Estimation device, estimation method, and program
The estimation device and method provide transparent AI judgments by presenting factors and reasons behind medical imaging results, addressing the 'black box' issue of machine learning models and improving user trust.
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- INSTITUTE OF SCIENCE TOKYO
- Filing Date
- 2025-11-28
- Publication Date
- 2026-06-04
AI Technical Summary
Existing machine learning models, particularly in medical imaging, operate as 'black boxes' where the factors leading to estimation results are unknown, making it difficult for users to trust and rely on the judgments, especially in critical fields like medicine and transportation.
An estimation device and method that includes a data acquisition unit, a first estimation unit to estimate factors using a first estimation model, a second estimation unit for classification, and a display unit to present both results and reasons behind the classification, utilizing pre-trained models to provide understandable factors and justifications.
Enables users to understand and validate the estimation results by providing clear factors and justifications, enhancing trust and reliability in AI-driven judgments.
Smart Images
Figure JP2025041617_04062026_PF_FP_ABST
Abstract
Description
Estimation Device, Estimation Method, and Program
[0001] The present disclosure relates to, for example, an estimation device, an estimation method, and a program capable of presenting factors (including reasons and bases) leading to an estimation result.
[0002] In recent years, the use of artificial intelligence (AI) including machine learning has been progressing in various fields such as robots, medical care, image understanding, automobiles, and speech recognition. For example, in the medical field, it is applied to assist in reading medical images that image the state inside a living body.
[0003] As an imaging method for medical images, for example, a CT (Computed Tomography) device is known. A CT device can image the X-ray absorption status of the human body and image the tissues and forms inside the human body. Also, as another imaging method, a magnetic resonance imaging (MRI: Magnetic Resonance Imaging) device is known. An MRI device, for example, applies a magnetic field to human tissues and uses the nuclear magnetic resonance (NMR: Nuclear Magnetic Resonance) phenomenon that occurs at that time to acquire two-dimensional or three-dimensional image information. An MRI device has excellent features such as being able to image tissues that cannot be imaged by a CT (Computed Tomography) device and having no radiation exposure. In these devices, methods for automatically diagnosing based on the imaged data have also been proposed (Patent Document 1).
[0004] Attempts have been made to input an image imaged by such a CT device or MRI device into a model pre-constructed by learning using machine learning to estimate lesions included in the image.
[0005] Japanese Patent Translation of PCT International Publication No. 2020-524018
[0006] However, common machine learning methods such as deep learning, as mentioned above, are end-to-end machine learning paradigms, where a single model learns everything from input data to the final result in one go. Therefore, while it is possible to obtain an estimated result by inputting unknown input data into the model, it is not possible to know the factors (including reasons and justifications) that led to that estimated result. In other words, common machine learning methods have the nature of a "black box." Furthermore, because models constructed by machine learning have complex structures, it is fundamentally difficult to theoretically analyze the internal representation of the model or analyze its internal state.
[0007] These shortcomings of machine learning can become obstacles when commercializing and implementing machine learning applications in society. In other words, if a model built using machine learning is implemented in society without understanding the basis or reasoning behind its estimation results, the model may behave unexpectedly, potentially having serious consequences in life-sustaining fields such as medicine and transportation. Furthermore, if a machine learning system provides users with judgment results without explaining the reasons or justifications, users will have no reason or basis to believe the judgment results, and therefore cannot trust or rely on the machine learning system, refusing to use it or being unable to use it effectively.
[0008] This disclosure is made in light of the circumstances described above, and aims to present the factors (including reasons and justifications) that led to the estimation when making estimations using models constructed with machine learning and artificial intelligence (AI).
[0009] An estimation device according to one aspect of the present disclosure includes: a data acquisition unit that acquires input data; a first estimation unit that inputs the input data to a first estimation model to estimate the factors of a measurement target included in the input data and outputs information indicating the estimated factors as a first estimation result; a second estimation unit that outputs a second estimation result indicating the result of classifying the measurement target based on the first estimation result; and a display unit that displays the first estimation result and the second estimation result.
[0010] An estimation device in one aspect of this disclosure is the estimation device described above, wherein the estimated factors include numerical values and attributes that represent the characteristics of the object to be measured.
[0011] An estimation device according to one aspect of the present disclosure is the estimation device described above, further comprising a report generation unit that outputs report information consisting of natural language including numerical values, indicating that the second estimation unit has come to output the second estimation result based on the first estimation result.
