System, program and method for providing cranial correction target data

Abstract

To provide a system, a program and a method for providing cranial correction target data that properly obtain cranial correction target data suitable for a case.SOLUTION: A system 10 for providing cranial correction target data includes: a storage section 70 for storing cranial shape data D16 based on a cranial shape of a subject 100; a pre-treatment section 21 for subjecting the cranial shape data D16 to pre-treatment for applying the data to a generation model D50 that has mechanically learned the cranial shape data D16, in order to generate pre-treated data; a generation model application section 22 for applying the pre-treated data to the generation model D50 in order to produce generation data; and a post-treatment section for generating cranial correction target data D20 obtained by subjecting the generation data to post-treatment. The generation model D50 is obtained by learning teaching data corresponding to a plurality of past cases. The teaching data includes cranial shape data D16 of an arbitrary subject and cranial correction target data D20 of the arbitrary subject, as one set.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a cranial correction target data providing system, a cranial correction target data providing program, and a cranial correction target data providing method. [Background technology]

[0002] Conventionally, a method is known for generating cranial shape correction helmet manufacturing data used to manufacture a cranial shape correction helmet, as described in Patent Document 1, based on cranial correction target data provided by a cranial correction target data providing system.

[0003] The method for generating cranial shape correction helmet production data described in Patent Document 1 generates cranial shape correction helmet production data for producing a cranial shape correction helmet for correcting a deformed skull using a 3D printer. This method includes an input step of importing corrective target cranial shape data indicating the corrective target cranial shape into 3D CAD software and generating mesh data indicating the corrective target cranial surface shape, an offset step of offsetting the mesh data by a predetermined amount in the expansion direction, a deletion step of deleting unnecessary portions from the mesh data, a thickness imparting step of imparting a predetermined amount of thickness to the mesh data, and an output step of outputting the mesh data that has been offset in the expansion direction, with the unnecessary portions deleted and the thickness imparted. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Patent Publication No. 2021-74897 Summary of the Invention [Problem to be solved by the invention]

[0005] The cranial shape correction helmet production data is created based on cranial correction target data corresponding to the cranial shape target, which is displayed as a correction result obtained by virtually removing distortions from the cranial shape of the subject to be corrected. In other words, the cranial correction target data is required in advance as a target for correcting the subject's cranial shape. Because this cranial correction target data is the result of correcting the subject's original head shape, it is created as a target that differs for each subject depending on the subject's original head shape. Therefore, to create the cranial correction target data for a subject, even if a template is appropriately selected from multiple cranial correction target data prepared as templates, it is modified to some extent to suit the subject's cranial shape. Even if such a template is modified to suit the subject's cranial shape, it is not easy to appropriately modify it to suit various cases of the subject, and the influence of the correction technician's number of cases experienced on the correction accuracy cannot be ignored.

[0006] The present invention has been made in consideration of the above-mentioned circumstances, and its purpose is to provide a cranial correction target data providing system, a cranial correction target data providing program, and a cranial correction target data providing method that can more appropriately obtain cranial correction target data that is appropriate for each case. [Means for solving the problem]

[0007] A cranial correction target data providing system that solves the above-mentioned problems comprises: a memory unit that stores cranial shape data, which is data indicating the cranial shape based on the results of measuring the cranial shape of a subject; a pre-processing unit that applies pre-processing to the cranial shape data to generate pre-processed data in order to apply the cranial shape data to a generative model generated by machine learning; a generative model application unit that applies the pre-processed data to the generative model to generate generated data; and a post-processing unit that post-processes the generated data to generate cranial correction target data adjusted to a size corresponding to the cranial shape data. The generative model generated by machine learning is a model obtained by learning training data corresponding to a plurality of past cases, and the training data is data that includes the cranial shape data of an arbitrary subject and the cranial correction target data of the arbitrary subject as a set. The generative model is a model generated by the machine learning in which the cranial shape data of the arbitrary subject is used as input data and the cranial correction target data of the arbitrary subject is used as output data.

[0008] A cranial correction target data provision program that solves the above-mentioned problem includes a step of storing cranial shape data, which is data indicating the cranial shape based on the results of measuring the cranial shape of a subject, in a memory unit; a step of applying preprocessing to the cranial shape data in a preprocessing unit to generate preprocessed data in order to apply the cranial shape data to a generative model generated by machine learning; a step of applying the preprocessed data to the generative model in a generative model application unit to generate generated data; and a step of applying post-processing to the generated data to generate cranial correction target data adjusted to a size corresponding to the cranial shape data in a post-processing unit.The program makes it possible to apply a generative model obtained by learning training data corresponding to a plurality of past cases as the generative model generated by machine learning, and the training data is data that includes the cranial shape data of an arbitrary subject and the cranial correction target data of the arbitrary subject as a set, and the generative model is a model generated by the machine learning in which the cranial shape data of the arbitrary subject is used as input data and the cranial correction target data of the arbitrary subject is used as output data.

[0009] A method for providing cranial correction target data that solves the above-mentioned problem includes the steps of: storing cranial shape data, which is data indicating the cranial shape based on the results of measuring the cranial shape of a subject, in a memory unit; applying preprocessing to the cranial shape data in a preprocessing unit to generate preprocessed data in order to apply the cranial shape data to a generative model generated by machine learning; applying the preprocessed data in a generative model application unit to the generative model to generate generated data; and applying postprocessing to the generated data in a postprocessing unit to generate cranial correction target data adjusted to a size corresponding to the cranial shape data. The method makes it possible to apply a generative model obtained by learning training data corresponding to a plurality of past cases as the generative model generated by machine learning, and the training data is data that includes the cranial shape data of an arbitrary subject and the cranial correction target data of the arbitrary subject as a set, and the generative model is a model generated by the machine learning in which the cranial shape data of the arbitrary subject is used as input data and the cranial correction target data of the arbitrary subject is used as output data.

[0010] With this configuration, cranial correction target data is generated using a generative model based on machine learning. In other words, by using the generative model, cranial correction target data appropriate for the cranial shape of the subject is generated from the cranial shape data of the subject. Furthermore, because the generative model is generated based on a large amount of training data, it reflects a variety of cases, making it possible to obtain cranial correction target data that appropriately corresponds to the case with high accuracy. This makes it possible to more appropriately obtain cranial correction target data that is appropriate for the case.

[0011] In a preferred configuration, the post-processing unit is configured to perform at least one of the following processes: reconstructing the generated data into a three-dimensional mesh, adjusting the size, and filling holes in the three-dimensional mesh; the machine learning is deep learning; and the generative model is a model trained by collectively learning the weights of all layers from the time the input data is input to the time the output data is output.

