Automated planned radiation-based therapy

Abstract

Deep learning methods automatically segment at least some breast tissue images, while non-deep learning methods automatically segment organs at risk. Both 3D CT imaging information and 2D orthogonal computed tomography (CT) imaging information can be used to determine the virtual skin volume. The aforementioned imaging information can also be used to automatically determine (205) at least a portion of the patient's body shape. The body shape, along with the virtual skin volume and registration information, can be used as input to automatically calculate (210) the radiotherapy platform trajectory, conflict detection information, and virtual rehearsal information for treatment delivery based on an optimized radiotherapy plan.

Description

[0001] Related applications

[0002] This application relates to two jointly pending and jointly owned U.S. patent applications: No. 16 / 833,778 entitled “AUTOMATICALLY-REGISTERED PATIENTFIXATION DEVICE IMAGES” (Attorney’s File No. 8632-144156-US(2018-061)) and No. 16 / 833,801 entitled “Conflicting METHOD AND APPARATUS TO DERIVE AND UTILIZEVIRTUAL VOLUMETRIC STRUCTURES FOR PREDICTING POTENTIAL COLLISIONS WHENADMINISTERING THERAPEUTIC RADIATION” (Attorney’s File No. 8632-144811-US(2018-065)), both filed on the same date as the U.S. patent application claiming priority to this PCT application, the contents of which are incorporated herein by reference in their entirety. Technical Field

[0003] These teachings generally concern the use of radiation as a treatment method, and more specifically concern the formulation and use of corresponding radiation therapy plans. Background Technology

[0004] The use of radiation to treat medical conditions encompasses known areas of effort in current technology. For example, radiation therapy includes several important components of treatments aimed at reducing or eliminating harmful tumors. Unfortunately, the radiation applied cannot inherently distinguish harmful substances from adjacent tissues, organs, or similar substances desired or even critical for prolonging a patient's survival. As a result, radiation is typically applied in a cautious manner to at least attempt to confine the radiation to a given target volume. So-called radiation therapy planning is used in the aforementioned contexts.

[0005] A treatment plan is used to specify any number of operational parameters relevant to the implementation of the treatment for a given patient and a specific real-world physical radiotherapy platform. This treatment plan is often optimized before use. (As used herein, "optimization" will be understood as referring to improvements on candidate treatment plans without ensuring that the optimized result is actually a uniquely optimal solution.) Multiple optimization methods employ an auto-incrementing methodology that sequentially calculates and tests various optimization results using various automatically modified (i.e., "incremental") treatment optimization parameters.

[0006] Radiation therapy planning workflows typically involve multiple manual and iterative steps. This is especially true when using backpropagation-based techniques such as IMRT and VMAT. Each step / choice made in a series of steps / choices can often influence later stages. Such factors can lead to overly complex planning workflows when only a simple 3-D plan is required (e.g., developing a treatment plan for breast cancer). In particular, physicians must often manually delineate target structures and organs of risk through a time-consuming process, which is also susceptible to observer variability. This variability is further exacerbated by the domain setting and the definition of appropriate objectives for the optimizer that must be manually determined before the optimization process begins.

[0007] Because exploring all or most possible combinations often involves a time-consuming and computationally expensive process, in some application settings, users can choose to utilize simpler methods or readily available, previously known practices. However, making this choice can significantly limit the options available and suitable for a particular patient.

[0008] Other limitations have also been found in the prior art. For example, while a prior art method estimates multiple script plans for different treatment units, those plans are not created with collision avoidance in mind and are generally neither patient-specific and ready for delivery nor properly tailored to a particular patient. Summary of the Invention

[0009] In one aspect, the present invention provides a method for automatically planning treatment volumes based on radiation, as defined in claim 1. Optional features are specified in the dependent claims of claim 1.

[0010] In another aspect, the present invention provides an apparatus for automatically planning treatment volumes based on radiation, as defined in claim 10. Optional features are specified in the dependent claims of claim 10. Attached Figure Description

[0011] The above needs are met at least in part by providing an automated, radiotherapy-based treatment apparatus and method as described in the detailed embodiments below, particularly when considered in conjunction with the accompanying drawings, wherein:

[0012] Figure 1 Including block diagrams configured according to various embodiments of these teachings;

[0013] Figure 2 Including flowcharts of various embodiments configured according to these teachings; and

[0014] Figure 3 This includes flowcharts of various embodiments configured according to these teachings.

[0015] The elements in the figures are shown for simplicity and clarity and are not necessarily drawn to scale. For example, the dimensions and / or relative positions of elements relative to other elements may be exaggerated to aid in understanding the various embodiments of this teaching. Furthermore, common but well-understood elements that are useful or necessary in commercially viable embodiments are often not depicted to facilitate a less obstructed view of the various embodiments of this teaching. An action and / or step may be described or depicted in a particular order of occurrence, and those skilled in the art will understand that such particularity in the order is not actually necessary. The terms and expressions used herein have the ordinary technical meaning consistent with those of those skilled in the art, unless otherwise specified herein. Unless otherwise specifically indicated, the word “or” as used herein should be interpreted as having a separating grammatical structure rather than a connecting grammatical structure. Detailed Implementation

[0016] In general, several embodiments of these various examples are provided for the automated planning of radiation-based treatment volumes for specific patients. As a particularly significant example of these aspects, this can include the automated planning of radiation-based treatment volumes within breast tissue of a specific patient. As a useful illustrative example, multiple or all of the activities supporting the foregoing are facilitated by the use of control circuitry.

