Respiratory function diagnostic device

Abstract

This respiratory function diagnostic device comprises: an image acquisition unit for acquiring a chest image as an X-ray image that includes the chest of a prediction subject (subject for whom a prediction is to be made); a respiration prediction unit for setting as an input value, with respect to a mathematical model for predicting from a chest image a respiration value serving as an index for respiration function, a chest image acquired for the prediction subject, and thereby predicting the respiration value for the prediction subject; and a learning unit for adjusting the mathematical model by using the chest image of an experimental subject as an input value and using the actual measurement value of the respiration value for the experimental subject as an output value.

Description

Respiratory function diagnostic device

[0001] The present invention relates to a technique for calculating predicted values ​​of indices related to respiratory function based on chest images.

[0002] Respiratory function testing is essential for diagnosing respiratory disorders. There is a technique called spirometry, which uses specialized equipment to perform respiratory function testing. Spirometry is a dynamic test performed while the patient inhales and exhales, allowing for functional evaluation. Spirometry can measure FVC (Forced Vital Capacity) and FEV1 (Forced Expiratory Volume (1 second)), which are typical indicators for determining the quality of respiratory function. FVC is the total volume of air exhaled when breathing in fully and then exhaling in one go. FEV1 is the volume of air exhaled in one second from the start of exhalation when breathing in fully and then exhaling in one go.

[0003] Spirometry, FVC, and FEV1 play an important role in the diagnosis of COPD (Chronic Obstructive Pulmonary Disease) and asthma. For example, FVC and FEV1 are used to diagnose and predict the prognosis of COPD. Continuous monitoring by regular spirometry is also highly recommended for the management of asthma.

[0004] Global Strategy for the Diagnosis, Management and Prevention of Chronic Obstructive Pulmonary Disease. 2023 Report: Global Initiative for Chronic Obstructive Lung Disease, 2022.Doherty DE. A Review of the Role of FEV1 in the COPD Paradigm. COPD: Journal of Chronic Obstructive Pulmonary Disease 2008; 5(5): 310-318.Global Strategy for Asthma Management and Prevention (2022 update): Global Initiative for Asthma - Global Initiative for Asthma - GINA, 2022.

[0005] As mentioned above, spirometry plays an important role in diagnosing respiratory disorders. However, spirometry is generally not performed frequently, such as during regular health checkups. Furthermore, spirometry requires the subject to inhale and exhale to the maximum extent possible. Therefore, for subjects such as young children, the elderly, or those with physical disabilities, the test may be difficult or physically demanding due to their inability to follow instructions.

[0006] On the other hand, one method for examining the lungs is to use X-ray images including the chest (hereinafter referred to as "chest imaging"). Chest imaging can be performed as long as the subject remains motionless for a short period of time. In other words, compared to spirometry, chest imaging is a static method. Therefore, it places almost no physical strain on the subject. Furthermore, chest imaging is frequently performed during regular health checkups, etc. Therefore, it would be desirable to have a respiratory function testing method based on chest imaging that can be used as an adjunct to spirometry and enable early detection of signs of respiratory disorders.

[0007] The present invention is an invention completed based on the above-mentioned viewpoints of the present inventors, and its main purpose is to provide predicted values ​​of respiratory function indicators (hereinafter referred to as "predicted respiratory values") by inputting chest images into a mathematical model.

[0008] A respiratory function diagnostic device in one aspect of the present invention includes an image acquisition unit that acquires chest images as X-ray images including the chest of a subject to be predicted, a respiratory prediction unit that predicts the respiratory value of the subject to be predicted by setting the chest image acquired for the subject to be predicted as an input value to a mathematical model that predicts a respiratory value as an index of respiratory function from the chest image, and a learning unit that adjusts the mathematical model by using the chest image of the subject to be an experimental subject as an input value and the actual measured value of the respiratory value of the subject to be experimental as an output value.

[0009] According to the present invention, a respiratory function test can be easily performed.

[0010] 1 is an overall configuration diagram of a respiratory function diagnostic system; FIG. 2 is a block diagram of a respiratory function diagnostic device; FIG. 3 is a schematic diagram showing the flow of acquiring predicted respiration values ​​based on chest images; FIG. 4 is a flowchart showing the flow of an embodiment of the present invention; FIG. 5 is a conceptual diagram of a respiration value prediction model; FIG. 6 is a flowchart showing the flow of processing in a learning stage; FIG. 7 is a flowchart showing the flow of processing in an implementation stage; FIG. 8 is a schematic diagram of a display screen related to predicted respiration values ​​of FVC and FEV1, etc.; FIG. 9 is a scatter plot showing the correlation between measured respiration values ​​and predicted respiration values; FIG. 10 is a schematic diagram showing an overview of mapping in a verification stage; FIG. 11 is a flowchart showing the flow of mapping in a verification stage; FIG. 12 is a schematic diagram showing mapping results in a verification stage; FIG. 13 is a schematic diagram showing mapping results for a healthy subject and an unhealthy subject.

