatial dose distribution for each corrected second image. d) A dose volume histogram calculation module that calculates the resulting expected dose volume histogram for each estimated spatial dose distribution. e) An evaluation module for calculating a quality measure indicating the expected quality of the treatment based on the calculated expected dose-volume histograms for each estimated spatial dose distribution. f) A gradient descent module that calculates a single next iterated transformation vector based on each value of expected quality determined for the modified transformation vector of the set of modified transformation vectors.

[0024] In operation, the third control module iteratively calculates the transformation vector until further iterations do not result in a substantial increase in the quality measure. Alternatively, a predetermined maximum number of iterations may be defined. As yet another alternative, the iterations stop as soon as one of these two conditions is met.

[0025] The simulation module in this example may estimate the spatial dose distribution by Monte Carlo simulation.

[0026] The quality measure may additionally be based on an estimate of normal tissue complication probability (NTCP), which indicates the probability that normal tissues are affected by the treatment. In one embodiment, the third module further includes a treatment effect calculation module that provides an indication of this probability associated with the spatial dose distribution estimated by the dose simulation module. In this calculation, the treatment effect calculation module may take into account one or more of RT structure information and information on prognostic factors. In this embodiment, the evaluation module is configured to calculate the quality measure based on the calculated expected dose-volume histogram and based on an estimate of the normal tissue complication probability.

[0027] According to a second aspect of the present disclosure, there is provided a training system for training a dose based control module, the system including: a) A first input for receiving a first image of a patient, the first image including tissue label data and dose specification data. b) A second input for receiving a second image of the patient, the second image including the designated portion of the patient being treated and taken at a later time than the first image. c) A third input for receiving a Golden Truth (GT) indicator of a transformation vector defining a correction of the patient's position and / or orientation such that the dose delivered to the patient best approximates the dose originally intended according to the first image. d) a transformation vector generator unit for generating a plurality of transformation vectors. e) an image transformation unit for obtaining a respective transformed second image by applying the transformation vector to the second image; f) an additional unit for calculating, for each transformed second image, a corresponding GT correction vector from the GT designator of the second image and the transformation vector (from which the transformed second image was acquired); g) A loss calculation module for determining a loss for each transformed second image based on a difference between the GT correction vector and the correction vector estimated by the dose-based control module when the first image and the transformed second image are provided as input. h) an update module for updating parameters of the dose based control module based on the calculated losses.

[0028] The update module is configured to update parameters of the dose based control module, for example by a back-propagation procedure.

[0029] According to a third aspect of the present disclosure, there is provided a method for training a neuromorphic dose-based control module, the method comprising the following operations. a) Providing a first image of a patient including tissue label data and dose specification data. b) Providing a second image of the patient including a designated portion of the patient to be treated, and also providing a golden truth (GT) designator of a transformation vector (xc, yc, zc) that defines a correction of the patient's position and / or orientation such that the dose delivered to the patient best approximates the dose originally intended according to the first image. c) generating an extended set of second images by applying mutually different geometric transformations (e.g. translation and / or rotation in three-dimensional space to the second images) and calculating for each species of the extended set of second images a golden truth for the required correction from the GT indicator of the second image and the transformation by which this species was obtained from the second image. d) calculating, by the dose based control module, a respective output vector for each pair of a first image and a species of a second image of the expanded set, the output vector indicating a correction to the position and / or orientation of the patient; e) For each species of second images of the augmented set, calculating a loss from the difference between the correction defined by the respective output vector and the respective GT. f) Updating parameters of the dose-based control module to reduce losses.

[0030] The training system according to the second aspect and the training method according to the third aspect enable efficient training of the dose-based control module in that for each manually labeled second image, multiple transformed second images used for training are automatically generated.

[0031] It should be noted that an exemplary embodiment of the improved position correction system including a third control module is configured to update parameters of the dose-based control module based on a loss defined by a difference between a correction vector determined by the third control module and a correction vector determined by the dose-based control module. The loss defined by the difference may be, for example, a distance measure (e.g., a block-based distance measure (L1), a Euclidean distance measure (L2), or a linear distance measure (L3)). oc distance measure). This allows further training of the dose-based control module to be done even when it is already in use.