[0012] With this configuration, even if the data consists of a non-3D mesh generated by a generative model, it can be reconstructed into a 3D mesh, its size adjusted, and holes in the 3D mesh filled, making it possible to make the cranial correction target data to which the generative model has been applied appropriate for the subject.

[0013] In a preferred configuration, the system further includes a learning unit that performs the machine learning, wherein the memory unit stores the training data, and the learning unit acquires the training data from the memory unit, uses the input data as cranial shape data of the arbitrary subject, and performs the machine learning using the output data as cranial correction target data of the arbitrary subject, thereby generating the generative model, and the generative model can be set in the generative model application unit so as to be applicable to generating the generative data.

[0014] With this configuration, the generative model is generated by machine learning using the cranial shape data of an arbitrary subject as input data and the cranial correction target data of an arbitrary subject as output data. In other words, the generative model is generated as an appropriate one based on the learning of many cases (examples).

[0015] In a preferred configuration, the preprocessing unit further has a configuration capable of preprocessing the input data and the output data, and the learning unit has a configuration to perform machine learning based on the preprocessed input data and the output data to generate the generative model.

[0016] With this configuration, the generation of a generative model through machine learning is performed based on input data and output data that have been preprocessed to be suitable for learning, thereby increasing the versatility of the generative model obtained through machine learning.

[0017] In a preferred configuration, the cranial shape data is three-dimensional data, the cranial correction target data is three-dimensional data, and the deep learning has a configuration in which an encoder-decoder structure type network is applied.

[0018] With this configuration, in deep learning performed by a network of the "encoder"-"decoder" structure type, the input data can be three-dimensional cranial shape data, and the output data can be three-dimensional cranial correction target data. [Effects of the Invention]

[0019] According to the present invention, it becomes possible to more suitably obtain cranial correction target data suited to a particular case. [Brief explanation of the drawings]

[0020] [Figure 1] 1 is a block diagram showing the functional configuration of a cranial correction target data providing system, a cranial correction target data providing program, and a cranial correction target data providing method according to an embodiment of the present invention. FIG. [Figure 2] FIG. 2 is a schematic diagram showing an example of the skull shape of a subject according to the embodiment. [Figure 3] FIG. 2 is a schematic diagram showing an example of a cranial shape correcting helmet in the same embodiment. [Figure 4] 4 is a flowchart showing the flow of the cranial correction target data providing system according to the embodiment. [Figure 5] 10 is a flowchart showing the flow of a cranial correction target data generation process in the embodiment. [Figure 6] 10 is a flowchart showing the flow of a post-processing process in the embodiment. [Figure 7] 10 is a flowchart showing the flow of a learning process in the embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0021] An embodiment of a cranial correction target data providing system, a cranial correction target data providing program, and a cranial correction target data providing method will be described with reference to FIGS.

[0022] As shown in Figure 1, cranial correction target data providing system 10 is a system that provides cranial correction target data D20 corresponding to corrected shape 121 (see Figure 2), which is the cranial shape after correction of cranial shape 101 (see Figure 2) of subject 100. In other words, cranial correction target data D20 provided by cranial correction target data providing system 10 is data used to manufacture cranial shape corrective helmet 61 that is worn on the head of subject 100 to correct cranial shape 101 (see Figure 2) to corrected shape 121 (see Figure 2) indicated in cranial correction target data D20.

[0023] First, the subject 100 is preferably an infant or the like having a correctable, flexible skull, more specifically, an infant between the ages of newborn and about six months old. Since such infants have soft skulls at birth, their heads are often flat or curved. Note that the subject 100 may be referred to as a patient when undergoing medical treatment.

[0024] Referring to Figure 3, cranial shape corrective helmet 61 includes an outer shell 62 made of a hard resin that is difficult to deform, and an inner lining 63 having flexibility such as elasticity and disposed between outer shell 62 and the skull of subject 100. Cranial shape corrective helmet 61 is provided with outer shell 62 and inner lining 63 so that an appropriate force is transmitted from cranial shape corrective helmet 61 to the skull of subject 100 (see Figure 1). In addition, in order to make cranial shape corrective helmet 61 lightweight and less burdensome to subject 100 (see Figure 1), it is preferable that outer shell 62 is made of a light material such as polystyrene foam, and inner lining 63 is made of a light material such as sponge or urethane.

[0025] The outer shell 62 is preferably a roughly shell-shaped structure that covers the skull and has a structure that reduces stuffiness inside, such as a structure with many ventilation holes. The inner lining 63 is preferably a cushion made of an elastically deformable material such as sponge or urethane to reduce the feeling of pressure on the head of the subject 100. The inner lining 63 is also preferably a structure that absorbs sweat and reduces dermatitis, and is preferably adjustable in thickness and elasticity to adjust the corrective force, which is the pressure applied to the skull, and is preferably replaceable to maintain cleanliness.

[0026] The cranial correction target data providing system 10 and devices connected thereto will be described in detail with reference to FIG.

[0027] The cranial correction target data providing system 10 has a main processing server 11. In addition, in the cranial correction target data providing system 10, the main processing server 11 may be connected to a network NW, and a subject-side terminal 31, a diagnosis-side terminal 41, and a manufacturing device 51 may be connected via this network NW so that information can be transmitted to and from each other. Furthermore, the main processing server 11, the subject-side terminal 31, the diagnosis-side terminal 41, and the manufacturing device 51, which are connected via the network NW, may be so close that at least two of them are bus-connected, or may be connected to each other via short-distance communication. Such two or more devices may have one of them connected to the network NW.

[0028] Subject-side terminal 31 outputs cranial shape measurement information, which is information required to calculate cranial shape 101 (see FIG. 2) of subject 100. Subject-side terminal 31 is an information processing device including a camera or the like that measures subject 100, such as a digital camera, a mobile phone, a smartphone, a tablet terminal, or a small computer. Subject-side terminal 31 can acquire information about cranial shape 101 (see FIG. 2) of subject 100 by measuring subject 100, and can also transmit information via network NW.

[0029] The subject-side terminal 31 can acquire, as the cranial shape measurement information, images or videos capturing the overall external shape of the skull of the subject 100. For example, the subject-side terminal 31 can capture images or videos of the entire circumference and top of the skull with the skin of the subject 100 visible. Note that the subject-side terminal 31 may also be a device that can measure the skull shape of the subject 100 more accurately than images or videos, such as a three-dimensional scanner using a laser.