[0011] First, the respiratory function diagnostic system of the present invention will be outlined. Next, the respiration value prediction model that calculates the predicted respiration value of the present invention will be outlined. Next, the details of the process performed by the respiration value prediction model will be explained in each embodiment of the learning stage, the implementation stage, and the verification stage.

[0012] First, an overview of a respiratory function diagnostic system 100 according to the present invention will be described. FIG. 1 is a diagram illustrating the overall configuration of the respiratory function diagnostic system 100. In the respiratory function diagnostic system 100, a respiratory function diagnostic device 101 is connected to a terminal 103 via a communication network 102. The terminal 103 is primarily intended to be a general-purpose computer such as a laptop PC found in a hospital examination room, but may also be a portable communication terminal such as a tablet or smartphone. The terminal 103 and the communication network 102 may be connected wirelessly or by wire. It is assumed that the devices constituting the respiratory function diagnostic system 100 are basically installed on the same premises of the same medical institution. The respiratory function diagnostic system 100 may also be provided as a so-called cloud service by installing the respiratory function diagnostic device 101 at an external service provider, etc.

[0013] 2 is a functional block diagram of the respiratory function diagnosis device 101. The components of the respiratory function diagnosis device 101 are implemented by hardware including computing units such as a CPU (Central Processing Unit) and various coprocessors, including a GPU (Graphics Processing Unit) that provides auxiliary functions to the CPU, storage devices such as memory and storage, and wired or wireless communication lines connecting these, as well as software stored in the storage device and supplying processing instructions to the computing units. The computer program may be composed of a device driver, an operating system, various application programs located at higher levels than these, and libraries that provide common functions to these programs. The blocks described below represent functional blocks, rather than hardware configurations.

[0014] The respiratory function diagnosis device 101 may be configured to include a web server, and the terminal 103 may be configured to include a portable communication terminal and a web browser installed thereon.

[0015] The respiratory function diagnosis device 101 includes a user interface processing unit 110, a data processing unit 120, a communication unit 130, and a data storage unit 140. The user interface processing unit 110 accepts operations from a user via an input interface such as a mouse, keyboard, or touch panel, and is responsible for processing related to the user interface, such as image display and audio output. It is assumed that the user of the respiratory function diagnosis device 101 is primarily a doctor. The communication unit 130 is responsible for communication processing with the terminal 103 via the communication network 102. The data storage unit 140 stores various data such as chest images 200. The data processing unit 120 performs various processes based on data acquired by the user interface processing unit 110 and the communication unit 130 and data stored in the data storage unit 140.

[0016] The user interface processing unit 110 includes an input unit 111 that receives input from the user, and an output unit 112 that outputs various types of information such as images and sounds to the user.

[0017] The communication unit 130 includes a receiving unit 131 and a transmitting unit 132. The receiving unit 131 receives a chest image 200 from the terminal 103 via the communication network 102. The transmitting unit 132 transmits various output data to the terminal 103 via the communication network 102.

[0018] The data processing unit 120 includes an image acquisition unit 121, a respiration prediction unit 122, a learning unit 123, an auxiliary diagnosis unit 124, an index generation unit 125, and an output generation unit 126. The image acquisition unit 121 acquires a chest image 200 from the data storage unit 140. The chest image 200 is an image that includes at least a portion of the lungs, and preferably an image that includes the entire lungs. The chest image 200 is assumed to be a so-called general X-ray photograph, more specifically, a two-dimensional still image of a bright-field image. The bright-field image here refers to an image that shows the internal structure of a subject (lungs) by changing the shade depending on the amount of X-rays transmitted through the subject when light (X-rays) is irradiated onto the subject. The respiration prediction unit 122 calculates a predicted respiration value based on the chest image 200 acquired by the image acquisition unit 121. Indicators for which predicted respiration values ​​are calculated include vital capacity, percent vital capacity, etc. in addition to FVC and FEV1, but hereinafter, the calculation targets for predicted respiration values ​​will be mainly FVC and FEV1. The learning unit 123 executes a learning process for a respiration value prediction model 300 (described later) so that predicted respiration values ​​can be calculated. The auxiliary diagnosis unit 124 identifies a part (hereinafter referred to as "influenced part") that has influenced the calculation result of the predicted respiration value from the chest image 200. The index generation unit 125 generates a predicted respiration value related to a new index based on the predicted respiration value calculated by the respiration prediction unit 122. The output generation unit 126 generates images, sounds, and characters for outputting information such as predicted respiration values ​​to the user.