[0032] BRIEF DESCRIPTION OF THE DRAWINGS These and other aspects are illustrated in greater detail with reference to the accompanying drawings. [Brief description of the drawings]

[0033] [Figure 1] 1 illustrates generally one embodiment of a system for positioning a patient for treatment within a beam of radiation from a radiation source. [Diagram 2] Another embodiment is shown. [Diagram 3] 1 shows the components of the embodiment. [Figure 4] 1 shows an exemplary portion of the component in further detail. [Diagram 5] 1 illustrates generally one embodiment of a method for training a neuromorphic dose-based control module. [Figure 6]1 illustrates generally one embodiment of a training system for training a dose based control module. [Figure 7] 2 illustrates another embodiment of a positioning system. [Figure 8] 1 shows exemplary components of the embodiment in greater detail. [Figure 9] 2 illustrates a schematic diagram of another embodiment of the system; [Figure 10A] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 10B] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 11A] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 11B] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 11C] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 11D] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 12A] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 12B] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 12C] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 12D] We show the dates of experiments obtained in the process of training the neural network in Fig. 4 . [Figure 13] Other experimental data is shown. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0034] Detailed Description of the Embodiments Fig. 1 shows a schematic diagram of a system 1 for positioning a patient for treatment in a beam of radiation from a radiation source 5 (e.g. a photon or proton radiation source according to patient-specific data). The patient-specific data includes a first image pCT of the patient, which includes tissue label data and dose specification data. As shown in Fig. 1, the system includes an imaging device 2, a positioning device 3 and an optimization controller 4.

[0035] A positioning device 3, such as a robotic couch, is provided for holding the patient in a variable position within a beam of radiation from a radiation source 5 for precise treatment to designated parts of the patient.

[0036] The patient-specific data is provided by the oncologist prior to treatment of the patient and typically defines a series of treatment sessions over a period of time, according to which the radiation source 5 is configured to emit a controlled beam of radiation having a predetermined distribution for a predetermined period of time.

[0037] This requires that the patient be properly registered, i.e. the patient must be properly positioned within the beam of radiation. For this purpose, the imaging device 2 is configured to provide a second image rCT of the patient comprising a designated part of the patient to be treated.

[0038] The optimization controller 4 is configured to provide positioning control data ΔP for guiding the positioning device 3 based on the first image pCT and the second image rCT.

[0039] One approach is to register the patient so that the positions of rigid structures within the body (such as bone tissue or markers identified in the second image) match those in the first image, pCT.

[0040] However, simply positioning the patient in a precise orientation as predicted by the oncologist during the treatment preparation is not always sufficient. This is due to the fact that the distribution of tissues in the patient's body may have changed in the time interval between the date of treatment preparation and the date of the treatment session. This may be the case, for example, if the patient has lost or gained weight or if the tumor to be treated has shrunk or instead increased in size.

[0041] As shown in more detail in FIG. 2, to take such morphological changes into account, the optimization controller 4 includes a dose-based control module 41 configured to calculate the patient registration such that the applied dose optimally matches the dose prescribed by the oncologist in the treatment plan.

[0042] As shown in Figure 3, in operation, the dose-based control module 41 performs neuromorphic data processing operations on first image data acquired from a first image pCT to render first feature map data F1. In operation, the dose-based control module 41 also performs the same neuromorphic data processing operations on second image data acquired from a second image rCTq to render second feature map data F2. The first image pCT and the second image rCTq include voxelized 3D x-ray data within a DICOM file of the CT modality.

[0043] Following these operations, the dose-based control module 41 concatenates the first feature map data F1 and the second feature map data F2 to provide concatenated feature map data Fc. The dose-based control module 41 then performs another neuromorphic data processing operation on the concatenated feature map data Fc to provide a correction vector Pd that indicates the required repositioning of the patient for dose-based optimization.