[0030] Subject-side terminal 31 includes input unit 32 and output unit 33. Input unit 32 is a unit for acquiring information from outside subject-side terminal 31, and is a unit for inputting information from a touch panel, camera, microphone, etc., and network NW. Output unit 33 is a unit for outputting information from inside subject-side terminal 31, and is a unit for outputting information to an image display device, audio device, etc., and network NW.

[0031] The diagnosis-side terminal 41 can refer to the subject information, the cranial shape data D16, and the cranial correction target data D20, and may output adjustment parameters indicating the amount of adjustment for the cranial shape data D16. The diagnosis-side terminal 41 is an information processing device including an image display device and an instruction input device, such as a smartphone, a tablet terminal, or a small computer. The diagnosis-side terminal 41 can display information about the subject 100 obtained from the main processing server 11 connected via the network NW on the image display device. The diagnosis-side terminal 41 can also set adjustment parameters including the amount of adjustment corresponding to the information about the subject 100 whose image is displayed, and may also be able to transmit the set adjustment parameters to the main processing server 11.

[0032] The diagnosis side terminal 41 includes an input unit 42 and an output unit 43. The input unit 42 is a unit for acquiring information from outside the diagnosis side terminal 41, and includes a unit for inputting information from a keyboard, mouse, camera, microphone, etc., and a network NW. The output unit 43 is a unit for outputting information from inside the diagnosis side terminal 41, and includes a unit for outputting information to an image display device, audio output device, etc., and the network NW. The diagnosis side terminal 41 can also register, for example, shape changes to the outer shell 62 and inner lining 63 of the cranial shape corrective helmet 61.

[0033] Based on the information obtained via the diagnosis terminal 41, the diagnostician 110 understands the cranial shape of the subject 100, determines whether correction is necessary, determines whether the shape of the cranial shape correcting helmet 61 when correcting, the interior 63 that provides the corrective force, etc. are appropriate, and gives new instructions. For example, if cranial shape correction is provided as a medical treatment, the diagnostician 110 should be a medical professional such as a doctor.

[0034] Furthermore, the diagnosis-side terminal 41 may be able to adjust or correct cranial correction target data D20, which is a target for correcting the distorted head of the subject 100, for example, a target with a more symmetrical head shape. In other words, the diagnosis-side terminal 41 may be able to set or correct the amount of correction for the cranial shape data D16 acquired from the main processing server 11. In this embodiment, the cranial correction target data D20 can be obtained with high accuracy, so that fewer corrections or modifications are required for the cranial correction target data D20, thereby enabling the time required for the correction or modification work to be shortened.

[0035] The main processing server 11 is an information processing device such as a computer or a server, and includes an input unit 12, an output unit 13, an information processing unit 14, and a storage unit .

[0036] The input unit 12 is an input interface into which necessary information is input, such as input from a keyboard, mouse, or touch panel, input from a camera or microphone, information input from other devices or servers via the network NW, and information input from an external storage device.

[0037] The output unit 13 is an output interface that outputs necessary information, such as to an image display device, a character display device, or an audio device, to other devices or servers via the network NW, or to an external storage device.

[0038] The information processing unit 14 is an information processing unit that is made up of a computer device that processes information in the main processing server 11. For example, the information processing unit 14 has a central processing unit, a volatile memory, a non-volatile memory, and an input / output interface. The input / output interface is capable of communicating information with the input unit 12, the output unit 13, the storage unit 70, etc.

[0039] The storage unit 70 is a part that can store various information such as subject information, and exchanges information to be stored, information to be read, instructions to delete information, etc. with the information processing unit 14. The storage unit 70 is a part that is made up of internal storage, external storage, or a combination thereof, and is made up of, for example, one or more of a hard disk, SSD, USB memory, etc. For example, the storage unit 70 may include a cloud system that exchanges information by information communication via the network NW.

[0040] For example, the storage unit 70 stores cranial shape data D16 and cranial correction target data D20. In addition to the cranial shape data D16 and cranial correction target data D20 of the subject 100, the storage unit 70 also stores the cranial shape data D16 and cranial correction target data D20 of a plurality of different subjects 100. The storage unit 70 also stores a plurality of training data. The training data is a set of the cranial shape data D16 and cranial correction target data D20 of an arbitrary subject 100, and the plurality of training data is a collection of training data of different arbitrary subjects 100. The number of data required as training data is preferably 1000 sets or more, more preferably 2000 sets or more, and even more preferably 3000 sets or more.

[0041] The main processing server 11 includes a subject information management unit 15, a shape data calculation unit 16, and a cranial correction target data generation unit 20, which perform their functions through program processing corresponding to the functions in the information processing unit 14. The main processing server 11 may also include a helmet shape data calculation unit 25, a helmet production data calculation unit 26, a cushion fitting information unit 27, a shipping instruction unit 28, and a learning unit 50, which perform their functions through program processing corresponding to the functions.

[0042] The subject information management unit 15 is a unit that manages information about the subject 100, including subject information that is information necessary for cranial shape correction of the subject 100, such as information that is recorded in a medical chart. The subject information management unit 15 also manages cranial shape measurement information for the cranial shape corresponding to the subject 100, cranial shape data D16, cranial correction target data D20, adjustment parameters, production data, various parameters, and the like. Note that these various data are stored in the storage unit 70, for example.

[0043] The shape data calculation unit 16 calculates skull shape data D16, which is data indicating the skull shape 101, from the skull shape measurement information. The shape data calculation unit 16 recognizes the skull shape 101 (see FIG. 2) based on information related to the skull shape 101 of the subject 100 input from the subject-side terminal 31, and generates the skull shape data D16 corresponding to the skull shape 101 (see FIG. 2) of the subject 100, for example, as three-dimensional data. For example, the skull of the subject 100 may be the part above the neck of the subject 100, including the so-called head. The information related to the skull shape 101 of the subject 100 input from the subject-side terminal 31 may be information obtained by a scanner that can accurately scan the skull shape 101, or may be a video or multiple images including the entire circumference and top of the subject 100 with the skin of the subject 100 visible.

[0044] The scan data included in the cranial shape measurement information is, for example, data acquired from a 3D (three-dimensional) scanner, and is scan data of the head of the subject 100 that has more or less distorted shape. The shape data calculation unit 16 may correct the scan data to make it have a more balanced cranial shape, and use the corrected data as cranial shape data D16.