[0019] 3 is a schematic diagram showing the flow of obtaining predicted respiration values ​​based on a chest image 200. A person who undergoes a respiratory function test according to the present invention is called a "subject." When a chest image 200 of the subject is input, the respiratory function diagnosis device 101 according to the present invention outputs a predicted respiration value of the subject. Specifically, when the chest image 200 of the subject is given, the respiratory prediction unit 122 of the respiratory function diagnosis device 101 calculates a predicted respiration value of the subject by using a respiration value prediction model 300.

[0020] 4 is a flowchart showing the flow of an embodiment of the present invention. The embodiment of the present invention is roughly divided into a "learning stage" for training the respiration rate prediction model 300 by performing a learning process, in other words, for improving calculation accuracy, and an execution stage for executing calculation processes etc. using the respiration rate prediction model 300. The execution stage is further divided into an "execution stage" for calculating predicted respiration rates using the respiration rate prediction model 300, and a "verification stage" for verifying the chest image 200 using the predicted respiration rates. The "learning stage", "execution stage", and "verification stage" will be described in detail later.

[0021] In the learning stage, a learning process is performed by inputting a large number of chest images 200 into the respiration rate prediction model 300 (S10). Hereinafter, a person who is the subject of the chest images 200 input into the respiration rate prediction model 300 for the learning process will be referred to as a "subject". By training the respiration rate prediction model 300 with chest images 200 collected from a large number of subjects and test results by spirometry, the respiration rate prediction model 300 becomes able to calculate predicted respiration rates from the chest images 200.

[0022] In the implementation stage, the user (doctor) calculates the predicted respiration value of the subject using the respiratory function diagnosis device 101. That is, the predicted respiration value is calculated by inputting the chest image 200 of the subject to the respiration value prediction model 300 that has passed the learning stage (S12). Note that Fig. 3 illustrates the process of the implementation stage. In addition, in the implementation stage, the predicted respiration value for the new index is calculated based on the predicted respiration value calculated by the respiration value prediction model 300.

[0023] In the verification stage, affected parts are identified from the chest image 200 of the subject (S14). Through the verification stage, the user (doctor) and the subject can visually understand which parts of the lungs are affecting the respiratory function.

[0024] 5 is a conceptual diagram of the respiration rate prediction model 300. In this embodiment, the respiration rate prediction model 300 is formed by a neural network. A neural network is a network formed by connecting nodes that mathematically model the functions of human brain nerve cells.

[0025] The respiration rate prediction model 300 includes an input layer 310, an intermediate layer 320, and an output layer 330. Each pixel of the chest image 200 is given to the input layer 310 as input information. Specifically, each pixel of the chest image 200 is associated one-to-one with each node (hereinafter referred to as an "input node") included in the input layer 310. In FIG. 5, the input node is represented by "I", and the input node corresponding to the n-th pixel of the chest image 200 is represented by "I ｎ In the embodiment of the present invention, the chest image 200 has 1024×1024=1048576 pixels, but for the sake of simplicity, the number of pixels is represented by "N" in FIG. 5.

[0026] The intermediate layer 320 receives input from the input layer 310 and performs calculation processing. Specifically, the nodes in the intermediate layer 320 (hereinafter referred to as "intermediate nodes") receive output values ​​from each input node as input, and determine their own output values ​​based on the weights set for the input values ​​and the thresholds set for themselves. In reality, in FIG. 5, the intermediate layer 320 is made up of multiple layers in which intermediate nodes are connected in multiple layers. In other words, the respiration rate prediction model 300 is a so-called deep learning model. The number of intermediate nodes can be set arbitrarily depending on the input and output targets of the respiration rate prediction model 300. The intermediate nodes determine their own output values ​​based on the output values ​​from the previous nodes such as the input layer.

[0027] The nodes included in the output layer 330 (hereinafter referred to as "output nodes") store the calculation results in the intermediate layer 320. In FIG. 5, the output nodes are represented by "O", and the output node corresponding to the nth type of output is represented by "O ｎ In the present invention, the output node O １ FVC, output node O ２ 5, the calculation results of the respiration rate prediction model 300 are output as FVC=X and FEV1=Y.