[0044] In one embodiment, the dose-based control module 41 is configured to perform the prescribed operations by a suitably programmed general-purpose processor. In the embodiment of Figure 3, the dose-based control module 41 includes a first convolutional neuromorphic processing branch 411, a second convolutional neuromorphic processing branch 412 that is identical to the first convolutional neuromorphic processing branch 411, a neuromorphic concatenation stage 413 and a fully connected neuromorphic stage 414.

[0045] The first convolutional neuromorphic processing branch 411 is configured to receive first image data acquired from the first image pCT and to provide first feature map data F1 based on the received first image data. The second convolutional neuromorphic processing branch 412 is identical to the first convolutional neuromorphic processing branch 411 and is configured to receive second image data acquired from the second image rCT and to provide second feature map data F2 based on the received second image data. The neuromorphic concatenation stage 413 is configured to receive the first feature map data F1 and the second feature map data F2 and to provide concatenated feature map data Fc based on the first feature map data and the second feature map data. The fully connected neuromorphic stage 414 is configured to receive the concatenated feature map data Fc and to provide an output vector Pd indicating a correction in the registration.

[0046] Referring again to FIG. 2, it is shown that the system for positioning 1 further includes a geometry-based control module 42 and a transformation module 43 .

[0047] In operation, the geometry-based control module 42 performs a geometry-rigorous registration to provide a first estimate ΔPg for the patient's requested repositioning based on the first image pCT and the second image rCT. In the embodiment shown, the module 43 in operation transforms the second image to compensate for the difference in registration indicated by the first estimate Pg. The dose-based control module 41 in operation then determines an estimate ΔPd for the patient's requested additional repositioning based on the first image pCT and the transformed second image rCTg. In the example shown, the dose-based control module 41 adds (adder 45) the first estimate ΔPg from the geometry-based control module 42 and the estimate Pd from the dose-based control module 41 to calculate the positioning control data P for guiding the positioning device.

[0048] In an alternative embodiment, the dose-based control module 41 directly calculates the positioning control data ΔP from the first image pCT and the second image rCT. This has the advantage that the conversion module 43 and the adder 45 are not required, simplifying the system 1. However, the embodiment of the system as shown in Fig. 2 has the advantage that the training of the dose-based control module 41 is simple.

[0049] An exemplary structure of a convolutional neuromorphic processing unit 6 that may be used for both the convolutional neuromorphic processing branch 411 and the convolutional neuromorphic processing branch 412 is shown in FIG. 4. In the illustrated embodiment, the convolutional neuromorphic processing unit 6 includes a plurality of n convolutional neuromorphic processing layers 6_1, ..., 6_n (e.g., several layers to several tens of layers). In the illustrated example, the selected value of n is equal to 5. Furthermore, in the illustrated example, the convolutional neuromorphic processing layers include a 3D convolutional layer 611, a batch normalization layer 612, an activation layer 613 (e.g., a ReLU activation layer such as a leaky ReLU activation layer) and a 3D pooling layer 614 (e.g., a max pooling layer). As an example, the 3D convolutional window of the 3D convolutional layer 611 has a size of 3x3x3 voxels, and the window used by the 3D pooling layer 614 has a size of 2x2x2 voxels. In this example, the window size is the same for all convolutional neuromorphic processing layers, but this is not necessarily the case. In the example shown, the 3D convolutional layer 611 of the first convolutional neuromorphic processing layer 6_1 has eight filters, and the 3D convolutional layer 611 of each subsequent convolutional neuromorphic processing layer has twice as many filters as the above filters. This causes the convolutional neuromorphic processing unit 6 to provide a 128-dimensional feature vector at its output.

[0050] 5 shows a schematic diagram of a method for training the neuromorphic dose-based control module 41. As shown therein, the method includes the following operations.

[0051] First three-dimensional image data (eg planning CT image data pCT of a given planning CT image) is provided to a first convolution neuromorphic processing branch 411 in step S1.