[0017] One approach provides access to imaging information for a treatment region, which includes the treatment volume of a specific patient. Control circuitry can then use this imaging information, along with deep learning, to automatically segment at least some breast tissue (and possibly heart) of a specific patient, and, along with non-deep learning, to automatically segment at least parts of some organs of danger (such as, but not limited to, parts of the lungs, spine, and chest wall) to provide automatically segmented patient content. Atlas-based and model-based approaches are examples of two non-deep learning methods. Atlas-based segmentation assumes that a given patient image can be segmented by propagating the structure from a manually segmented atlas. The atlas image is deformed to match the patient image using one or more deformable image registration algorithms, and the structure is propagated using a deformable vector field that maps voxels from the atlas image to voxels in the patient image. Model-based structural segmentation depicts the structure by directly detecting edges and points on the patient image. In these respects, various image processing techniques are frequently used. Manually defining the volume is suitable for the user in many cases. Overall, these methods combine deep learning and density with heuristic search algorithms. The latter is an algorithm sometimes used in existing treatment planning systems. However, the combination of density-based and heuristic tools with deep learning was previously unknown to the applicant.

[0018] The control circuitry can also use the imaging information to automatically determine a virtual skin volume corresponding to at least a portion of the aforementioned breast tissue (specifically, the skin therein). By means of a method, when automatically optimizing a radiotherapy plan for the aforementioned treatment volume of a particular patient, the control circuitry uses automatically segmented patient content and the virtual skin volume as input to provide an optimized radiotherapy plan for that particular patient.

[0019] In one method, the aforementioned imaging information includes three-dimensional computed tomography (CT) imaging information. In another method, the aforementioned imaging information may further include two-dimensional topogram imaging information. In this case, when the control circuit uses the imaging information to automatically determine the virtual skin volume corresponding to at least a portion of the breast tissue of a specific patient, this may at least partially include using both the three-dimensional CT imaging information and the two-dimensional topogram imaging information to determine the virtual skin volume.

[0020] One approach allows the control circuitry to automatically segment at least some cardiac tissue from a specific patient using imaging information along with deep learning (as opposed to non-deep learning). In general, the inventors have determined that some anatomical structures, such as the lungs, spinal cord, bones, and eye structures, can be easily and accurately segmented using standard non-deep learning methods, while other structures, especially those without clearly defined boundaries, are more problematic. In the latter case, deep learning-based methods can produce better results faster.

[0021] In various radiotherapy application settings, the patient is supported by a patient support surface (such as, but not limited to, a treatment bed). In some cases, a patient fixation device may also be used in place of or in combination with the aforementioned to spatially fix a portion of the patient's body, thereby maintaining the body portion in a relatively stable position / orientation. With this in mind, these teachings may further include, through a method, automatically registering at least one of the patient support surface and the patient fixation device to at least some of the imaging information in the imaging information to provide registration information and use that registration information as further input when automatically optimizing the aforementioned radiotherapy plan. As a non-limiting illustrative example of these aspects, the foregoing may include automatically registering a model of at least one of the patient support surface and the patient fixation device to at least some of the imaging information in the imaging information.

[0022] By means of a method, these teachings will be further applied to automatically determining the body outline for at least a portion of a patient using the aforementioned imaging information. For example, in this case, control circuitry can use the body outline, the aforementioned virtual skin volume, and the registration information described above as inputs to automatically calculate the trajectory of the radiotherapy platform, conflict detection information, and virtual rehearsal information for treatment delivery according to the optimized radiotherapy plan.

[0023] This configuration, and through a method, facilitates the automation of several time-consuming steps that currently prevent users from utilizing alternative technologies other than conformal radiotherapy. Through a method, these teachings are used to automatically generate multiple scored plans for various forms of beam-based radiotherapy. Each of these plans may include conflict-free geometry definition, beam energy selection, dosimetry optimization, and dose calculation. It can also be noted that these teachings utilize deep learning for breast CT data segmentation and also play a significant role in improving the creation and utilization of pills, as described below.

[0024] These and other benefits become clearer after a comprehensive review and study of the detailed description below.

[0025] However, before describing the foregoing teachings in more detail, it may be helpful to the reader to first provide a general description of an example of a current reverse planning workflow for treating breast cancer. This example will help illustrate at least some current processes that, while relying on a degree of automation, must frequently invoke human intervention, subjective judgment, and supervision.

[0026] The current example of a reverse planning workflow begins with a CT simulation. This simulation can begin with a scan including a chest panel (which may include triangulation or leveling of the patient), lead lines for the boundaries, and / or a wireframe for the entire breast tissue. Technicians then use CT simulation software to create the necessary 3D images and contours.