[0012] An estimation device according to one aspect of the present disclosure is the estimation device described above, wherein the display unit further displays the report information in addition to the first and second estimation results.
[0013] An estimation device in one aspect of the present disclosure is the estimation device described above, wherein if the portion that draws attention to the viewer of the report information consists of characters including a numerical value, the display unit displays the characters using character decoration to distinguish them from other characters.
[0014] An estimation device in one aspect of the present disclosure is the estimation device described above, wherein the character decoration includes any of the following, and any combination of some or all of the following: underlining the character; displaying the character in bold; displaying the character with a font different from the font of other characters; displaying the character with a display color different from the display color of other characters; using a background color different from the background color of other characters for the character; superimposing a predetermined pattern on the character.
[0015] An estimation device according to one aspect of this disclosure is the estimation device described above, wherein the report generation unit is configured to output the report information to an external source.
[0016] An estimation method in one aspect of the present disclosure acquires input data, inputs the input data into a first estimation model to estimate the factors of a measurement target included in the input data, outputs information indicating the estimated factors as a first estimation result, outputs a second estimation result showing the result of classifying the measurement target based on the first estimation result, and displays the first estimation result and the second estimation result.
[0017] One aspect of the present disclosure is a program that causes a computer to perform the following steps: acquire input data; input the input data into a first estimation model to estimate the factors of the object to be measured included in the input data and output information indicating the estimated factors as a first estimation result; output a second estimation result which is the result of classifying the object to be measured based on the first estimation result; and display the first estimation result and the second estimation result.
[0018] According to this disclosure, when making estimations using models built with artificial intelligence, including machine learning, it is possible to present factors including the reasons and basis for the estimation.
[0019] This is a schematic diagram showing the configuration of the estimation device according to Embodiment 1. This is a flowchart showing the estimation process in the estimation device according to Embodiment 1. This is a diagram showing an example of the estimation process in the estimation device according to Embodiment 1. This is a schematic diagram showing the configuration of the estimation device according to Embodiment 2. This is a flowchart showing the estimation process in the estimation device according to Embodiment 2. This is a diagram showing an example of the estimation process in the estimation device according to Embodiment 2. This is a diagram showing an example of the configuration of a computer for realizing the estimation device.
[0020] The following describes specific embodiments in detail with reference to the drawings. However, the invention is not limited to the embodiments described below. Also, for clarity, the following descriptions and drawings have been simplified as appropriate. Furthermore, the same elements are denoted by the same reference numerals, and redundant explanations are omitted.
[0021] Embodiment 1 This embodiment describes an estimation device that estimates the classification of an estimation target and the factors leading to the classification result for an estimation target included in the input data IN. Figure 1 is a schematic block diagram showing the configuration of the estimation device 10 according to Embodiment 1. The estimation device 10 includes a data acquisition unit 11, a factor estimation unit 12, a classification estimation unit 13, and a display unit 14.
[0022] The data acquisition unit 11 reads the input data IN. Then, it outputs the read input data IN to the factor estimation unit 12. The data acquisition unit 11 can read input data IN from various devices such as storage devices, CT scanners and MRI scanners, and various sensors such as signal sensors and image sensors.
[0023] The factor estimation unit 12 has a pre-constructed estimation model M1. The factor estimation unit 12 estimates the factors to be estimated that are included in the input data IN by inputting the input data IN into the estimation model M1. The factor estimation unit 12 then outputs information indicating the estimated factors as factor estimation result OUT1. The factors may be features described by the user for the target of estimation, or numerical values measured for the target of estimation. Hereafter, the factor estimation unit 12 will also be referred to as the first estimation unit. The estimation model M1 will also be referred to as the first estimation model. The factor estimation result OUT1 will also be referred to as the first estimation result.
[0024] At this time, the factor estimation unit 12 expresses the estimated factors in natural language that can be easily understood by the user. Furthermore, for each estimated factor, the factor estimation unit 12 expresses the degree to which the estimated target fits a predetermined factor numerically, for example, as a probability. The factor estimation unit 12 then outputs a factor estimation result OUT1 which shows the numerical values and attributes of the estimated target factor expressed in natural language extracted from the image, and the correspondence between the degree of fit of the estimated target factor to the predetermined factor.