[0052] Second 3D image data (e.g., from predetermined repeated CT image data) are provided in step S2, and an extended set of predetermined second image data {rCT1,...,rCTn} is obtained in step S3 by applying mutually distinct transformations {ΔP1,...,ΔPn} in 3D space to the predetermined repeated CT images rCT. In addition, golden truths {ΔPg1,...,ΔPgn} for the required corrections for each species of the extended set of second images are calculated from the GT indicators of the second images and the transformations applied to that species.

[0053] In another example, the three-dimensional image data is synthetic CT (sCT) image data (eg, based on CBCT or MRI).

[0054] Each of the second image data {rCT1, . . . , rCTn} is then provided to a convolution neuromorphic processing branch 412 in step S4.

[0055] As a result, the dose-based control module 41 calculates, for each pair of first image data pCT and a species rCTi of second image data of the extended set, an output vector {ΔPo1, . . . , ΔPon} indicating a correction during registration.

[0056] In step S5, for each pair i, the loss Li is calculated from the difference between the correction defined by the output vector ΔPoi calculated by the dose-based control module 41 and the respective GT ΔPgi of that pair.

[0057] Next, the weights of the first and second mutually identical convolutional neuromorphic processing branches 411, 412, the neuromorphic concatenation stage 413 and the fully-connected neuromorphic stage 414 are updated in step S6 (e.g., by using backpropagation) to minimize the loss Li.

[0058] FIG. 6 shows a schematic of a training system for training the dose-based control module 41. As shown in FIG. 6, the training system includes a first input IP for receiving a first image of the patient (in this case a planning CT image pCT including tissue label data and dose specification data). The training system also includes a second input IR for receiving a second image rCT of the patient. The second image rCT includes a specified portion of the patient to be treated and is taken at some subsequent time point. An oncologist H determines which corrections (e.g. patient position corrections (xc, yc, zc)) are required to optimally deliver the dose as originally intended by the treatment plan. A transformation vector generation unit 10 generates a number of transformations (e.g. translations (xi, yi, zi) that are applied by an image transformation unit 12 to the second image rCT to obtain a transformed second image rCTi. The corrections (xc, yc, zc) are used to calculate a ground truth (GT) of optimal corrections for the second image rCT. Based on this GT value and the known transformation applied to the second image rCT to obtain the transformed second image rCTi, the summing unit 14 calculates the corresponding GT value (xcj, ycj, zcj) of each transformed second image rCTi. The trained dose-based control module 41 compares each transformed second image rCTj with the first image pCT and estimates the required correction vector (xoj, yoj, zoj). The loss calculation module 16 determines the difference between the estimated correction vector (xoj, yoj, zoj) and the corresponding GT value (xcj, ycj, zcj) and outputs a loss value Lj (e.g., a value indicating the distance between the estimated correction vector (xoj, yoj, zoj) and the corresponding GT value (xcj, ycj, zcj) in the space defining the transformation vector). The distance measure can be, for example, Manhattan distance, Euclidean distance or L oc The current network parameters of the dose-based control module 41 are then updated (eg, by using backpropagation) to minimize the loss.

[0059] 7 shows another embodiment of the positioning system further comprising a third control module 44 configured to calculate the full dose guided optimization by applying a gradient descent GD algorithm. The third control module 44 receives an output vector ΔP0 indicating the corrections in registration as calculated by the dose-based control module 41. In an alternative embodiment, the input of the third control module 44 is the output P calculated in the embodiment shown in FIG.

[0060] 8 shows an embodiment of the third control module 44 in more detail. As shown in the embodiment, a transformation vector corrector 440 is provided which generates a set of modified transformation vectors {ΔPj1, ..., ΔPjn} for each transformation vector ΔPj at its input. Thus, for a transformation vector P0 received from the dose-based control module 41, the transformation vector corrector 440 generates a set of modified transformation vectors {ΔP01, ..., ΔP0n}. The image transformation module 441 provides a set of transformed second image data {rCTj1, ..., rCTjn}. Each type of modified second image data rCTjk is obtained by transforming the second image data (rCT) according to the respective modified transformation vector ΔPjk.

[0061] Based on a given radiation therapy plan (RT plan), the dose simulation module 442 then provides an estimated spatial dose distribution D for each corrected repetitive CT image data r. The simulation module 442 uses, for example, Monte Carlo simulation to estimate the spatial dose distribution D.