[0028] [1. Learning Phase] Fig. 6 is a flowchart showing the process flow in the learning phase. In the learning phase, the learning unit 123 sets the subject's chest image 200 as input to the respiration rate prediction model 300 (S16). The respiration rate prediction model 300 calculates a predicted respiration rate based on the chest image 200. In addition, the actual measured value of an index related to the subject's respiratory function (hereinafter referred to as "measured respiration rate") is set as a comparison target for the predicted respiration rate. Specifically, the learning unit 123 sets the subject's actual measured respiration rate of FVC to the predicted respiration rate of FVC (S18). In addition, the learning unit 123 sets the subject's actual measured respiration rate of FEV1 to the predicted respiration rate of FEV1 (S20). At this time, the error between the predicted respiration rate and the actual measured respiration rate for each index is calculated using MSE (mean square error) or the like, and the threshold value and weight of the intermediate node are adjusted by the error backpropagation method to reduce this error (S22). In this way, the learning stage aims to improve the prediction accuracy of the respiration rate prediction model 300. In this embodiment, 130,000 chest images 200 collected from 75,000 people at multiple medical institutions in Osaka Prefecture and spirometry test results were used as input for learning.

[0029] 7 is a flowchart showing the flow of processing in the implementation stage. In the implementation stage, the chest image 200 of the subject is given as input to the respiration rate prediction model 300, which has achieved sufficient prediction accuracy through the learning stage, to output a predicted respiration rate.

[0030] Specifically, the image acquisition unit 121 acquires one chest image 200 of the subject from the data storage unit 140. The respiration prediction unit 122 inputs the chest image 200 of the subject acquired by the image acquisition unit 121 to the respiration value prediction model 300 (S24). That is, each pixel of the chest image 200 is provided as input information to each input node of the input layer 310. Upon receiving this input, the intermediate layer 320 performs calculation processing (S26). The output layer 330 outputs the predicted respiration values ​​of FVC and FEV1 as calculation results, and the respiration prediction unit 122 acquires these (S28).

[0031] 8 is a schematic diagram of a display screen 400 relating to predicted respiration values ​​such as FVC and FEV1. The output generation unit 126 generates the display screen 400 to display the predicted respiration values ​​of FVC and FEV1 acquired by the respiration prediction unit 122 on the output unit 112. In addition to the predicted respiration values ​​of FVC and FEV1, the display screen 400 may also display the difference between these predicted respiration values ​​and the average value, the risk of lung disease, and the like. The average value may be based on the test results of all subjects or on general test results using spirometry. Subjects whose FVC and other values ​​are within the reference values ​​of spirometry are estimated to have good respiratory function.

[0032] The display screen 400 in FIG. 8 displays that a subject's predicted FVC respiration value is X and their predicted FEV1 respiration value is Y. It also displays that the predicted FVC respiration value X is lower than the average value by A, and that the predicted FEV1 respiration value Y is lower than the average value by B. Furthermore, it displays that the predicted FVC and FEV1 respiration values ​​are lower than the reference values, i.e., respiratory function is inferior to the average, and therefore the risk of pulmonary disease is somewhat high. It is well known that there is a correlation between the likelihood of restrictive pulmonary disease and the measured FVC value. Therefore, the likelihood of restrictive pulmonary disease can be predicted from the magnitude of the predicted FVC respiration value. Regarding the risk of restrictive pulmonary disease, the respiration prediction unit 122 compares the predicted respiration value with the FVC threshold value corresponding to the likelihood of disease stored in the data processing unit 120, and determines whether the risk is "high," "slightly high," "slightly low," or "low."

[0033] In the implementation stage, the index generating unit 125 generates a new index based on the predicted respiration value calculated by the respiration value prediction model 300. For example, the index generating unit 125 generates a "rate in one second" by dividing FEV1 by FVC. That is, the rate in one second indicates the ratio of the exhaled air volume in the first one second of exhalation to the total exhaled air volume when a subject takes in a maximum breath and exhales in one go. The higher the rate in one second, the greater the instantaneous movement of the lungs, allowing for stronger exhalation, and therefore the rate in one second is an index for measuring lung flexibility. The rate in one second may be displayed on the display screen 400. Various indexes for assessing the subject's health condition may be generated by further performing predetermined calculations on the multiple predicted respiration values ​​output from the respiration value prediction model 300, just like the rate in one second.

[0034] If the calculation result in the implementation stage shows that the predicted respiration value is significantly lower than the average value, that is, if an abnormality in respiratory function is suspected, the output generation unit 126 may generate a display screen 400 with a message recommending that the subject consult a specialist. This allows the subject to recognize that there may be an abnormality in their respiratory function and to take early measures to prevent respiratory disorders.