[0025] The estimation model M1 is constructed by pre-training training data that shows the correspondence between pre-prepared input data IN and factor estimation results given as ground truth data. The estimation model M1 may be a model having various structures, such as various deep learning models, neural networks, support vector machines, regression models, and random forests. Various deep learning models may include, for example, Convolutional Neural Networks (CNN), Shift-Invariant Neural Networks, Deep Belief Networks (DBN), Deep Neural Networks (DNN), Fully Convolutional Neural Networks (FCN), U-Net, V-Net, Massive-Training Artificial Neural Network (MTANN), Multi-Resolution Massive-Training Artificial Neural Networks, Multiple Expert Massive-Training Artificial Neural Networks, SegNet, VGG-16, LeNet, AlexNet, Residual Network (ResNet), Autoencoders and decoders, Generative Adversarial Networks (GAN), Recurrent Neural Networks (RNN), Recursive Neural Networks, Long Short-Term Memory (LSTM), Transformers, and Vision Transformers.
[0026] The classification estimation unit 13 has a pre-constructed estimation model M2. The classification estimation unit 13 inputs the factor estimation result OUT1 to the estimation model M2. As a result, the estimation model M2 estimates which class the estimation target included in the input data IN belongs to based on the factors estimated by the factor estimation unit 12, and outputs the estimation result as the classification estimation result OUT2. Hereafter, the classification estimation unit 13 will also be referred to as the second estimation unit. The estimation model M2 will also be referred to as the second estimation model. The classification estimation result OUT2 will also be referred to as the second estimation result.
[0027] In this case, if the estimation result OUT1 includes multiple factors estimated by the factor estimation unit 12, the classification estimation unit 13 may select and use the factors to be used as input for the estimation model M2. For example, consider a case where the estimation result OUT1 includes N factors estimated by the factor estimation unit 12. If the estimation model M2 uses a predetermined number of M factors as input, the classification estimation unit 13 may select M factors from the N factors included in the estimation result OUT1 to be used as input for the estimation model M2, and input the selected M factors into the estimation model M2. Note that selection of factors is not mandatory, and if the estimation model M2 can input all of the multiple factors included in the estimation result OUT1, the classification estimation unit 13 may input all factors into the estimation model M2.
[0028] The estimation model M2 is constructed by pre-training it with training data that defines the correspondence between pre-prepared factor estimation results and the labels of the target to be estimated, which are given as ground truth data. The estimation model M2 may be a model with various structures, such as various deep learning models, neural networks, support vector machines, regression models, and random forests. Various deep learning models may include, for example, Convolutional Neural Networks (CNN), Shift-Invariant Neural Networks, Deep Belief Networks (DBN), Deep Neural Networks (DNN), Fully Convolutional Neural Networks (FCN), U-Net, V-Net, Massive-Training Artificial Neural Network (MTANN), Multi-Resolution Massive-Training Artificial Neural Networks, Multiple Expert Massive-Training Artificial Neural Networks, SegNet, VGG-16, LeNet, AlexNet, Residual Network (ResNet), Auto Encoders and Decoders, Generative Adversarial Networks (GAN), Recurrent Neural Networks (RNN), Recursive Neural Networks, Long Short-Term Memory (LSTM), Transformers, and Vision Transformers.
[0029] The display unit 14 displays the factor estimation result OUT1 received from the factor estimation unit 12 and the classification estimation result OUT2 received from the classification estimation unit 13 on a display device such as a display so that it can be recognized by the user.
[0030] Next, the estimation process in the estimation device 10 according to Embodiment 1 will be explained based on a specific example. Figure 2 is a flowchart showing the estimation process in the estimation device 10 according to Embodiment 1. Figure 3 is a diagram showing an example of the estimation process in the estimation device 10 according to Embodiment 1.
[0031] In this embodiment, as an example, CT image data of the subject's abdomen is used as the input data IN to be estimated. The subject to be estimated is a nodule contained in the CT image data that is suspected to be a malignant or benign tumor.
[0032] In this example, the estimation model M1 of the factor estimation unit 12 is constructed by pre-training data on the correspondence between nodule features extracted from previously acquired CT images and the types of nodules, such as whether they are malignant or benign. If a deep learning model is used for the estimation model M1, it is constructed by pre-training data on the correspondence between the CT images themselves and the types of nodules.
[0033] In this example, the estimation model M2 of the classification estimation unit 13 is constructed by pre-training it with training data that defines the correspondence between pre-acquired training factor estimation results and correct labels indicating the types of nodes.