[0027] The technician (or another technician) working on a properly equipped contouring workstation imports the aforementioned CT / body content imaging information and automatically contours it. (Volume-modulated arc radiotherapy (VMAT) users can also manually expand the body image to fit the "skin flash" during optimization.) This contouring may include contouring both the left and right lungs, the heart, the contralateral breast, the spinal cord, the lymph node chain, and the patient target volume (PTV).

[0028] Technicians can also add bolus content if they wish to increase the skin dose or as a workaround for skin flashes. (For VMAT, users can add bolus content during optimization to achieve skin flashes and then remove the bolus content before dose calculation.) The latter activity can include defining both the thickness and shape of the bolus (e.g., by selecting a predefined shape or creating a custom shape, for example, one designed to cover the entire irradiated area). (In radiotherapy, a bolus is a substance that, when irradiated, has properties equivalent to a given tissue such as breast tissue. Bolus content is frequently used to reduce or alter the dose to a radiotherapy target. For example, bolus content can be used to compensate for missing or irregularly shaped tissue and / or modify the radiation dose at the skin.)

[0029] The results of the profiling process are then transferred from the profiling process to the external beam planning workbench and the corresponding technician. The external beam planning process typically creates a forward plan (i.e., a baseline dose plan) that includes calculated and normalized parameters for multiple treatment areas. In some cases, this activity may also include converting isodose lines into corresponding structures to obtain an optimized patient treatment volume. The specific calculations and steps for external beam planning will vary in part depending on whether the plan corresponds to an IMRT plan or a VMAT plan. In some cases, the planning process may also include using a skin flash tool to add flash to one or more areas.

[0030] External beam planning processes typically work in conjunction with corresponding optimization processes to iteratively calculate the implemented dose. In any case, the resulting plan is subsequently evaluated by an external beam planning (EB) or plan evaluation (PE) workstation. This evaluation may include assessing the location of the minimum and maximum doses achieved by the plan, the monitoring unit (MU) used (a monitoring unit is a measurement of machine output from a clinical accelerator for radiotherapy, such as a linear accelerator).

[0031] Filled with platform switching and frequent or occasional ongoing human involvement, this entire process, while often yielding useful results, is time-consuming, subject to human vulnerability, and inherently fails to realize the potential synergies of its constituent activities.

[0032] Now refer to the attached diagram, especially in Figure 1 In this context, an illustrative device 100 compatible with several of these teachings will now be presented.

[0033] In a particular example, enabling device 100 includes control circuitry 101. As a “circuit”, control circuitry 101 therefore includes a structure comprising at least one (and typically multiple) conductive paths (e.g., including conductive metals such as copper or silver) that carry electrical power in an ordered manner, and these paths(multiple) will also typically include corresponding electronic components (including, where appropriate, both passive (e.g., resistors and capacitors) and active (e.g., any of the various semiconductor-based devices)) to allow the circuitry to implement the control aspects of these teachings.

[0034] Such control circuitry 101 may include a fixed-purpose hardware platform (including, but not limited to, application-specific integrated circuits (ASICs) (which are custom integrated circuits designed for a specific purpose rather than intended for general-purpose use), field-programmable gate arrays (FPGAs), etc.), or may include a partially or fully programmable hardware platform (including, but not limited to, microcontrollers, microprocessors, etc.). These architectural options for such a structure are known and understood in the art and need not be described further herein. Control circuitry 101 is configured (e.g., by means of corresponding programming that will be well understood by those skilled in the art) to perform one or more steps, actions, and / or functions described herein.

[0035] In this illustrative example, control circuitry 101 is operatively coupled to memory 102. Memory 102 may be integrated into control circuitry 101 or physically discrete from control circuitry 101 (whole or partially) as needed. Memory 102 may also be local to control circuitry 101 (where, for example, both share a common circuit board, chassis, power supply, and / or housing), or may be remote relative to control circuitry 101 (where, for example, memory 102 is physically located in another facility, metropolitan area, or even country compared to control circuitry 101).

[0036] The memory 102 can be used, for example, to non-transitory store computer instructions that, when executed by the control circuit 101, cause the control circuit 101 to act as described herein. (As used herein, the reference to "non-transitory" will be understood to refer to a non-transient state of the stored content (and therefore excludes when the stored content consists only of signals and waves), rather than the volatility of the storage medium itself, and therefore includes both non-volatile memories (such as read-only memory (ROM) and volatile memories (such as dynamic random access memory (DRAM)).)

[0037] In this example, the memory 102 can also be used to store imaging information 103. This imaging information may include information related to an image (such as the "image" itself) of a treatment area for a patient, including a treatment volume 105 for a specific patient. For illustrative purposes, it will be assumed here that the treatment volume 105 is within the patient's breast tissue 104.