[0062] For each estimated spatial dose distribution Djk, the treatment effect calculation module 443 calculates an indicator NTCPjk of the effect of the estimated spatial dose distribution on the patient. This calculation takes into account RT configuration information and information on prognostic factors.

[0063] In addition, the dose volume histogram (444) calculation module calculates RT structure information for each estimated spatial dose distribution Djk by considering the expected dose volume histogram DVHjk.

[0064] The assessment module 445 calculates an overall quality measure Qjk indicating the expected quality Qjk of the treatment based on the calculated indicators NTCPjk and DVHjk of the estimated spatial dose distribution Djk.

[0065] The gradient descent module 446 calculates a single next iterative transformation vector Pj+1. To this end, the gradient descent module 446 determines which direction in the space in which the transformation vectors are defined produces the greatest improvement in expected quality.

[0066] As an example, in the case where the translation vector ΔPj includes components (Δxj, Δyj, Δzj), the translation vector corrector 440 may generate a set of modified translation vectors {ΔPj, ..., ΔPjn}), where the modified translation vector has components (Δxj+dxk, Δyj+dyk, Δzj+dzk), where the displacements dxk, dyk, dzk are uniformly distributed in a volume centered on the origin in the space of translation vectors. The gradient descent module 446 may then select one of the modified translation vectors ΔPjk corresponding to the highest value of Qjk as the next iterative translation vector ΔPj+1.

[0067] In the above embodiments, the transformation vector space is assumed to be a three-dimensional space. The transformation vectors in this case are three-dimensional vectors in mutually orthogonal spatial directions x, y, z. In these embodiments, the dose-based control module (41) is configured to provide positioning control data P (a three-dimensional control vector that causes the positioning device to assume a corrected position in the three-dimensional space). In practice, this already achieves a significant improvement. In other embodiments, the transformation vector is a higher-dimensional vector that not only contains components in the spatial directions x, y, z, but also one or more components related to the orientation in space. In these other embodiments, the dose-based control module 41 is configured not only to provide positioning control data for causing the positioning device to assume a corrected position in the three-dimensional space, but also to provide orientation control data for causing the positioning device to assume a corrected orientation in the three-dimensional space. This allows further improvements in the dose control.

[0068] Fig. 9 shows a schematic representation of another embodiment of the system. As in the embodiment of Fig. 2, a geometry-based control module 42 is included. In this embodiment, the geometry-based control module 42 provides its calculated correction vector ΔPg to an input of the dose-based control module 42 as well as to an input of the control module 44. Based on the correction vector ΔPg provided at their respective inputs, the dose-based control module 42 and the further control module 44 each calculate a correction vector P0, ΔPg for the dose-based positioning, respectively. A quality assessment module 46, additionally included in this embodiment of the system, compares the correction vector ΔPg calculated by the further control module 44 with the correction vector ΔP0 calculated by the dose-based control module 42. Based on this comparison, the quality assessment module 46 determines a final correction vector ΔPf, and the system positions the patient according to the final correction vector ΔPf thus obtained.

[0069] Experimental Results In a first experiment, the neural network shown in and described with reference to FIG. 3 was trained for the purpose of providing a geometry-based correction vector.

[0070] The first study was performed with one patient / image pair (pCT+rCT). The rCT images were enhanced 1000 times by applying a composite shift of -10 to 10 cm across the x, y and z directions.

[0071] The results of the training process are shown in Figures 10A and 10B.

[0072] 10A and 10B show the training loss expressed as mean square error (MSE) and mean absolute error (MAE), respectively, as a function of the number of training epochs. Once training was complete, the neural network estimated a geometry-based correction vector with a mean absolute error of less than 1 mm for all axes.

[0073] In the second experiment, a second training / validation set was used, where data from 45 patients were used. Each patient had 1 pCT and 5 repeat CT for a total of 225 image pairs. These images were then enhanced 5 times with a synthetic shift of -5 to +5 cm in the x, y, and z axes. The total number of images used was 1125 images.