[0034] Step S11 The data acquisition unit 11 reads the CT image to be estimated as input data IN. The data acquisition unit 11 then outputs the read input data IN to the factor estimation unit 12. The data acquisition unit 11 may read the input data IN directly from the CT device, or it may read the input data IN stored in a storage device (not shown).
[0035] Step S12 The factor estimation unit 12 inputs the input data IN to the estimation model M1. The estimation model M1 then estimates the factors from the CT image of the input data IN and expresses the estimated factors in natural language that can be easily understood by the user. The factor estimation unit 12 also provides a numerical value representing the factor and a probability (corresponding to confidence level or reliability level) for each estimated factor. Figure 3 shows an example in which the factor estimation unit 12 estimates the size, shape, contour of the nodule, and pixel density. The factor estimation unit 12 then outputs a factor estimation result OUT1, which shows the factors that determine the type of nodule estimated and the probability (corresponding to confidence level or reliability level) for the estimated factor, expressed in natural language.
[0036] Step S13 The classification estimation unit 13 inputs the factor estimation result OUT1 to the estimation model M2. The estimation model M2 then estimates what type the nodule is in the CT image of the input data IN, and outputs the estimation result and the probability of that estimation result as the classification estimation result OUT2. For example, in the example in Figure 3, the classification estimation unit 13 may estimate that the nodule is "benign" based on the factor estimation result OUT1, on the fact that the shape of the lesion is round, the size is small, and the density inside is uniform.
[0037] Step S14 The display unit 14 displays the factor estimation result OUT1 and the classification estimation result OUT2 in a way that is recognizable to the user. When displaying, a specific display method, such as text decoration, may be used to indicate each of the multiple categories to be classified. For example, for characters containing numerical values that should draw the attention of viewers of the report information, text decoration may be used to distinguish the characters containing numerical values from other characters. Text decoration may include underlining the characters, using a different font from the font of other characters, using a different display color from the display color of other characters, using a different background color from the background color of other characters, overlaying a predetermined pattern on the characters such as shading or an outer frame, or any combination of some or all of these. For example, when highlighting a category with color, by displaying a predetermined color for each description related to each category (judgment result, factors, etc.), the user can intuitively and instantly understand the judgment and its factors (basis and reason). For example, the color for Category 1 (malignant) can be red, and Category 2 (benign) can be green. The degree to which factors contribute to Category 1 can be indicated by color coding or varying the intensity of the color. Figure 6 shows an example where text is underlined. This emphasizes the category, allowing users to intuitively and quickly understand the content.
[0038] Based on the above, the estimation device 10 allows the user to receive the factors that led the classification estimation unit 13 to the classification estimation result OUT2 as the factor estimation result OUT1.
[0039] As described above, models constructed using general end-to-end machine learning can estimate the name, type, attributes, and associated numerical values of lesions by inputting data into a pre-trained model obtained through supervised learning of CT and MRI images. However, general models can only estimate the type of nodule in the input image data, and the basis or reasoning for obtaining the estimation result is unknown. Therefore, users can only know the type of nodule as an estimation result, and they cannot obtain information to judge whether the estimated type of nodule is valid. This makes it difficult for users such as doctors to trust the estimation results and make a diagnosis.
[0040] In contrast, with the estimation device 10, the user can recognize that the model has estimated the nodule to be "benign" based, for example, on the fact that the lesion is round in shape, small in size, and has uniform internal density. This allows the user to compare the estimation factors with the estimated type of nodule and consider whether the classification estimation result OUT2 is valid. This enables the user to evaluate the classification estimation result OUT2 from a professional standpoint and to determine whether the classification estimation result OUT2 is reliable.
[0041] Embodiment 2 In Embodiment 1, the factor estimation result OUT1 was provided to the user as information displaying numerical values for each factor in the classification estimation result OUT2. However, some users may find it time-consuming to analyze the correspondence between factors and numerical values, and may not be able to fully utilize the factor estimation result OUT1. For example, it is conceivable that a physician may not have time to analyze the correspondence between factors and numerical values in a clinical setting, and therefore cannot reflect the factor estimation result OUT1 in their diagnosis. Therefore, this embodiment describes an estimation device that creates a report explaining the factors leading to the classification estimation result OUT2 in natural language or the like, based on the factor estimation result OUT1.