[0038] These teachings are flexible in practice and will be applicable to a variety of imaging information. For example, by one method, imaging information 103 includes three-dimensional computed tomography (CT) imaging information provided by a corresponding CT device 106. As another example, instead of or in combination with the foregoing, imaging information 103 may include two-dimensional tomographic localization film imaging information provided by the CT device 106 or a correspondingly suitable imaging device 107. (Tomographic localization film is well understood in the art and is sometimes referred to as a sunnt view or survey view; a tomographic localization film is a two-dimensional image generated by tomography without being reconstructed as a slice.)

[0039] It should also be understood that imaging information 103 may include one or more organs at risk for a particular patient (in Figure 1 The imaging information is provided by organs from the first dangerous organ 108 to the Nth dangerous organ 109 (where "N" is an integer greater than "1"). A dangerous organ is part or all of a non-target organ that is at risk of radiation injury when the treatment volume 105 is treated with radiation. Examples of such dangerous organs (when the treatment volume 105 is within breast tissue 104) include one or both lungs, part of the spine and part of the chest wall, and the heart.

[0040] Furthermore, it will be understood that the imaging information 103 may also include imaging information for the patient support device 110 (such as a so-called treatment bed) and / or one or more patient fixation devices 111 used to hold some part of the patient's body in a fixed position relative to the radiotherapy beam 112.

[0041] If desired, the control circuitry 101 can also be operatively coupled to a network interface (not shown). The control circuitry 101 configured in this way can communicate with other components (both within and outside the device 100) via this network interface. Network interfaces, including both wireless and non-wireless platforms, are well understood in the art and do not require particular detailed description here.

[0042] Alternatively, the control circuit 101 can also be operatively coupled to a user interface (not shown) via an alternative method (instead of or in combination with the aforementioned method). This user interface may include any variety of user input mechanisms (such as, but not limited to, keyboards and keypads, cursor control devices, touch-sensitive displays, voice recognition interfaces, gesture recognition interfaces, etc.) and / or user output mechanisms (such as, but not limited to, visual displays, audio transducers, printers, etc.) to facilitate the receipt of information and / or instructions from the user and / or the provision of information to the user.

[0043] As described in more detail below, control circuitry 101 is at least partially configured to optimize radiotherapy plans, thereby producing one or more optimized radiotherapy plans 113. For example, the optimized radiotherapy plan 113 is used to control a radiotherapy platform 114 that may include a radiation source 115, which may (if desired) be operatively coupled to and responsive to control circuitry 101. Radiation sources are well understood in the art and do not need to be further described herein.

[0044] With this configuration, the corresponding radiation beam 116 emitted by the radiation source 115 can be selectively turned on and off by the control circuit 101. These teachings will also apply to enabling the control circuit 101 to control the relative intensity of the radiation beam 116. The radiotherapy platform 114 can also be configured to move the radiation source 115 during the treatment phase, thus delivering radiation from various directions (“fields”). In this case, the control circuit 101 can also be configured to control this movement.

[0045] As an alternative approach, as illustrated herein, the radiotherapy platform 114 may further include one or more beamforming devices 117. For example, this device 117 is used to modify the radiation beam 116 by shaping it and / or otherwise modulating it, thereby producing a corresponding output radiation beam 112 to which the treatment volume 105 is exposed. Examples of known beamforming devices include, but are not limited to, clips, collimators, and multi-leaf collimators.

[0046] For illustrative purposes, this document assumes that the aforementioned control circuit 101, in conjunction with the application settings described above, implements at least some (and possibly all) of the actions, steps, and / or functions described herein. Figure 2 The process presented is consistent with much of what these teachings contain.200

[0047] At box 201, the process 200 provides access to imaging information (such as the aforementioned imaging information 103) for a treatment area including a treatment volume 105 for a specific patient. This illustrative example assumes that the treatment volume 105 is located within the patient's breast tissue 104. Again, for illustrative purposes, this example assumes that the imaging information 103 specifically includes three-dimensional CT imaging information and two-dimensional tomographic imaging information. In addition to the treatment volume 105 itself, the imaging information 103 may include imaging information that includes at least portions of one or more organs of danger 108, 109, such as the patient's lungs, portions of the patient's spine, and / or portions of the patient's chest wall. Furthermore, in appropriate application settings, the imaging information 103 may include imaging information for one or more of the patient support device 110 and / or one or more patient fixation devices 111.

[0048] One method allows one or more items of imaging information 103 to be captured at particularly needed times (i.e., when preparing and optimizing a radiotherapy plan for a specific patient) and when the patient is on the radiotherapy platform 114 itself. Another method allows one or more items of imaging information 103 to be captured at an earlier time, and possibly when the patient is not on the radiotherapy platform 114 itself.

[0049] At box 202, the process 200 provides, while using imaging information 103, the automatic segmentation of at least some breast tissue 104 of a specific patient using deep learning, and the automatic segmentation of at least portions of some organs of danger 108, 109 using non-deep learning, to provide automatically segmented patient content. This process will also be applicable to the automatic segmentation of any high-density artifacts within the patient's body.

[0050] One method for this activity may include defining the patient's body shape. Alternatively, the activity may include automatically detecting and defining the location of lines and / or radiographic markings on the patient's body surface. Based on this detection and definition, the activity may include automatically removing specific content from 3D CT images.