[0074] The results of the training process are shown in Figures 11A-11D.

[0075] 11A and 11B show the training loss expressed as the mean squared deviation (MSE) and the mean absolute error (MAE), respectively, as a function of the number of training epochs during a short training period (100 epochs).

[0076] 11C and 11D show the training loss expressed as mean squared deviation (MSE) and mean absolute error (MAE), respectively, as a function of the number of training epochs during the long training period (600 epochs). Upon completion of the long training period, the neural network estimated a geometry-based correction vector with a mean absolute error of less than 3 mm for all axes.

[0077] In a third experiment, a neural network as shown in and described with reference to FIG. 3 was trained for the purpose of providing a dose-based correction vector. In this case, the daily dose is first estimated at the current patient position in the rCT image. The network is then trained with the planned dose and the daily dose to optimize the previously obtained translational shift by gradient descent.

[0078] The results of the training process are shown in Figures 12A-12D.

[0079] 12A and 12B show the training loss expressed as the mean squared deviation (MSE) and the mean absolute error (MAE), respectively, as a function of the number of training epochs during a short training period (30 epochs).

[0080] 12C and 12D show the training loss expressed as mean squared deviation (MSE) and mean absolute error (MAE) as a function of the number of training epochs during the long training period (80 epochs), respectively. Upon completion of the long training period, the neural network estimated dose-based correction vectors with mean absolute errors of less than 1.2 mm for all axes. In any case, a relatively modest number of training examples is sufficient for a network that is already trained to compute geometry-based corrections (since training does not need to start from scratch).

[0081] Figure 13 shows a comparison of the calculation speeds involved in the three approaches to achieve dose-based optimization. The comparison was performed on 10 head and neck cancer patients previously treated with proton radiotherapy at GPTC. For each patient, these calculations require imaging data (planning CT and repeat CT, segmented files) and treatment data (radiotherapy plan) in DICOM format.

[0082] The three methods include: 1) The RS-MC dose engine from RaySearch's Treatment Planning System (TPS) software called RayStation. 2) Gradient descent using a fast Monte Carlo code developed at Massachusetts General Hospital (MGH) called gPMC, a GPU-based dose engine for proton therapy focused solely on dose and LET calculations. 3) Calculation by CNN as defined with reference to Figure 4.

[0083] For each patient, the left column indicates the calculation time involved.

[0084] Calculations with methods 1 and 2 were performed on a GPU-based clinical RayStation customer configuration that included: ○ 12 physical cores dual socket Intel 2.6GHz CPU ○NVIDIA RTX8000 GPU ○48GB DDR3 ECC RAM ○ 2 x 150 SSDs in RAID 1

[0085] The CNN-based optimization was performed using the CNN in Figure 4 implemented on the clinical QA platform server at GPTC, which includes: ○24 physical core dual socket Intel 3.0GHz CPU ○Two NVIDIA GTX 1080 GPUs ○64GB DDR4 RAM ○ 4 x 250 GB SSD

[0086] FIG. 13 shows the computational time involved in each of methods 1), 2) and 3) per patient.

[0087] The results are summarized in the table below.

[0088] [Table 1]

[0089] In summary, it can be seen that a very fast dose-based optimization is achieved by the CNN-based method (3). Even if the CNN-based calculations are performed by a GPU-based clinical RayStation customer configuration, this will still be significantly faster than the RS-MC-based method. The optimization results obtained by this method (3) can be independently verified by the gMC-based method (2), which is also substantially faster than the RS-MC-based method. Thereby, the verified optimization results can also be achieved substantially faster than possible by the RS-MC-based method.