[0045] The shape data calculation unit 16 may output the skull shape data D16 of the subject 100 to the diagnosis-side terminal 41. In other words, the skull shape data D16 may be made available for reference on the diagnosis-side terminal 41.

[0046] Furthermore, the shape data calculation unit 16 may use adjustment parameters set by the diagnostician 110 on the diagnostic terminal 41 to readjust the skull shape data D16.

[0047] For example, in this embodiment, the data set used for the skull shape data D16 is a three-dimensional data format that stores edge and direction coordinates and can be saved as an object file ".obj." This data set is three-dimensional data that represents the skull shape of the subject 100, and each data set may include measured data (scan data) and edited data (edited data). In this embodiment, for example, well-known image processing techniques can be used to generate the skull shape data D16 in the ".obj" file format as three-dimensional data from images or videos obtained as skull shape measurement information.

[0048] The cranial correction target data generation unit 20 generates cranial correction target data D20 corresponding to the cranial shape data D16 by applying a trained generative model D50 generated by machine learning to the cranial shape data D16.

[0049] Referring to FIG. 2, cranial correction target data generating unit 20 generates cranial correction target data D20 indicating a corrected shape 121 of subject 100 as a cranial correction target from cranial shape data D16 indicating cranial shape 101 of subject 100, for example.

[0050] For example, the cranial shape data D16 is stored as data for a specified reference point C0. The cranial shape data D16 includes a center line L0 extending in the anterior-posterior direction through the center of the skull, and two symmetric inclined lines L1 and L2 inclined at a predetermined angle (e.g., 30 degrees) relative to the center line L0. The cranial correction target data generating unit 20 may then obtain post-correction cranial correction target data D20 corresponding to the cranial shape 101 at the intersection of the inclined lines L1 and L2.

[0051] Referring to FIG. 1, the cranial correction target data generating unit 20 includes a pre-processing unit 21, a generative model applying unit 22, and a post-processing unit .

[0052] The preprocessing unit 21 applies preprocessing to the cranial shape data D16 to generate preprocessed data in order to apply the cranial shape data D16 to the generative model D50. The cranial shape data D16 is input to the preprocessing unit 21 as data in the ".obj" file format. The preprocessing unit 21 or the like may also perform a basic analysis of the cranial shape data D16 to calculate the volume ratio, brachycephalic ratio, CA (Cranial Asymmetry), CVAI (Cranial Vault Asymmetry Index), and the like.

[0053] The preprocessing unit 21 converts the skull shape data D16 to generate preprocessed data. That is, the preprocessing unit 21 generates the preprocessed data by converting the skull shape data D16, which is mesh data in the ".obj" file format, into voxel data. For example, a predetermined library that can convert mesh data into voxel data at a custom resolution can be used.

[0054] Here, voxel data is image data with a three-dimensional extent, also known as a three-dimensional image, and is equivalent to multiple layers of two-dimensional image data. Voxel data is also used in CT scans, MRIs, seismic tomography, etc. The smallest component of voxel data is a voxel, and the value of the element is the voxel value.

[0055] The generative model application unit 22 applies the preprocessed data to the generative model D50 to generate generative data. In other words, the generative model application unit 22 uses preprocessed data based on the cranial shape data D16 of the subject 100 as input data for the generative model D50, and obtains generative data as output data of the learning result for the input data. The generative data is data that becomes cranial correction target data D20 after being post-processed.

[0056] In this embodiment, the input data of the generative model application unit 22 is voxel data, and the output data is also voxel data.

[0057] The generative model application unit 22 has an "encoder"-"decoder" type network, and by associating this network with the generative model D50, the generative model D50 is applied to the input preprocessed data and the generated data is output.

[0058] More specifically, the generative model application unit 22 can obtain a feature representation of the condensed 3D data using an "encoder." The generative model application unit 22 also processes the feature representation of the 3D data obtained from the "encoder" using a "decoder" to generate and output 3D data corresponding to the feature representation.

[0059] The post-processing unit 23 performs post-processing on the generated data to obtain cranial correction target data D20 adjusted to a size corresponding to the cranial shape data D16.

[0060] More specifically, the post-processing unit 23 converts (reconstructs) the generated data from voxel data into mesh data (3D mesh). Furthermore, the post-processing unit 23 performs at least one of the following processes on the generated data converted into mesh data: changing the size, filling holes in the mesh data, and moving the coordinates of the mesh data. The post-processing unit 23 then outputs the mesh data obtained based on the processing performed on the generated data as, for example, a file in ".obj" format.

[0061] Furthermore, the volume ratio, brachycephalic ratio, CA, CVAI, etc. may be calculated by performing a basic analysis on the ".obj" format file output by the post-processing unit 23 or the like.

[0062] The helmet shape data calculation unit 25 calculates the helmet shape of the cranial shape corrective helmet 61 based on the subject information and the cranial correction target data D20. The helmet shape data calculation unit 25 may reflect adjustment parameters set by the diagnostician 110 in the calculation of the helmet shape data. The helmet shape may include data on the outer shell 62 and data on the interior 63.

[0063] The helmet production data calculation unit 26 calculates production data used to manufacture the helmet outer shell 62 based on the helmet shape data. This production data is data necessary for the manufacturing device 51, which is a 3D printer, to manufacture the outer shell 62 of the cranial shape corrective helmet 61. The production data is output from the output unit 13 to the manufacturing device 51 that manufactures the cranial shape corrective helmet 61.

[0064] Note that well-known techniques (for example, the technique disclosed in Japanese Patent Application Laid-Open No. 2021-74897) can be used to calculate the manufacturing data. The calculated helmet shape data and manufacturing data may be managed as information about the cranial shape corrective helmet 61 for the subject 100, and as information about the subject 100.

[0065] Furthermore, the cushion fitting information unit 27 calculates information regarding the manufacture and selection of the interior lining 63 that is provided so that the appropriate force is transmitted from the cranial shape correcting helmet 61 to the skull of the subject 100. By incorporating the interior lining 63 into the outer shell 62, the skull correction in the correction using the cranial shape correcting helmet 61 is performed by the corrective force applied via the flexible interior lining 63, thereby reducing the burden on the skull.

[0066] The shipping instruction unit 28 outputs instructions to the helmet manufacturer, cushion manufacturer, etc. to ship the cranial shape corrective helmet 61 so that the outer shell 62 and the interior lining 63 are a set for the subject 100. At this time, if the outer shell 62 and the interior lining 63 can be recognized as a set, the outer shell 62 and the interior lining 63 may be shipped together or separately.