[0035] The output generation unit 126 may generate information such as predicted respiration values ​​as sound and output it from the output unit 112. The output generation unit 126 may also transfer this sound or a display screen 400 to the transmission unit 132 and output it from the terminal 103 via the communication network 102. Alternatively, the output generation unit 126 may generate a report as a deliverable that compiles information such as predicted respiration values ​​and a mapping image (described later).

[0036] FIG. 9 is a scatter diagram 500 showing the correlation between measured respiration values ​​and predicted respiration values. The scatter diagram 500 shows the results of an implementation example of the present invention, with the horizontal axis representing measured respiration values ​​and the vertical axis representing predicted respiration values. The points on the scatter diagram 500 are distributed in a generally linear pattern. In this embodiment of the present invention, a learning process was performed using spirometry results collected from 75,000 individuals at multiple medical institutions in Osaka Prefecture and a massive number of 130,000 chest images 200 collected in connection with these results. To verify the results of the learning process, the respiration value prediction model 300 was used to calculate predicted respiration values ​​for 7,000 chest images 200 collected from another 6,000 individuals (hereinafter referred to as "identification subjects") at multiple medical institutions in Osaka Prefecture. When the correlation between the measured respiration values ​​and predicted respiration values ​​of the identification subjects was confirmed, the correlation coefficient for FVC was 0.91, and the correlation coefficient for FEV1 was also 0.91. From the above, it was verified that there is a very strong correlation between the measured respiration value and the predicted respiration value, and therefore the respiration value prediction model 300 can estimate the subject's respiratory function with the same accuracy as spirometry.

[0037] 10 is a schematic diagram showing an outline of mapping in the verification stage. In the verification stage, mapping is performed to identify the part that has affected the calculation of the predicted respiration rate, i.e., the affected part, from the chest image 200.

[0038] A chest image 200 of a subject is input to the respiration rate prediction model 300, and as a result of calculation, the predicted respiration rate of FVC (hereinafter referred to as "predicted FVC") is X １ In this case, the auxiliary diagnosis unit 124 divides the chest image 200 into an arbitrary number of blocks to generate a divided image 201. The blocks may be generated in pixel units, or may be generated by grouping multiple pixels together.

[0039] Next, the auxiliary diagnosis unit 124 masks (blocks) one of the divided blocks to generate a masked image 202. The respiration prediction unit 122 calculates the predicted respiration value of FVC (hereinafter referred to as "masked FVC") in the masked image 202 using the respiration value prediction model 300. In FIG. 10, block A is masked. At this time, as a result of calculation using the respiration value prediction model 300, the masked FVC is X ２ The auxiliary diagnosis unit 124 calculates the predicted FVC=X １ and mask FVC = X ２ and calculates the difference. If the difference is equal to or greater than a predetermined threshold, that is, if block A has a large influence on the prediction, the auxiliary diagnosis unit 124 determines that the masked block A is an affected part and applies a predetermined color to this block A to visualize the magnitude of the influence. In FIG. 10, predicted FVC=X １ and mask FVC = X ２ Since the difference between the two is equal to or greater than a predetermined threshold, block A is colored. Thereafter, the auxiliary diagnosis unit 124 and the breathing prediction unit 122 perform the above series of operations on all blocks. As a result of this, the auxiliary diagnosis unit 124 generates a mapping image 203. In other words, mapping refers to generating a mapping image 203 that visualizes affected parts by coloring each block of the segmented image 210.

[0040] FIG. 11 is a flowchart showing the flow of mapping in the verification stage. Details of the mapping are provided below. First, the auxiliary diagnosis unit 124 acquires a chest image 200 and a predicted FVC from the data storage unit 140 (S30). The auxiliary diagnosis unit 124 divides the acquired chest image 200 into blocks to generate a divided image 201 (S32). The auxiliary diagnosis unit 124 also sets a block counter to 1 (S34). The block counter is a variable that indicates the number of blocks in the divided image 201 for which mapping is currently being performed. In other words, the block counter continues to increment until mapping is completed for all divided blocks.

[0041] The auxiliary diagnosis unit 124 performs masking on one block of the divided image 201 to generate a masked image 202 (S36). The respiration prediction unit 122 acquires the masked image 202 and calculates a masked FVC using the respiration rate prediction model 300 (S38).

[0042] The auxiliary diagnostic unit 124 acquires the predicted FVC and the masked FVC and calculates the difference between them (S40). The auxiliary diagnostic unit 124 compares the calculated difference with a predetermined threshold (S42). If the result of the comparison shows that the calculated difference is equal to or greater than the predetermined threshold (Y in S42), the auxiliary diagnostic unit 124 determines that the masked block is an affected area and colors this block (S44). On the other hand, if the calculated difference is equal to or less than the predetermined threshold (N in S42), the auxiliary diagnostic unit 124 does not perform any action on the masked block.