[0051] Segmentation involves well-understood activities and includes the careful identification of specific organs or artifacts (and the outer edges of the structure), thereby allowing one organ (or artifact) to be distinguished from another.

[0052] CT images often feature low soft tissue contrast, and the overall appearance of the treated volume and the organs at risk often appear visually similar. The applicant has determined that deep learning techniques can provide useful results to support the automated segmentation of a patient's breast tissue, while non-deep learning techniques can provide useful results to support the automated segmentation of a patient's organs at risk. (That is, the applicant has also determined that utilizing imaging information 103 in conjunction with deep learning to automatically segment at least a portion of a specific patient's cardiac tissue is useful.)

[0053] Deep learning (sometimes also referred to as hierarchical learning, deep neural learning, or deep structural learning) is generally limited to a subset of machine learning in artificial intelligence. Deep learning features networks capable of unsupervised learning from unstructured or unlabeled data. That is, deep learning can also be supervised or semi-supervised if needed. Deep learning frameworks include deep neural networks, deep belief networks, recurrent neural networks, and convolutional neural networks.

[0054] Deep learning uses multiple layers to progressively derive higher-level features from the raw input. In a typical configuration, each layer learns to transform its input data into a more abstract and complex representation. As a simple example, in image recognition applications, the initial raw input might be a matrix of pixels. The first representation layer extracts the pixels and encodes the edges; the second layer assembles and encodes the permutations of those edges; the third layer encodes specific features such as the nose and eyes; and the fourth layer identifies that the image contains a face. Generally, the deep learning process learns on its own which features are best placed in which layers.

[0055] One approach allows deep learning algorithms to be trained in a supervised learning setting using images and real-world contours of hundreds of patients. For example, useful data can be obtained from healthcare providers worldwide. The images in the training set can be selected to represent a practical range of common anatomical artifacts. If needed, real-world contours can be created by human anatomy experts as part of the algorithm development.

[0056] While expert-level performance can be achieved using traditional models and / or atlas-based algorithms for automatic segmentation, this performance is only achievable in a limited number of organs (therefore, further human-based editing is required before it can be clinically acceptable). The inventors have determined that, by comparison, the aforementioned deep learning-based methods can produce significantly better results, especially for structures without well-defined learning settings using images and real contours from hundreds of patients.

[0057] At box 203, the process 200 provides for automatically determining a virtual skin volume corresponding to at least a portion of the breast tissue 104 of a particular patient, also using imaging information 103. This activity may include, in a manner that utilizes at least part of the aforementioned three-dimensional CT imaging information and two-dimensional orthogonal tomographic localization image information, to determine the virtual skin volume. (As used herein, the expression "virtual skin volume" will be understood to refer to the patient volume formed by an external incision around the skin, obtained through calculation (and therefore "virtual").)

[0058] In typical radiotherapy application settings, the patient will lie or sit on one or more patient support devices 110, such as a treatment bed, chair, armrest, or similar. In conjunction with or instead of the foregoing, one or more parts of the patient may be held in a fixed position by one or more patient fixation devices 111. Patient fixation devices are used to hold parts of the patient in a fixed position during radiation administration to keep the treatment volume and / or organs at risk in a fixed position relative to the radiation source 115. Patient support devices and patient fixation devices are well understood in the art and do not require further detailed description herein.

[0059] Considering the foregoing, at optional box 204, the process 200 can provide automatic registration of at least one of the patient support surface 110 and the patient fixation device 111 with at least some of the imaging information in the imaging information 103 to provide corresponding result registration information. Automatic registration of the device to the imaging information 103 can include automatically registering a model of at least one of the patient support surface and the patient fixation device to at least some of the imaging information. "Registration" refers to aligning and spatially corresponding items. In this context, a representative model of the artifact in question is aligned and spatially corresponding to one or more features in the imaging information. The latter can be used as further input when automatically optimizing a radiotherapy plan as described below.

[0060] Alternatively, in conjunction with or instead of the foregoing, in optional box 205, process 200 uses imaging information 103 to automatically determine the shape of at least a portion of the patient's body. This body shape represents the patient's outermost physical edge.

[0061] At box 206, the process 200 provides input 207 to the automatically segmented patient content and virtual skin volume as described above, which is used to automatically optimize the radiotherapy plan for the treatment volume 105 for that particular patient, thereby providing an optimized radiotherapy plan 113 for that patient.

[0062] Various methods for optimizing radiotherapy plans are known in the art. However, it can be noted that, as mentioned above, the aforementioned registration information 208 can also be used for further input when automatically optimizing radiotherapy plans.

[0063] At the end of the optimization process, process 200 produces an optimized radiotherapy plan 113 for a specific patient, as indicated in box 209. (As will be described in more detail below, process 200 may actually produce multiple treatment plans, including, for example, three plans that are different from each other regarding their specific radiation delivery methods and techniques.) If desired, and as shown in optional boxes 210 and 211, process 200 may further include using the aforementioned body shape, virtual skin volume, and registration information as input to automatically calculate the trajectory of the radiotherapy platform, collision detection, and virtual rehearsal information for treatment delivery according to the optimized radiotherapy plan 113.