[0042] Figure 4 is a schematic block diagram showing the configuration of the estimation device 20 according to Embodiment 2. The estimation device 20 has a configuration in which a report generation unit 21 is further provided in addition to the estimation device 10.
[0043] The report generation unit 21 has a report generation model M3 that creates report information RP explaining the correspondence between factor estimation result OUT1 and classification estimation result OUT2, based on factor estimation result OUT1 and classification estimation result OUT2.
[0044] The report generation model M3 is constructed as a natural language processing model that learns the correspondence between factor estimation results and classification estimation results prepared in advance as training data, and generates report information in natural language text that shows the correspondence. By using correspondences determined by the user for training, it becomes possible to create reports that match the user's sensibilities and are highly compatible with the user. The report generation model M3 may be a model with various structures, such as Recurrent Neural Networks (RNN), Recursive Neural Networks, Long Short-Term Memory (LSTM), Transformers, neural networks, support vector machines, regression models, random forests, Convolutional Neural Networks (CNN), Shift-Invariant Neural Networks, Deep Belief Networks (DBN), Deep Neural Networks (DNN), Fully Convolutional Neural Networks (FCN), U-Net, V-Net, Massive-Training Artificial Neural Network (MTANN), Multi-Resolution Massive-Training Artificial Neural Networks, Multiple Expert Massive-Training Artificial Neural Networks, SegNet, VGG-16, LeNet, AlexNet, Residual Network (ResNet), Auto Encoders and Decoders, Generative Adversarial Networks (GAN), and Vision Transformers.
[0045] The report creation unit 21 inputs the factor estimation result OUT1 and the classification estimation result OUT2 into the report creation model M3, and creates report information RP that explains, in natural language, the reason why the report creation unit 21 arrived at the classification estimation result OUT2 based on the factor estimation result OUT1. For example, the report creation unit 21 creates the report information RP so that it includes a sentence indicating the reason why the classification estimation unit 13 estimated that the nodule shown in the CT image of the input data IN is the type of nodule indicated by the classification estimation result OUT2 based on the factor estimation result OUT1. Note that the report information RP may include not only sentences but also numerical values, mathematical formulas, images, and charts for facilitating user understanding.
[0046] Next, the estimation process in the estimation device 10 according to the first embodiment will be described based on a specific example. FIG. 5 is a flowchart showing the estimation process in the estimation device 20 according to the second embodiment. FIG. 6 is a diagram showing an example of a report created in the estimation device 20 according to the second embodiment.
[0047] Steps S11 to S13 Steps S11 to S13 in FIG. 5 are the same as steps S11 to S13 in FIG. 2, respectively, so duplicate explanations are omitted.
[0048] Step S21 The report creation unit 21 inputs the factor estimation result OUT1 and the classification estimation result OUT2 into the report creation model M3, and creates report information RP that explains, in a sentence, the correspondence between the factor estimation result OUT1 and the classification estimation result OUT2. In FIG. 6, as an example, the report information RP1 when the tumor shown in the CT image of the input data IN is malignant and the report information RP2 when it is benign are displayed.
[0049] Step S22 The display unit 14 displays the factor estimation result OUT1, the classification estimation result OUT2, and the report information RP in a recognizable manner to the user. Note that instead of the display unit 14, it may be configured by an information transfer unit that transfers the report information RP to another system.
[0050] As described above, with the estimation device 20, the user can understand the correspondence between the factor estimation result OUT1 and the classification estimation result OUT2 in text form by viewing the report information RP. Therefore, compared to the case where only the factor estimation result OUT1 is used as in Embodiment 1, the correspondence between the factor estimation result OUT1 and the classification estimation result OUT2 can be understood more easily.
[0051] Furthermore, if it is required to record the correspondence between factor estimation result OUT1 and classification estimation result OUT2 in writing, report information RP may be recorded. This reduces the effort required for the user to create a written explanation of the analysis results of factor estimation result OUT1 and classification estimation result OUT2.
[0052] For example, if the user is a medical professional such as a doctor, they are required to record the findings of the diagnostic imaging in writing in the medical record. However, with the estimation device 20, by recording report information RP in the medical record, the amount of writing required for the doctor to record the findings can be reduced. This is expected to significantly reduce the burden on doctors in clinical practice, for example.