[0067] The learning unit 50 generates a generative model D50 through machine learning. The learning unit 50 sets the generated generative model D50 in the generative model application unit 22, and enables the generative model application unit 22 to apply the generative model D50 to generating generative data. In other words, the generative model application unit 22 generates generative data based on the applicability of the generative model D50.

[0068] In other words, the generative model D50 is a model obtained based on machine learning of training data corresponding to multiple past cases.

[0069] The training data is data including, as a set, the cranial shape data D16 of any subject 100 and the cranial correction target data D20 applied to that subject 100. The training data may include the cranial shape data D16 and the cranial correction target data D20 stored in the storage unit 70 as a set for each subject 100.

[0070] The teacher data (data set) is divided into two groups: a training set and a validation set, and the ratio of the divided numbers can be set to 9:1, for example.

[0071] The generative model D50 is a model generated based on machine learning in which the input data is the cranial shape data D16 of an arbitrary subject 100 and the output data is the cranial correction target data D20 of the arbitrary subject 100.

[0072] The learning unit 50 performs machine learning using deep learning. The learning unit 50 uses machine learning to collectively learn the weights of all layers from when input data is received until when output data is output, and generates the generative model D50.

[0073] For example, the machine learning in the learning unit 50 uses an "encoder"-"decoder" type deep learning network. In deep learning, the "encoder" obtains a feature representation of condensed three-dimensional data, and the "decoder" generates and outputs three-dimensional data corresponding to the obtained feature representation, learning a model that can output the data.

[0074] The manufacturing device 51 is a device that manufactures the outer shell 62 of the cranial shape correction helmet 61 based on production data. The manufacturing device 51 is, for example, a 3D printer that can manufacture products made of foamed resin such as polystyrene foam using a resin material. The manufacturing device 51 may also be capable of selecting a material that can manufacture a cranial shape correction helmet 61 that is lightweight and has some elasticity. The manufacturing device 51 may also manufacture the cranial shape correction helmet 61 by resin molding or cutting it out from a raw material.

[0075] The manufacturing apparatus 51 includes an input unit 52 and an output unit 53. The input unit 52 is a unit for acquiring information from outside the manufacturing apparatus 51, and includes a keyboard, mouse, touch panel, microphone, and the like, and a unit for inputting information from the network NW. The output unit 53 is a unit for outputting information from inside the manufacturing apparatus 51, and includes an image display device, audio output device, and the like, and outputs information to the network NW.

[0076] The flow of processing in the cranial correction target data providing system 10 will be described with reference to FIGS.

[0077] First, referring to Fig. 4, cranial correction target data provision system 10 has a cranial shape data calculation process (step S100 in Fig. 4) and a cranial correction target data generation process (step S200 in Fig. 4). In addition, cranial correction target data provision system 10 has a helmet shape data calculation process (step S300 in Fig. 4), a helmet production data calculation process (step S310 in Fig. 4), a helmet production process (step S320 in Fig. 4), a cushion fitting process (step S330 in Fig. 4), and a shipping process (step S340 in Fig. 4).

[0078] The skull shape data calculation process (step S100 in FIG. 4) is a process executed by, for example, the shape data calculation unit 16 of the main processing server 11, and is a process for calculating skull shape data D16, which is three-dimensional data. In the skull shape data calculation process (step S100 in FIG. 4), skull shape data D16 corresponding to the skull shape 101 recognized based on information about the skull shape 101 of the subject 100 is generated as three-dimensional data.

[0079] The cranial correction target data generation process (step S200 in FIG. 4) generates cranial correction target data D20 corresponding to the cranial shape data D16. In the cranial correction target data generation process (step S200 in FIG. 4), the cranial correction target data generation unit 20 generates the cranial correction target data D20 by sequentially applying pre-processing, generative model application, and post-processing to the input three-dimensional cranial shape data D16.

[0080] Referring to FIG. 5, the cranial correction target data generation process (step S200 in FIG. 4) includes a cranial shape data input process (step S210 in FIG. 5), a pre-processing process (step S220 in FIG. 5), a generative model application process (step S230 in FIG. 5), a post-processing process (step S240 in FIG. 5), and a cranial correction target data output process (step S250 in FIG. 5).

[0081] When the cranial correction data generation process (step S200 in FIG. 4) is started, the cranial shape measurement data input process (step S210 in FIG. 5) is started. The cranial shape data input process (step S210 in FIG. 5) is a process executed by, for example, cranial correction target data generation unit 20, and cranial shape data D16 is input from storage unit 70 or the like.

[0082] The preprocessing step (step S220 in FIG. 5) is a step executed by, for example, the preprocessing unit 21, in which input skull shape data D16 is processed to generate preprocessed data so that it can be applied to the generative model D50. Examples of data processing include changing the size of an object and converting mesh data into voxel data. Voxel data is a data format that is often selected as one of the standard data formats for input and output to the generative model D50.

[0083] The object size change process described above is a process for adjusting the size of the skull shape data D16 to a "predetermined reference size" suitable for training the generative model D50. The generative model D50 is a model generated by training to adjust the skull shapes of the subject 100, which originally vary in size, to a "predetermined reference size" by scaling them. Therefore, by adjusting the size of the skull shape data D16 to which the generative model D50 is applied to the "predetermined reference size," the generative model D50 can be appropriately applied, and the accuracy of the output can be improved.

[0084] The above-mentioned mesh data to voxel data conversion process converts mesh data, which is a 3D object representation made up of a collection of vertices and polygons, into voxel data. Voxel data is easy to compare and learn because it represents an image of a 3D spatial region limited by a given coordinate size as a 3D array. The mesh data to voxel data conversion process can convert all mesh data into voxel data adjusted to the desired size (resolution), except when the mesh is so corrupted that it affects subsequent processing (for example, when a large hole exists in the model).

[0085] The mesh data is in a format that is read from the skull shape data D16 as an ".obj" file, for example. However, each file has a different number of "edges" and "points," which makes it difficult to combine the data into a single generative model D50 by machine learning, and also makes it difficult to apply the generative model D50.

[0086] Voxel data have their own nodal coordinates in an accepted coordinate system, their own form, their own state parameters that indicate their belonging to the modeled object, and the properties of the modeled region. Voxel data have the property that, because they represent values ​​on a regular grid in 3D space, the voxels themselves do not explicitly contain coordinate values, like pixels in a 2D bitmap, for example.

[0087] For these reasons, when voxelizing input mesh data in the preprocessing step, it is possible to record a value corresponding to the actual physical length (e.g., [mm]) of each side of the voxel data. The voxelized data becomes dimensionless data, but this recorded corresponding value allows it to be converted to the actual length at any time. This "corresponding value" can be called, for example, a scaling factor.