[0043] More specifically, two thresholds, a positive threshold (threshold 1) and a negative threshold (threshold 2), both of which have equal absolute values, are set. The auxiliary diagnostic unit 124 subtracts the masked FVC from the predicted FVC. If the difference is greater than threshold 1, i.e., if masking significantly reduces the predicted FVC, the auxiliary diagnostic unit 124 colors the masked block red. If the difference is less than threshold 2, i.e., if masking significantly increases the predicted FVC, the auxiliary diagnostic unit 124 colors the block blue. If the difference is between threshold 1 and threshold 2, the auxiliary diagnostic unit 124 does nothing for the block.

[0044] Next, the auxiliary diagnostic unit 124 determines whether the block counter is at its maximum value (total number of blocks) (S44). That is, it determines whether the calculation of the difference between the predicted FVC and the masked FVC has been completed for all divided blocks. If the block counter is not at its maximum value (N in S46), the auxiliary diagnostic unit 124 increments the block counter by 1 (S48). The auxiliary diagnostic unit 124 then moves on to the next block (S50) and performs masking on this block. On the other hand, if the block counter has reached its maximum value (Y in S46), mapping is complete.

[0045] Fig. 12 is a schematic diagram showing the mapping result in the verification stage. When mapping is performed on a chest image 200 for which a predicted FVC has been calculated, a mapping image 203 as shown in Fig. 12 is obtained. Blocks painted in two different colors are scattered on the mapping image 203. Furthermore, each of the colored blocks has different shades of color.

[0046] For example, in Figure 12, colored block B on the mapping image 203 indicates that the predicted FVC is larger than the mask FVC. In other words, it is identified as a positively influenced part that has had an effect of increasing the FVC. On the other hand, block C, which is colored a different color from block B, indicates that the predicted FVC is smaller than the mask FVC. In other words, it is identified as a negatively influenced part that has had an effect of decreasing the FVC. In reality, block B is colored red and block C is colored blue.

[0047] Hereinafter, blocks showing positively affected areas will be referred to as "positive blocks," blocks showing negatively affected areas will be referred to as "negative blocks," and blocks that are neither positive nor negative blocks (uncolored blocks) will be referred to as "neutral blocks."

[0048] Blocks painted darkly on the mapping image 203 indicate a large difference between the difference between the predicted FVC and the masked FVC and the predetermined threshold. In other words, they are influential parts that have a stronger influence on the FVC calculation. On the other hand, blocks painted lightly indicate a small difference between the difference between the predicted FVC and the masked FVC and the predetermined threshold. In other words, they are influential parts that have less influence on the FVC calculation. For example, in FIG. 12, blocks B and D are both positive blocks, but block B is painted darker than block D. In other words, it can be said that block B has a greater influence than block D on the calculation of the predicted FVC.

[0049] The mapping image 203 can be used to estimate the possibility of a subject suffering from a specific disease. For example, mapping images 203 are generated for multiple subjects suffering from pneumonia. When comparing the mapping images 203, it is determined that block E is a negative block in all of them. At a different time, a mapping image 203 is generated for subject P. It is assumed that block E in the mapping image 203 of subject P is a negative block. Even if subject P has not been diagnosed with pneumonia, it can be estimated that subject B is also likely to suffer from pneumonia because the same characteristics as subjects suffering from pneumonia are observed in the mapping image 203. In this way, mapping images 203 of subjects suffering from a specific disease are collected, and trends in the distribution of affected areas are identified. This allows for a simple determination of the possibility that other subjects are suffering from the same disease.

[0050] FIG. 13 is a schematic diagram showing mapping results for a healthy subject and an unhealthy subject. Visually comparing a mapping image 203 of a healthy subject with normal respiratory function (hereinafter referred to as a "healthy subject image 204") with a mapping image 203 of an unhealthy subject with abnormal respiratory function (hereinafter referred to as a "unhealthy subject image 205") can provide clues for diagnosis. For example, block F in the healthy subject image 204 is a negative block, but block F is a neutral block in the unhealthy subject image 205. Conversely, block G in the healthy subject image 204 is a neutral block, but block G is a positive block in the unhealthy subject image 205. In other words, it can be estimated that a subject in whom block F is not a negative block or block G is not a neutral block is likely to be an unhealthy subject. Thus, when visually comparing the mapping images 203, doctors can recognize areas that require attention during diagnosis based on the differences in coloring of blocks at the same position.