[0064] It should be understood that these plans are readily available and can be used accordingly to administer radiation to patients via the aforementioned platform. In one method, at least one such outcome plan is used to deliver therapeutic radiation to treat the patient's therapeutic volume.

[0065] Now for reference Figure 3 Specific explanations of these teachings will be presented. For illustrative examples, Figure 3 This process 300 is presented in the context of a breast cancer planning workflow (i.e., developing an implementable beam-based radiotherapy to treat a therapeutic volume 105 comprising a cancerous tumor in the thymus tissue 104 of a specific patient). It will be understood that the specific details of this example are intended for illustrative purposes and are not intended to suggest any particular limitation on these teachings.

[0066] At box 301, the process 300 begins the workflow with CT images. After properly positioning the patient on the chest plate according to well-understood existing technical practices, the activity includes obtaining at least one tomographic localization film and one CT scan. By one method, the former includes obtaining at least two orthogonal tomographic localization films, the at least two tomographic images including the patient's arms and chest plate, the chest plate being as wide as possible in field given the limitations of available equipment and other application settings. By one method, at least one CT scan includes an upward view including a slice above the patient's shoulders and a downward view below the patient's breast tissue 104, as well as a scan including the treatment volume 105. In typical application settings, these images are in a digital format as previously described.

[0067] Overall, the equipment and processes used to acquire these computed tomography and CT scan images are well understood in the prior art. While the acquisition of specific views and subsequent uses described herein need not necessarily conform to existing art practices, for the sake of brevity, further detailed descriptions of the image acquisition itself are unnecessary.

[0068] At box 302, the aforementioned image is imported into an automatic contouring activity. Contouring, of course, includes identifying and / or specifying the shape of individual organs, tissues, or other anatomical structures and patient artifacts, such as, but not limited to, part or all of the patient's breast tissue, the patient's lungs, heart, and / or chest wall. This contouring activity includes, for example, automatically contouring the patient's body shape excluding the sternum and any accompanying lines / conductors that may be present. Alternatively, in place of or in combination with the aforementioned method, this contouring activity includes automatically contouring the sternum and / or the aforementioned lines / conductors that are separate from and distanced from other image content.

[0069] One method may include expanding available CT images outward to include the patient's external body shape. In this configuration, the CT image content can be expanded to include the peripheral edges of the patient's torso, neck, and / or part or all of the patient's arms.

[0070] This automated profiling activity utilizes relevant models through a method. For example, relevant models of the thoracic plate and / or patient support treatment bed can be used to assist in these aspects.

[0071] At box 303, and following the aforementioned contouring activity, process 300 provides an external beam planning process. This external beam planning activity may, if desired, include the use of an automated or user-initiated breast planning creation wizard. (It will be understood that the wizard comprises software that automates a complex task by asking the user a series of easily answered questions, the answers of which then drive the customized execution of the task.)

[0072] Based on the contouring information, the box 303 may include automatically selecting whether the depicted treatment target (or multiple targets) represents the left breast, the right breast, or both breasts. This selection may be based, for example, on structural codes assigned during the contouring process. (If desired, the process 300 will be adapted to provide the technician with the opportunity to override and modify this automatic selection.) Overall, the process also includes automatically defining a specific patient orientation. That is, if desired, the opportunity may be provided for participating technicians to select including lymph nodes.

[0073] The process then provides a template for automatically selecting a specific starting point for breast treatment. This selection can be based on various criteria. One approach includes criteria that incorporate the patient's clinical goals (including, where applicable, any established ordering and / or prioritization of these goals). Another approach includes criteria that include an associated dosage prescription, the identification of a default linear accelerator, and / or any energy specification (where the latter can be automatically defined, user-specified, or overridden on demand). Yet another approach includes criteria that may include pill size (including, for example, the number of any corresponding portions). Yet another approach includes criteria that may include information on the size of the skin flash (e.g., in millimeters or centimeters). And yet another approach includes criteria that may include specifying a particular RapidPlan. TM (RP) model to facilitate the prediction of histogram dose-volume information.

[0074] When considering pills, the process may, if desired, include opening a display window on the user interface to facilitate defining the pill. One method would be suitable for defining only a fraction of the pill rather than the entire pill. When only a fraction of the pill is present, a method could be used to automatically create at least six corresponding plans. ("Fragment" refers only to a portion of the total dose represented, such as a dose administered from a specific angle or field.)

[0075] This activity involves selecting a specific imaging template and automatically creating a field with corresponding settings, which can be used as the initial starting point for treatments using a specific accelerator. Examples include, but are not limited to, cone-beam computed tomography (CBCT), megavolts (MV), and kilovolts (kV).