[0053] Other Embodiments The present invention is not limited to the embodiments described above, and can be modified as appropriate without departing from the spirit of the invention. In the embodiments described above, the factor estimation unit 12 and the classification estimation unit 13 were described as separate configurations. However, this is merely an example, and similar processing may be performed by a single configuration that encompasses the functions of the factor estimation unit 12 and the classification estimation unit 13. That is, the estimation models M1 and M2 may be separate models, or they may be integrated into a single model.
[0054] In the embodiments described above, estimation related to CT images was explained, but the images to be estimated are not limited to CT images; they may also be other medical images such as MRI images, X-ray images, ultrasound images, and nuclear medicine images. Furthermore, it is possible to apply this method to estimation of images in fields other than medical images.
[0055] Furthermore, the images are not limited to the medical field, but may also be images from various fields such as manufacturing and other industrial sectors. For example, in the industrial field, in a process of inspecting the appearance of a product using images, the images may be input to an estimation device to provide a classification estimation result OUT2 of a mode exceeding quality, and a factor estimation result OUT1.
[0056] In the embodiments described above, the input data IN to be input to the estimation device was explained as being an image, but this is merely an example. In other words, the input data IN is not limited to an image and may be various types of data. For example, the input data IN may be various types of time-series data such as electrocardiograms or process control data. The input data IN may be various types of numerical data such as blood test results or inspection results. The input data IN may be data showing the delayed sequence of DNA.
[0057] In the embodiments described above, the estimation device according to this disclosure has been described mainly as a hardware configuration, but is not limited thereto. The estimation device according to this disclosure can be realized by having a computer execute a computer program to perform any processing. These processing may be realized by having a computer, which includes at least one processor (e.g., a microprocessor, CPU, GPU, MPU, or DSP (Digital Signal Processor)), execute a program. Specifically, one or more programs including a set of instructions for causing a computer to perform algorithms related to these transmission signal processing or reception signal processing can be created and supplied to the computer.
[0058] Computer programs can be stored and supplied to a computer using various types of non-transitor computer-readable media. Non-transitor computer-readable media include various types of tangible storage media. Examples of non-transitor computer-readable media include magnetic recording media (e.g., flexible disks, magnetic tapes, hard disk drives), magneto-optical recording media (e.g., magneto-optical disks), CD-ROMs (Read Only Memory), CD-Rs, CD-R / Ws, and semiconductor memory (e.g., mask ROMs, PROMs (Programmable ROMs), EPROMs (Erasable PROMs), flash ROMs, RAMs (random access memory)). Furthermore, programs may be supplied to the computer by various types of temporary computer-readable media. Examples of temporary computer-readable media include electrical signals, optical signals, and electromagnetic waves. Temporary computer-readable media can be supplied to the computer via wired communication channels such as electric wires and optical fibers, or via wireless communication channels.
[0059] The following shows an example of a computer configuration for realizing the estimation device according to the above-described embodiment. Figure 7 is a diagram showing an example of a computer configuration for realizing the estimation device. The estimation device can be realized by a computer 9000 such as a dedicated computer or a personal computer (PC). However, the computer does not need to be physically single; there may be multiple computers when performing distributed processing. As shown in Figure 7, the computer 9000 has, for example, a processor 9001, a ROM (Read Only Memory) 9002, a RAM (Random Access Memory) 9003, a storage unit 9004, a communication interface 9005, and a user interface 9006.
[0060] The processor 9001, ROM 9002, RAM 9003, storage unit 9004, communication interface 9005, and user interface 9006 are connected to each other via bus 9007 so that they can communicate with one another. While the OS software and other components necessary to operate the computer are not described here, they are also installed in computer 9000 as appropriate.
[0061] ROM is composed of, for example, a non-volatile semiconductor memory device. ROM 9002 stores information such as various programs used in computer 9000.
[0062] The storage unit 9004 is composed of various storage devices, such as hard disks and solid-state disks. Furthermore, the storage unit 9004 is not limited to storage devices installed in the computer 9000, but may also be external storage devices. External storage devices may include various communication means, such as cloud storage connected to the computer 9000 via a network. The storage unit 9004 stores information such as various programs and data used by the computer 9000.
[0063] The RAM 9003 is composed of volatile semiconductor memory devices and the like. Programs and data used by the processor 9001 are loaded into the RAM 9003 as needed from either or both of the ROM 9002 and the storage unit 9004.