[0051] [Summary] The respiratory function diagnosis device 101 has been described above based on the embodiments of the learning stage, implementation stage, and verification stage. When a chest image 200 is provided as input to the respiratory function diagnosis device 101, predicted respiration values ​​such as FVC and FEV1 can be obtained through calculations using the respiration value prediction model 300. The respiratory function diagnosis device 101 also makes it possible to know lung flexibility based on the predicted respiration values ​​of FVC and FEV1. Furthermore, the respiratory function diagnosis device 101 can estimate the possibility of a subject suffering from a specific disease based on the chest image 200 and predicted respiration values.

[0052] In the past, doctors have used chest images 200 to diagnose diseases such as lung cancer and pneumothorax based on visual features seen in the chest images 200. However, estimation of information with dynamic characteristics, such as respiratory function, from a still image such as a chest image 200 has not been attempted. The idea of ​​such estimation had not been conceived before, and it was unclear what aspects of the chest image 200 to focus on and how to determine respiratory function estimation in humans. Based on this background, the inventor of the present invention hypothesized that respiratory function could be estimated by inputting the chest image 200 into a mathematical model and calculating the respiration value. In this invention, a simple examination method for respiratory function has been realized by calculating a predicted respiration value from the chest image 200 using a mathematical model using a neural network.

[0053] Chest images 200 are frequently obtained, for example, during regular medical checkups. In other words, there are many situations where it is not necessary to take the trouble to acquire chest images 200 in order to implement the examination method of the present invention. In the examination method of the present invention, chest images 200 obtained during regular medical checkups and other occasions can also be reused. In other words, since data obtained during other examinations and the like are effectively utilized as a secondary measure, the cost of the examination is inevitably reduced.

[0054] As described above, the chest image 200 is a bright-field image. In the chest image 200, tissues (affected areas), such as lung cancer, that are not present in the lungs of a healthy individual appear as shadows of irradiated light (X-rays). This allows doctors to visually confirm abnormalities in the lungs. Furthermore, the chest image 200 is a two-dimensional still image. A still image can show the structure of the lungs and affected areas with higher resolution than a moving image of the same volume, and thus, by inputting the image into the respiratory function diagnosis device 101, a more precise predicted respiration rate can be calculated. Furthermore, compared to moving images, still images can be captured in a shorter time, thereby shortening the time the subject is confined and reducing the subject's physical and mental burden.

[0055] It is recommended that respiratory function tests such as spirometry be performed regularly. Measurement of FEV1 is particularly important because a decline in FEV1 strongly influences the 10-year mortality rate of COPD. To reduce the risk of death from COPD, it is necessary to measure and monitor FEV1 frequently to prevent a decline in FEV1. However, performing spirometry frequently is cumbersome for both doctors and patients, and is not practical. On the other hand, the testing method of the present invention can be easily performed without the need for additional testing using specialized equipment, allowing for earlier detection of respiratory function abnormalities.

[0056] Note that CT (Computed Tomography) may be used to capture the chest image 200. However, since CT uses a dedicated device to irradiate the entire body with X-rays, X-ray imaging limited to the chest is preferable as a means that requires less X-ray exposure.

[0057] According to the present invention, it is also possible to diagnose subjects in remote locations. For example, clinics in remote locations, such as rural areas or remote islands, often lack dedicated equipment for respiratory function tests but have equipment capable of capturing chest images 200. Therefore, chest images 200 captured at such clinics can be sent via communication network 102 to a hospital or cloud that has a respiratory function diagnostic device 101, enabling simple respiratory function diagnosis. Furthermore, by sending the diagnostic results via communication network 102 to a terminal 103 at the remote clinic, doctors and subjects can easily check the diagnostic results.

[0058] According to the present invention, by inputting a chest image 200 into the respiration rate prediction model 300, a predicted respiration rate is calculated, and respiratory function can be estimated. Furthermore, the user (doctor) can visually check the part that the respiration rate prediction model 300 focused on when calculating the predicted respiration rate, that is, the affected part, by looking at the mapping image 203. The affected part may show a new point of view that has not been seen in previous respiratory function diagnoses by doctors. Therefore, by checking the affected part at the time of diagnosis, it is expected that the diagnostic ability of doctors will be improved.

[0059] The present invention is not limited to the above-described embodiments and modifications, and the components can be modified without departing from the spirit of the invention. Various inventions can be formed by appropriately combining multiple components disclosed in the above-described embodiments and modifications. Furthermore, some components can be omitted from all the components shown in the above-described embodiments and modifications.