[0076] In this example, the external beam planning includes three plans: generation, optimization, and calculation (or, when using pellets for a small fraction, six plans). These plans include the TO-VMAT (Volume-Modulated Intensity-Modulated Radiotherapy Trajectory Optimizer) plan, the TO-IMRT (Intensity-Modulated Radiotherapy Trajectory Optimizer) plan, and the iComp irregular surface compensation technique plan. These different plans use different techniques depending on the specific circumstances. These teachings will also apply to other techniques such as, for example, hybrid combinations of VMAT and IMRT, or hybrid combinations of open-field and IMRT, if desired.

[0077] The geometry definition of the VMAT arc and IMRT field uses a trajectory optimizer (TO) that supports dynamic or static collimator angle determination, coplanar and non-coplanar fields, energy selection, and isobath placement. (Those skilled in the art will recognize that, as understood in the art, trajectory optimization is an extension of the segmentation method used in radiotherapy.) In particular, the field sequence is adjusted for planning quality, conflict avoidance, and efficient delivery. Additional modifications include rack slowdowns for radians, restricted leaf sequence, and modified control point weights; these teachings will be applied to help improve planning quality.

[0078] If desired, optimizing the leaf sequence and flux can take into account the skin flash margin to handle possible target changes and movements.

[0079] The geometry of an irregular surface compensator program can be defined based on a target volume that has been contoured, including iso-depth point placement. Flux optimization can be achieved by minimizing the dose to defined at-risk organs while maintaining the desired coverage of the target volume, taking into account physician-prioritized clinical objectives. A method allows for the automatic addition of a predetermined skin flash margin to the flux to accommodate potential target changes or shifts in the program. Once the flux is optimized, blade movement and dose calculations can be performed automatically.

[0080] If desired, automated interactive optimization of volumetric arc-shaped intensity-modulated radiotherapy (IMRT) plans (as understood in the art) can be utilized during the optimization process for some or all of the plans to achieve Pareto optimal dose.

[0081] TO-VMAT program creation

[0082] This process can automatically optimize the radiation source trajectory by defining one or more arc and collimator positions as a function of sorting considerations for one or more organs of risk, including the breast template and the patient's treatment volume (the latter considering the presence or absence of nodes as needed and / or specified). A method allows conflict detection to run in the background to check that the trajectory is conflict-free. A method transforms available clinical objectives into optimization goals, which are then used to guide the iterative optimization process. A method limits the accelerator photon energy.

[0083] After optimization, the process can calculate the dosage to be implemented.

[0084] TO-IMRT program creation

[0085] This process can automatically optimize the radiation source trajectory, at least in part, by defining the static field as a function of the ordering considerations of one or more organs of risk, in the breast template and the patient's treatment volume (the latter considering the presence or absence of nodes as needed and / or specified). One method allows collision detection to run in the background to check that the trajectory is collision-free. Another method utilizes this objective to guide the iterative optimization process. A third method limits the energy.

[0086] After optimization, the process can calculate the blade movement and dosage to be applied when using a multi-blade collimator.

[0087] iComp plans to create

[0088] This process can automatically identify tangent fields based on the patient's treatment volume. These fields are parallel and relative. The number of fields typically depends on the size of the target. While two fields are usually sufficient, additional fields can be added. If lymph nodes are included, additional adjacency fields are sometimes added to facilitate node treatment. An additional conflict detection can be run in the background to check that the path is conflict-free. A method is used to translate the available clinical objective into an optimization objective, which is then used to guide the iterative optimization process. A method is used to limit the energy.

[0089] Conflict-free geometry can be calculated based on the patient geometry of the treatment bed and (multiple) fixation devices in the arcuate plane relative to the relevant gantry. Conflict regions are calculated using a method and an image-based model. This method allows for the automated search and generation of more complex tracks, which can, in turn, improve program quality and efficacy.

[0090] After optimization, the process can calculate the blade movement and dosage used when using a multi-blade collimator.

[0091] At box 304, process 300 provides an evaluation of one or more of the aforementioned plans. Similarly, in a typical application setting, the aforementioned process provides three separate plans (or six, when selecting pills for a small subset). By a method, no results are presented to the technician for consideration until all plans are ready.

[0092] The assessment includes presenting the calculated results of all three breast plans when calculating results corresponding to physician intent and other indicators of interest ranked by priority, such as heterogeneity index (HI), gradient index (GI), monitoring units, intensity-modulated complexity, and treatment time. When presenting dual plans that take into account the presence of pills, plans can be created for both plans with and without pills. In any case, all three comprehensive breast plans can be presented with scores based on corresponding clinical objectives and other indicators of interest. For example, the scores may include pass / fail indicators for each clinical objective.

[0093] With this configuration, technicians can review proposed plans and identify the optimal plan for a patient. This plan, in turn, can lead to the approval of a specific plan for that particular patient.

[0094] If desired, these teachings will be applicable to virtual rehearsals of treatment delivery in a defined treatment field via 3D animation, allowing technicians to perform such rehearsals. These virtual rehearsals may include representations of treatment delivery in the dimensions of actual treatment machines, treatment beds, fixation devices, and patients. If desired, the hill-shaped view provides indication of the clearances between the patient and the application-set machines and / or between each machine during treatment delivery.