[0088] In addition, in the pre-processing step, before the input mesh data is converted into voxels, the position determined as the center of the input data (input object) may be made to coincide with the center position (0,0,0) on the three-dimensional coordinate system.

[0089] For example, in the programming language Python, the Trimesh package can be used to read ".obj" files in mesh format, and then the Mesh-to-sdf package can be used to convert the mesh data to voxel data.

[0090] The generative model application step (step S230 in FIG. 5) is, for example, a step executed by the generative model application unit 22, and obtains generated data as output for input preprocessed data.

[0091] In the generative model application process (step S230 in FIG. 5), training is performed using a deep learning network. Preprocessed data is provided as input to this network, and generative data is obtained as output from the network. For convenience of explanation, the network will be explained in detail in the learning process, and will not be explained here.

[0092] The post-processing step (step S240 in FIG. 5) is a step executed by, for example, the post-processing unit 23, and performs post-processing on the generated data generated in the generative model application step (step S230 in FIG. 5) to generate cranial correction target data D20. In the post-processing step (step S240 in FIG. 5), mesh data is converted into voxel data, and post-processing is performed to address scale changes, coordinate shifts, and holes that occur when converting voxel data into mesh data. One reason for performing post-processing is to compensate for the inconvenience of losing some of the information contained in the mesh data, such as position and object size, when the mesh data is converted into voxel data by the pre-processing unit 21.

[0093] 6, the post-processing step (step S240 in FIG. 5) includes a generated data input step (step S241 in FIG. 6), a 3D mesh reconstruction step (step S242 in FIG. 6), a size adjustment step (step S243 in FIG. 6), a hole filling processing step (step S244 in FIG. 6), and a post-processed data output step (step S245 in FIG. 6). That is, the generated data input step (step S241 in FIG. 6), the 3D mesh reconstruction step (step S242 in FIG. 6), the size adjustment step (step S243 in FIG. 6), the hole filling processing step (step S244 in FIG. 6), and the post-processed data output step (step S245 in FIG. 6) included in the post-processing step (step S240 in FIG. 5) are steps executed by the post-processing unit 23 of the main processing server 11, for example.

[0094] When the post-processing step (step S240 in FIG. 5) is started, the generated data input step (step S241 in FIG. 6) is started. The generated data input step (step S241 in FIG. 6) inputs the generated data generated in the generative model application step (step S230 in FIG. 5).

[0095] The 3D mesh reconstruction step (step S242 in FIG. 6) converts (reconstructs) the generated data as voxel data output from the generative model D50 into mesh data as a 3D mesh. For example, when using the programming language Python for the conversion, a combination of the Scikit-Image package and the Trimesh package may be used.

[0096] The voxel data output from the trained model in the generative model application process is dimensionless data (for example, data that does not have the dimension of a physical quantity such as [mm]). Therefore, as a post-processing step, for example, using the scaling factor described above, a physical quantity unit can be set for each side of the output voxel data, and the voxel data can be made to have a physical quantity.

[0097] The size adjustment step (step S243 in FIG. 6) is a step of returning the data reconstructed as mesh data to the origin and to the original size.

[0098] In the size adjustment process (step S243 in FIG. 6), the size of the reconstructed data is adjusted to be the same as the original skull shape data D16. Also, in the size adjustment process (step S243 in FIG. 6), the origin of the reconstructed data may be adjusted so that the center of the object is at the position (0,0,0).

[0099] Voxel data does not include some of the properties required for mesh data. As a result, the mesh data obtained by converting the generated data (voxel data) does not have the same object position or size as the original data (e.g., preprocessed data). For example, the center of an object in the converted mesh data may not be at (0,0,0). Furthermore, the size of an object may differ from the size of the preprocessed data or the size of the skull shape data D16.

[0100] Furthermore, since the center position of the voxel data (output object) output from the trained model in the generative model application process is not guaranteed to match the center of the input object, post-processing may be performed to align the center of the output object with the center (0,0,0) of the input object described above.

[0101] The hole filling process (step S244 in FIG. 6) is a process for searching for and closing (filling) holes that have occurred in mesh data (here, generated data whose size has been adjusted) due to the process of converting voxel data into mesh data. In the process of converting voxel data of the skull shape into mesh data in this embodiment, it is predicted that the converted mesh data will often have holes in positions such as the vertex of the head and the back of the neck, and based on this prediction, it may be possible to inspect at least these positions with a focus on filling the holes.

[0102] Furthermore, a smoothing process may be added to the hole filling process (step S244 in FIG. 6) to smooth the surface of the converted mesh data.

[0103] In the post-processed data output step (step S245 in FIG. 6), the processed mesh data is output as post-processed data in the form of an ".obj" file. The post-processed data is, in other words, cranial correction target data D20.

[0104] 5, when the post-processing step (step S240 in FIG. 5) outputs post-processed data (i.e., cranial correction target data D20), the cranial correction target data output step (step S250 in FIG. 5) outputs the cranial correction target data D20 and stores it in the storage unit 70. At this time, the cranial correction target data D20 may include some data in addition to the post-processed data. In other words, necessary data or arbitrary data may be added to the cranial correction target data D20 in the cranial correction target data output step (step S250 in FIG. 5) or the like.

[0105] Referring to FIG. 4, the cranial correction target data providing system 10 executes a helmet shape data calculation step (step S300 in FIG. 4).

[0106] The helmet shape data calculation process (step S300 in FIG. 4) is a process executed, for example, by the helmet shape data calculation unit 25, and calculates helmet shape data indicating the shape of the helmet required to correct the skull shape of the subject 100 to a shape corresponding to the cranial correction target data D20, based on the cranial correction target data D20.

[0107] In the helmet shape data calculation step (step S300 in FIG. 4), the helmet shape data may be calculated from, for example, the subject information and the cranial correction target data D20.

[0108] The helmet production data calculation process (step S310 in Figure 4) is a process that is executed, for example, by the helmet production data calculation unit 26, and is a process in which the manufacturing device 51 calculates the modeling data necessary to manufacture the cranial shape correction helmet 61 based on the subject information and helmet shape data.

[0109] In the helmet production data calculation process (step S310 in FIG. 4), the shaping data may be calculated taking into consideration the characteristics of the manufacturing apparatus 51 and the characteristics of the raw materials. Furthermore, in the helmet production data calculation process (step S310 in FIG. 4), the shaping data may be calculated taking into consideration other information.