Claims

1. 1. A system (1) for positioning a patient for treatment within a beam of radiation according to patient-specific data including a first image (pCT) of the patient including tissue label data and dose specification data, the system comprising: an imaging device (2) configured to provide a second image (rCT) of the patient including a designated portion of the patient to be treated; a positioning device (3) for holding the patient in a variable position and / or orientation within the beam of radiation for precise treatment of a designated portion of the patient; and an optimization controller (4) configured to provide registration control data (ΔP) for guiding the positioning device (3) based on the first image (pCT) and the second image (rCT); The optimization controller (4) includes a dose-based control module (41) configured to calculate an optimized registration of the patient by performing a dose-guided optimization; The dose based control module: performing a neuromorphic data processing operation on first image data acquired from said first image (pCT) to render first feature map data (F1); performing said neuromorphic data processing operation on second image data acquired from said second image (rCT) to render second feature map data (F2); concatenating said first feature map data (F1) and said second feature map data (F2) to provide concatenated feature map data (Fc); and performing another neuromorphic data processing operation on the concatenated feature map data (Fc) to provide a correction vector indicative of the patient's required registration for dose-based optimization, the system comprising: The system includes another control module (44) configured to calculate correction vectors for dose guidance optimization by applying a gradient descent (GD) algorithm; The further control module (44) comprises: a translation vector corrector (440) that generates a set of corrected translation vectors ({ΔPj1, ..., ΔPjn}) for each translation vector (ΔPj) at its input; an image transformation module (441) for providing a set of modified second images ({rCTj1, ..., rCTjn}) by transforming the second image (rCT) according to each modified transformation vector (ΔPjk) of the set of modified transformation vectors ({ΔPj1, ..., ΔPjn}); a dose simulation module (442) that provides an estimated spatial dose distribution (Djk) for each corrected second image (rCTjk); a dose-volume histogram (444) calculation module that calculates the resulting expected dose-volume histogram (DVHjk) for each estimated spatial dose distribution (Djk); an evaluation module (445) for calculating a quality measure (Qjk) indicative of the expected quality (Qjk) of the treatment based on the calculated expected dose-volume histogram (DVHjk) for each estimated spatial dose distribution (Djk); a gradient descent module (446) that calculates a single next iterative transformation vector (ΔPj+1) based on the respective values ​​of the expected qualities (Qjk) determined for the modified transformation vectors (ΔPjk) of the set of modified transformation vectors ({ΔPj1, ..., ΔPjn}).

2. a geometry-based control module (42) for performing a geometry-rigid registration to provide a first estimate of a desired registration correction (ΔPg) for the patient based on the first image (pCT) and the second image (rCT); 2. The system (1) of claim 1, further comprising a transformation module (43) for transforming either the first image or the second image to compensate for differences in registration indicated by the first estimate (ΔPg), The system (1), wherein the dose-based control module (41) is configured to determine an estimate (ΔPd) of the patient's required additional registration correction based on the first image (pCT) and the second image (rCT), either one of which is transformed.

3. The dose-based control module (41) comprises: a first convolutional neuromorphic processing branch (411) configured to receive the first image data obtained from the first image (pCT) and to provide the first feature map data (F1) based on the received first image data; a second convolutional neuromorphic processing branch (412) identical to the first convolutional neuromorphic processing branch (411), configured to receive the second image data acquired from the second image (rCT) and to provide the second feature map data (F2) based on the received second image data; a neuromorphic concatenation stage (413) configured to receive the first feature map data (F1) and the second feature map data (F2), and to provide the concatenated feature map data (Fc) based on the first feature map data and the second feature map data; 3. The system of claim 1, further comprising a fully connected neuromorphic stage (414) configured to receive the connected feature map data (Fc) and to provide the correction vector (ΔPd) that indicates the correction during registration.

4. 4. The system of claim 3, wherein the mutually identical convolutional neuromorphic processing branches (411, 412) each include a plurality of stages (6_1, ..., 6_n) having convolutional neuromorphic layers (611).

5. 5. The system of claim 4, wherein each stage (6_1, ..., 6_n) of the convolutional neuromorphic processing branch includes a batch normalization layer (612).

6. The system of claim 4 , wherein each stage (6_1, . . . , 6_n) of the convolutional neuromorphic processing branch includes a leaky ReLU activation layer (613).

7. 5. The system of claim 4, wherein each stage (6_1, . . . , 6_n) of the convolutional neuromorphic processing branch comprises a pooling layer (614).