[0064] The processor 9001 may be composed of, for example, a CPU (Central Processing Unit). Alternatively, the processor 9001 may include a GPU (Graphics Processing Unit) in addition to the CPU. A GPU is suitable for parallel processing of routine tasks, and can improve processing speed compared to a CPU, for example, by being used in neural network processing. The processor 9001 executes various processes based on various programs stored in the ROM 9002, or various programs and data held in the RAM 9003. The processor 9001 may also store the data created by the processing in the RAM 9003 or the storage unit 9004 as appropriate.
[0065] The communication interface 9005 is an interface that connects the computer 9000 to a communication network such as the Internet or an intranet via various wired or wireless communication means. This allows the computer 9000 to communicate with other devices, systems, and sensors connected to the communication network.
[0066] The user interface 9006 includes, for example, a display unit that provides information to the user in a way that the user can perceive, such as through a display device, and an audio output unit that provides audio. The user interface 9006 also includes an input unit that allows the user to input information to the computer 9000 through user operation, such as a keyboard, mouse, and touch panel. Furthermore, the user interface 9006 may include devices such as sensors that acquire information useful to the user.
[0067] Here, the computer 9000 has been described as a single device, but this is merely an example. The computer 9000 may consist of multiple physically separated devices. Some of these devices may be portable, while others may be stationary.
[0068] Each drawing is merely illustrative to illustrate one or more embodiments. Each drawing may be associated with one or more other embodiments, rather than being associated with only one specific embodiment. As those skilled in the art will understand, various features or steps described with reference to any one drawing can be combined with features or steps shown in one or more other drawings, for example, to create embodiments not explicitly shown or described. Not all features or steps shown in any one drawing to illustrate an exemplary embodiment are necessarily required, and some features or steps may be omitted. The order of steps described in any of the drawings may be changed as appropriate.
[0069] This application claims priority based on Japanese Patent Application No. 2024-208313, filed on 29 November 2024, and incorporates all of its disclosures herein.
[0070] 10, 20 Estimation device 11 Data acquisition unit 12 Factor estimation unit 13 Classification estimation unit 14 Display unit 21 Report creation unit 9000 Computer 9001 Processor 9002 ROM 9003 RAM 9004 Storage unit 9005 Communication interface 9006 User interface 9007 Bus IN Input data M1 Estimation model M2 Estimation model M3 Report creation model OUT1 Factor estimation result OUT2 Classification estimation result RP Report information
Claims
1. An estimation device comprising: a data acquisition unit that acquires input data; a first estimation unit that inputs the input data into a first estimation model to estimate the factors of the object to be measured included in the input data and outputs information indicating the estimated factors as a first estimation result; a second estimation unit that outputs a second estimation result indicating the result of classifying the object to be measured based on the first estimation result; and a display unit that displays the first estimation result and the second estimation result.
2. The estimation device according to claim 1, wherein the estimated factors include numerical values and attributes that indicate the characteristics of the object to be measured.
3. The estimation apparatus according to claim 1 or 2, further comprising a report generation unit that outputs report information consisting of natural language including numerical values, indicating that the second estimation unit has come to output the second estimation result based on the first estimation result.
4. The estimation device according to claim 3, wherein the display unit further displays the report information in addition to the first and second estimation results.
5. The estimation device according to claim 4, wherein, if the portion of the report information intended to draw the viewer's attention consists of characters including numerical values, the display unit displays the characters using character decoration to distinguish them from other characters.
6. The estimation device according to claim 5, wherein the character decoration includes any of the following, and any combination of some or all of the following: underlining the character; displaying the character in bold; displaying the character in a font different from the font of other characters; displaying the character in a display color different from the display color of other characters; using a background color different from the background color of other characters for the character; and superimposing a predetermined pattern on the character.
7. The estimation device according to claim 3, wherein the report generation unit is configured to output the report information to an external source.
8. An estimation method comprising: acquiring input data; inputting the input data into a first estimation model to estimate the factors of the object to be measured included in the input data; outputting information indicating the estimated factors as a first estimation result; outputting a second estimation result as the result of classifying the object to be measured based on the first estimation result; and displaying the first estimation result and the second estimation result.
9. A program that causes a computer to perform the following steps: acquire input data; input the input data into a first estimation model to estimate the factors of the object to be measured included in the input data, and output information indicating the estimated factors as a first estimation result; output a second estimation result which is the result of classifying the object to be measured based on the first estimation result; and display the first estimation result and the second estimation result.