[0110] The helmet manufacturing process (step S320 in FIG. 4) is a processing step performed by, for example, the manufacturing device 51, and manufactures the cranial shape corrective helmet 61 based on the modeling data.

[0111] In the helmet manufacturing process (step S320 in FIG. 4), for example, the outer shell 62 of the cranial shape corrective helmet 61 is manufactured using a three-dimensional printer based on the modeling data. The outer shell 62 may be manufactured as a single component or may be manufactured as a component made up of multiple components.

[0112] The cushion attachment process (step S330 in Figure 4) is a processing process performed, for example, by a manufacturing device 51, in which a cushion is produced or selected as the interior 63 of the cranial shape correction helmet 61 so that the interior 63 can be properly positioned inside the outer shell 62 of the cranial shape correction helmet 61.

[0113] The cushion attachment process (step S330 in FIG. 4) may combine the outer shell 62 and the interior lining 63 of the cranial shape corrective helmet 61, or may make the outer shell 62 and the interior lining 63 combinable, as long as the outer shell 62 and the interior lining 63 can be recognized as a set.

[0114] The shipping process (step S340 in FIG. 4) is, for example, a process executed by the shipping instruction unit 28, and causes the cranial shape correction helmet 61 manufactured by the manufacturing device 51 to be delivered to the subject 100 via the diagnoser 110. If the subject 100 is an infant, the shipping process (step S340 in FIG. 4) may also be such that the cranial shape correction helmet 61 is delivered to a person registered as a party capable of handling the cranial shape correction helmet 61, such as a relative, the doctor in charge, or a person in charge.

[0115] Next, the flow of the learning process of the cranial correction target data providing system 10 will be described with reference to FIG.

[0116] 7, the learning process of the cranial correction target data providing system 10 is a process executed by the learning unit 50, and is a process of generating a generative model D50 by machine learning. For example, the learning process includes a teacher data preparation process (step S500 in FIG. 7), a preprocessing process for learning (step S510 in FIG. 7), a machine learning process (step S520 in FIG. 7), and a generative model output process (step S530 in FIG. 7).

[0117] When the learning process is started, the teacher data preparation process (step S500 in FIG. 7) is started.

[0118] The teacher data preparation step (step S500 in FIG. 7) is, for example, a step in which the learning unit 50 acquires teacher data for any of a plurality of subjects 100 from the storage unit 17. The teacher data includes cranial shape data D16 and cranial correction target data D20, which are stored in the storage unit 70 as a set for each of the subjects 100.

[0119] In the machine learning of this embodiment, a dataset consisting of multiple pieces of teacher data is divided into two parts: a training set for training the machine learning and a validation set for verifying the results of the machine learning. For example, the ratio of the training set to the validation set can be 9:1.

[0120] The learning preprocessing step (step S510 in FIG. 7) is a step in which, for example, learning unit 50 performs learning preprocessing on the skull shape data D16 and skull correction target data D20 included in the training data. Note that the learning preprocessing step (step S510 in FIG. 7) is similar to the preprocessing step (step S220 in FIG. 7) executed by preprocessor 21, and therefore a detailed description thereof will be omitted. That is, in the learning preprocessing step (step S510 in FIG. 7), voxel data corresponding to the skull shape data D16 and skull correction target data D20 is generated.

[0121] In machine learning, the input is voxel data, and the output is also voxel data, so it is preferable that both the cranial shape data D16 and the cranial correction target data D20 be voxel data in both training and verification.

[0122] The machine learning process (step S520 in FIG. 7) is a process in which, for example, learning unit 50 performs machine learning using deep learning to generate generative model D50. Here, the machine learning is performed using an "encoder"-"decoder" type deep learning network, where the input data is pre-processed data corresponding to cranial shape data D16 and the output data is pre-processed data corresponding to cranial correction target data D20. That is, in deep learning, the encoder obtains a feature representation of condensed 3D data of cranial shape data D16 (pre-processed). Then, in deep learning, the decoder generates 3D data of cranial correction target data D20 (pre-processed) according to the feature representation obtained from the encoder.

[0123] Furthermore, in the machine learning step (step S520 in FIG. 7), a generative model D50 is generated by machine learning, learning the weights of all layers collectively from when input data is received until when output data is output.

[0124] For example, the network may consist of a contractive encoder that analyzes the entire image and a successively expanded decoder that produces a full-resolution segmentation. The network may also accept 3D data as input and process it using corresponding 3D operations, particularly 3D convolution, 3D max-pooling, and 3D up-convolution layers.

[0125] The generative model output process (step S530 in FIG. 7) is a process executed, for example, by the learning unit 50, in which the generated generative model D50 is set in the generative model application unit 22 so that the generative model application unit 22 can generate generative data.

[0126] (action)

[0127] As described above, the cranial correction target data providing system, cranial correction target data providing program, and cranial correction target data providing method of this embodiment make it possible to more suitably obtain cranial correction target data D20 suited to the case.

[0128] As described above, the cranial correction target data providing system, cranial correction target data providing program, and cranial correction target data providing method according to this embodiment provide the following effects.

[0129] (1) The cranial correction target data D20 is generated using a generative model D50 based on machine learning. In other words, by using the generative model D50, cranial correction target data D20 suited to the cranial shape of the subject 100 is generated from the cranial shape data D16 of the subject 100. Furthermore, because the generative model D50 is generated based on a large amount of training data, it reflects a variety of cases, and cranial correction target data D20 suited to each case can be obtained with high accuracy. This makes it possible to more appropriately obtain cranial correction target data D20 suited to each case.

[0130] (2) Even if the data consists of a non-three-dimensional mesh generated by the generative model D50, by performing processes to reconstruct it into a three-dimensional mesh, adjust the size, and fill in holes in the three-dimensional mesh, the cranial correction target data D20 to which the generative model D50 is applied can be made appropriate for the subject 100.

[0131] (3) The generative model D50 is generated by machine learning using the learning unit 50, in which input data is the cranial shape data of an arbitrary subject and output data is the cranial correction target data of an arbitrary subject. In other words, the generative model D50 is generated as an appropriate one based on learning of many cases (examples).

[0132] (4) The generation of the generative model D50 by machine learning is performed based on input data and output data that have been preprocessed to be suitable for learning. This increases the versatility of the generative model D50 obtained by machine learning.

[0133] (5) In deep learning performed by a network of the “encoder”-“decoder” structure type, the input data can be three-dimensional data of skull shape data D16, and the output data can be three-dimensional data of skull correction target data D20.