8. The system of claim 5, wherein the pooling layer (614) is a max-pooling layer.

9. The system of claim 1 , wherein the simulation module (442) estimates the spatial dose distribution (Djk) by Monte Carlo simulation.

10. 2. The system of claim 1, further comprising a treatment effect calculation module (443) for providing an estimate of a normal tissue damage probability (NTCP) associated with each estimated spatial dose distribution (D) on the human body by considering one or more of RT structure information and information on prognostic factors, The evaluation module (445) is configured to calculate the quality measure (Qjk) based on the calculated predicted dose-volume histogram (DVHjk) and based on the estimate of the probability.

11. 2. The system of claim 1, configured to update parameters of the dose-based control module (41) based on a loss defined by a difference between the correction vector determined by the other control module (44) and the correction vector determined by the dose-based control module (41).

12. 2. The system of claim 1, wherein the further control module (44) is configured to receive at its input the correction vector (ΔP0) provided by the dose-based control module (41), and the system is configured to position the patient according to the correction vector calculated by the further control module.

13. 3. The system of claim 2, wherein the further control module (44) is configured to receive at its input the correction vector (ΔPg) provided by the geometry-based control module (42), The system further includes a quality assessment module (46) for comparing the correction vector calculated by the further control module (44) with the correction vector calculated by the dose-based control module (42); The system is configured to position the patient according to a correction vector (ΔPf) determined by the quality assessment module (46) based on the comparison.

14. A training system for training a dose-based control module (41), comprising: a first input (IP) for receiving a first image (pCT) of a patient, said first image including tissue label data and dose specification data; a second input (IR) for receiving a second image (rCT) of the patient, the second image (rCT) including a designated portion of the patient to be treated and taken at a later time than the first image; a third input (H) for receiving a ground truth (GT) designator of a transformation vector defining a correction (xc, yc, zc) of the patient's position and / or orientation, such that the dose delivered to the patient best approximates the dose originally intended according to the first image; a transformation vector generator unit (10) for generating a plurality of transformation vectors (xi, yi, zi); an image transformation unit (12) for obtaining each transformed second image (rCTi) by applying the transformation vectors (xi, yi, zi) to the second images (rCT); an addition unit (14) for calculating, for each transformed second image (rCTi), the corresponding GT correction vector (xcj, ycj, zcj) from the GT indicator of the second image (rCTi) and the transformation vector (xi, yi, zi) with which the transformed second image (rCTi) was obtained; a loss calculation module (16) for determining a loss for each transformed second image based on a difference between the corresponding GT correction vector and the correction vector (x, y, zo) estimated by the dose-based control module when the first image and the transformed second image are provided as inputs; an update module (18) for updating parameters of the dose-based control module (41) based on the calculated losses.

15. A method for training a neuromorphic dose-based control module (41), comprising: Providing a first image (pCT) of a patient including tissue label data and dose specification data (S1); providing (S2) a second image (rCT) of the patient including a designated portion of the patient to be treated, along with a ground truth (GT) designator of a transformation vector defining a correction (xc, yc, zc) of the patient's position and / or orientation, so that the dose delivered to the patient best approximates the dose originally intended according to the first image; generating (S3) an extended set of second images ({rCT1, ..., rCTn}) by applying mutually distinct geometric transformations {ΔP1, ..., ΔPn} in three-dimensional space to the second images (rCT), and calculating, for each species of second images of the extended set, the ground truth for the required correction from the GT indicators of the second images and the transformations applied to that species; calculating (S4) by the dose-based control module (41) a respective output vector {ΔPo1, ..., ΔPon} indicating a correction (xc, yc, zc) of the patient position and / or orientation for each pair of the first image (pCT) and a seed (rCTi) of the expanded set of second images ({rCT1, ..., rCTn}); calculating (S5) for each species (rCTi) of the extended set of second images ({rCTi, ..., rCTn}) a loss (Li) from the difference between the correction defined by the respective output vector and the respective GT; updating (S6) parameters of said dose-based control module (41) to reduce said losses (Li).

Images