[0095] Those skilled in the art will recognize that various modifications, alterations, and combinations can be made with respect to the embodiments described above without departing from the scope of the invention, and that such modifications, alterations, and combinations are considered to be within the scope of the concept of the invention.

Claims

1. A method for automatically planning a treatment volume within breast tissue of a specific patient using radiation-based therapy, the method comprising: Control circuit: Access imaging information for a treatment area, which includes the treatment volume for the specific patient; Using the imaging information and using deep learning to automatically segment at least some breast tissue of the specific patient and using non-deep learning to automatically segment at least some parts of some organs at risk, to provide automatically segmented patient content; The imaging information is used to automatically determine the virtual skin volume that corresponds at least a portion of the breast tissue of the specific patient; Using the automatically segmented patient content and the virtual skin volume as input, the radiotherapy plan for the treatment volume of the specific patient is automatically optimized, thereby providing an optimized radiotherapy plan for the specific patient.

2. The method according to claim 1, wherein the imaging information includes three-dimensional computed tomography imaging information.

3. The method of claim 2, wherein the imaging information further comprises two-dimensional localization slice imaging information, and wherein using the imaging information to automatically determine at least a portion of a virtual skin volume corresponding to at least a portion of the breast tissue of the particular patient comprises: The virtual skin volume is determined using both the three-dimensional computed tomography imaging information and the two-dimensional localization film imaging information.

4. The method according to claim 1, 2, or 3, wherein using the imaging information and deep learning to automatically segment at least some breast tissue from the specific patient further comprises: The imaging information and deep learning are used to automatically segment at least some of the heart tissue of the specific patient.

5. The method according to any one of claims 1 to 3, wherein the dangerous organ comprises at least one of the following: the lung, a portion of the spine, and a portion of the chest wall.

6. The method according to any one of claims 1 to 3, further comprising: At least one of the patient support surface and the patient fixation device is automatically registered with at least some of the imaging information to provide registration information, and the registration information is used as further input to automatically optimize the radiotherapy plan.

7. The method of claim 6, wherein automatically registering at least one of the patient support surface and the patient fixation device with at least some of the imaging information comprises: The model of at least one of the patient support surface and the patient fixation device is automatically registered with at least some of the imaging information in the imaging information.

8. The method according to claim 6, further comprising: The imaging information is used to automatically determine at least a portion of the patient's body shape.

9. The method according to claim 8, further comprising: Using at least the body shape, the virtual skin volume, and the registration information as input, virtual rehearsal information for radiotherapy platform trajectory, conflict detection, and treatment delivery is automatically calculated based on the optimized radiotherapy plan.

10. A device for automatically planning the intramural treatment volume of breast tissue in a specific patient, the device comprising: A memory containing imaging information for a treatment area, the treatment area including the treatment volume of the particular patient; Control circuitry, operably coupled to the memory and configured to: Access imaging information for a treatment area, which includes the treatment volume for the specific patient; Using the imaging information and using deep learning to automatically segment at least some breast tissue of the specific patient and using non-deep learning to automatically segment at least some parts of some organs at risk, to provide automatically segmented patient content; The imaging information is used to automatically determine the virtual skin volume that corresponds at least a portion of the breast tissue of the specific patient; Using the automatically segmented patient content and the virtual skin volume as input, the radiotherapy plan for the treatment volume of the specific patient is automatically optimized, thereby providing an optimized radiotherapy plan for the specific patient.

11. The apparatus of claim 10, wherein the imaging information includes three-dimensional computed tomography imaging information.

12. The apparatus of claim 11, wherein the imaging information further includes two-dimensional localization film imaging information, and wherein the control circuitry is configured to: automatically determine, at least partially, a virtual skin volume corresponding to at least a portion of the breast tissue of the particular patient using the imaging information to determine the virtual skin volume by using both the three-dimensional computed tomography imaging information and the two-dimensional localization film imaging information.

13. The apparatus of claim 10, 11 or 12, wherein the control circuitry is further configured to use deep learning to automatically segment at least some cardiac tissue of the particular patient.

14. The device according to any one of claims 10 to 12, wherein the dangerous organ comprises at least one of the following: a lung, a portion of the spine, and a portion of the chest wall.

15. The apparatus according to any one of claims 10 to 12, wherein the control circuit is further configured to: At least one of the patient support surface and the patient fixation device is automatically registered with at least some of the imaging information to provide registration information, and the registration information is used as further input to automatically optimize the radiotherapy plan.

16. The apparatus of claim 15, wherein the control circuit is configured to automatically register at least one of the patient support surface and the patient fixation device with at least some of the imaging information by automatically registering a model of at least one of the patient support surface and the patient fixation device with at least some of the imaging information.

17. The apparatus of claim 15, wherein the control circuit is further configured to: The imaging information is used to automatically determine at least a portion of the patient's body shape.

18. The apparatus of claim 17, wherein the control circuit is further configured to: Using at least the body shape, the virtual skin volume, and the registration information as input, virtual rehearsal information for radiotherapy platform trajectory, conflict detection, and treatment delivery is automatically calculated based on the optimized radiotherapy plan.

Images