[0134] In the above embodiment, the cranial correction target data provision system 10 has been described as including the learning unit 50, but this is not limited thereto, and the learning unit 50 may not be included as long as the generative model D50 can be used.

[0135] In the above embodiment, the cranial correction target data providing system 10 may be configured such that the main processing server 11 is configured to be able to execute at least a storage unit 70, a pre-processing unit 21, a generative model application unit 22 to which the generative model D50 has been applied, and a post-processing unit 23, and that the cranial shape data D16 is input to these units and the cranial correction target data D20 is output. The cranial correction target data providing system 10 may also be configured to include a device that is communicably connected.

[0136] The main processing server 11 in the above embodiment may be a single processing device or may be a processing device made up of multiple devices.

[0137] In the above embodiment, the input data and output data of the generative model application unit 22 are voxel data. However, the input data and output data of the generative model application unit 22 do not have to be limited to voxel data.

[0138] For example, point cloud data constructed by sampling "points" from the faces and edges of mesh data under given conditions can be used as input data or output data for the generative model application unit 22. For the encoder processing of this point cloud data, a method for obtaining feature representations from point cloud data, such as "Charles R Qi, et al., "Pointnet: Deep learning on point sets for 3D classification and segmentation," CVPR (2017)" may be used.

[0139] In addition, the decoder processing can be realized by combining a model in which the generative model (so-called learning model) itself is configured as an implicit function representation of the output model, rather than explicitly obtaining voxel data, such as "Z. Chen, et al., 'Learning implicit fields for generative shape modeling', CVPR (2019)". [Explanation of symbols]

[0140] 10...Cranial correction target data providing system, 11...Main processing server, 12...Input unit, 13...Output unit, 14...Information processing unit, 15...Subject information management unit, 16...Shape data calculation unit, 17...Memory unit, 20...Cranial correction target data generation unit, 21...Preprocessing unit, 22...Generative model application unit, 23...Postprocessing unit, 25...Helmet shape data calculation unit, 26...Helmet production data calculation unit, 27...Cushion fitting information unit, 28...Shipping instruction unit, 31 ...Subject side terminal, 32...input unit, 33...output unit, 41...diagnostic side terminal, 42...input unit, 43...output unit, 50...learning unit, 51...manufacturing equipment, 52...input unit, 53...output unit, 61...cranial shape correction helmet, 62...outer shell, 63...interior, 70...memory unit, 100...subject, 101...cranial shape, 110...diagnostician, 121...corrected shape, D16...cranial shape data, D20...cranial correction target data, D50...generative model, NW...network.

Claims

1. a storage unit that stores cranial shape data that is data indicating the cranial shape of the subject based on the results of measuring the cranial shape of the subject; a preprocessing unit that applies preprocessing to the cranial shape data to generate preprocessed data in order to apply the cranial shape data to a generative model generated by machine learning; a generative model application unit that applies the preprocessed data to the generative model to generate generative data; a post-processing unit that performs post-processing on the generated data to generate cranial correction target data adjusted to a size corresponding to the cranial shape data, The generative model generated by the machine learning is a model obtained by learning training data corresponding to a plurality of past cases, the training data being data including the cranial shape data of an arbitrary subject and the cranial correction target data of the arbitrary subject as a set, and the generative model is a model generated by the machine learning in which the cranial shape data of the arbitrary subject is used as input data and the cranial correction target data of the arbitrary subject is used as output data. A cranial correction target data providing system.

2. the post-processing unit is configured to perform at least one of a process of reconstructing the generated data into a three-dimensional mesh, a process of adjusting a size, and a process of filling holes in the three-dimensional mesh; the machine learning is deep learning, The generative model is a model trained by collectively learning the weights of all layers from the time when the input data is input to the time when the output data is output. The cranial orthodontic target data providing system according to claim 1 .

3. further comprising a learning unit that performs the machine learning; The storage unit stores the teacher data, The learning unit is capable of generating the generative model by acquiring the training data from the storage unit, using the input data as cranial shape data of the arbitrary subject, and performing the machine learning using the output data as cranial correction target data of the arbitrary subject, and is capable of setting the generative model so that it can be applied to the generation of the generative data by the generative model application unit. The cranial orthodontic target data providing system according to claim 2 .

4. the preprocessing unit further has a configuration capable of preprocessing the input data and the output data, The learning unit is configured to perform the machine learning based on the preprocessed input data and the preprocessed output data to generate the generative model. The cranial orthodontic target data providing system according to claim 3 .

5. the skull shape data is three-dimensional data, The cranial correction target data is three-dimensional data, The deep learning has a configuration in which an encoder-decoder structure type network is applied.

5. The cranial correction target data providing system according to claim 3 or 4.

6. storing cranial shape data in a storage unit, the cranial shape data being data indicating the cranial shape based on the results of measuring the cranial shape of the subject; a step of applying preprocessing to the cranial shape data in a preprocessing unit to generate preprocessed data in order to apply the cranial shape data to a generative model generated by machine learning; a step of applying the preprocessed data to the generative model in a generative model application unit to generate generative data; and a step of generating cranial correction target data adjusted to a size corresponding to the cranial shape data by performing post-processing on the generated data in a post-processing unit, The generative model generated by the machine learning is obtained by learning training data corresponding to a plurality of past cases, and the training data is data including the cranial shape data of an arbitrary subject and the cranial correction target data of the arbitrary subject as a set, and the generative model is a model generated by the machine learning in which the cranial shape data of the arbitrary subject is used as input data and the cranial correction target data of the arbitrary subject is used as output data. A cranial correction target data providing program.

7. storing cranial shape data in a storage unit, the cranial shape data being data indicating the cranial shape based on the results of measuring the cranial shape of the subject; a step of applying preprocessing to the cranial shape data in a preprocessing unit to generate preprocessed data in order to apply the cranial shape data to a generative model generated by machine learning; a step of applying the preprocessed data to the generative model in a generative model application unit to generate generative data; and a step of generating cranial correction target data adjusted to a size corresponding to the cranial shape data by performing post-processing on the generated data in a post-processing unit, The generative model generated by the machine learning is obtained by learning training data corresponding to a plurality of past cases, and the training data is data including the cranial shape data of an arbitrary subject and the cranial correction target data of the arbitrary subject as a set, and the generative model is a model generated by the machine learning in which the cranial shape data of the arbitrary subject is used as input data and the cranial correction target data of the arbitrary subject is used as output data. A method for providing cranial correction target data.

Images