[0060] [Modification] In the present invention, each pixel of the chest image 200 is described as being input to the respiration rate prediction model 300. As a modification, in addition to each pixel of the chest image 200, the subject's medical history, treatment history, smoking history, age, surgical history, gender, race, blood test results, physiological test results, pathological test results, and the like may also be provided as input. This information may be acquired by the input unit 111 of the respiratory function diagnostic device 101. Note that medical history refers to the history of illnesses the subject has suffered in the past. For example, information such as "having had COVID-19" corresponds to medical history. Furthermore, medical history refers to the history of treatments the subject has received for illnesses and the like that the subject has suffered in the past. For example, information such as "taking bronchodilators for asthma" corresponds to medical history. When conducting a respiratory function test, doctors also check information such as medical history and treatment history. Furthermore, it has been found that this information also affects respiratory function. Therefore, by inputting this information together with the chest image 200 into the respiration rate prediction model 300, the accuracy of respiration rate prediction is further improved.

[0061] In this embodiment, it has been explained that a doctor can recognize areas that should be focused on during diagnosis from the difference in coloring of blocks at the same position in both images by visually comparing the image of the healthy subject 204 and the image of the unhealthy subject 205. As a modified example, the respiratory function diagnosis device 101 may determine the possibility of a specific disease from the difference in coloring of blocks at the same position in the image of the healthy subject 204 and the image of the unhealthy subject 205.

[0062] For example, the respiratory function diagnosis apparatus 101 may include a disease determination unit (not shown). The disease determination unit calculates a correlation coefficient r as the similarity between pixel values ​​(color, density, etc.) of all pixels or pixels in a predetermined region of the healthy subject image 204 of the healthy subject A and pixel values ​​in the same target region of the unhealthy subject image 205 of the unhealthy subject suffering from COPD. More specifically, the disease determination unit calculates the correlation coefficient r by calculating a difference value 1 between the pixel value of pixel 1 in the healthy subject image 204 and the pixel value of pixel 1 at the same position in the unhealthy subject image 205, a difference value 2 between the pixel value of pixel 2 in the healthy subject image 204 and the pixel value of pixel 2 at the same position in the unhealthy subject image 205, and so on. The disease determination unit may determine that the healthy subject A has COPD if the correlation coefficient r is equal to or greater than a threshold value T. This method can reveal that individuals who appear healthy based on their appearance and behavior may actually be suffering from a specific disease, allowing doctors to take prompt action to prevent the condition from worsening.

Claims

1. A respiratory function diagnosis device comprising: an image acquisition unit that acquires chest images as X-ray images including the chest of the subject to be predicted; a respiratory prediction unit that predicts the respiratory value of the subject to be predicted by setting the chest image acquired of the subject to be predicted as an input value to a mathematical model that predicts the respiratory value as an index of respiratory function from the chest image; and a learning unit that adjusts the mathematical model by using the chest image of the subject to be an experimental subject as an input value and the actual measured respiratory value of the subject to be an experimental subject as an output value.

2. The respiratory function diagnostic device according to claim 1, wherein the chest image is a two-dimensional still image.

3. The respiratory function diagnostic device of claim 1, further comprising an auxiliary diagnostic unit that identifies locations of influence on the respiratory value based on the predicted respiratory value, using a chest image of the subject to be predicted as the target.

4. A respiratory function diagnostic device as described in claim 3, wherein the auxiliary diagnostic unit displays a mapping image showing the locations of influence on the respiration value based on a chest image of the subject to be predicted.

5. A respiratory function diagnosis device as described in claim 3 or 4, wherein the auxiliary diagnostic unit identifies the affected area based on the difference between the respiration value predicted by the mathematical model from the chest image and the respiration value predicted by the mathematical model when a portion of the chest image is obstructed.

6. A respiratory function diagnostic device as described in claim 4, further comprising a disease determination unit that determines the possibility of the subject to be assessed being affected by the specific disease by comparing the distribution of the affected areas in the mapping image of the subject suffering from the specific disease with the distribution of the affected areas in the mapping image of the subject to be assessed.

7. The respiratory function diagnostic device of claim 1, wherein the respiratory prediction unit sets, in addition to the chest image, one or more of the medical history, medical treatment history, smoking history, age, surgical history, gender, race, blood test results, physiological test results, and pathological test results of the subject to be predicted as input values ​​for the mathematical model.

8. The respiratory function diagnostic device according to claim 1, wherein the respiration prediction unit outputs multiple types of respiration values ​​based on the mathematical model.

9. A respiratory function diagnostic device as described in claim 1, wherein the respiratory prediction unit predicts the possibility of the subject suffering from restrictive pulmonary disease based on the difference between the average respiratory value obtained from a predetermined number of subjects and the predicted respiratory value.

10. The respiratory function diagnosis device of claim 8, further comprising an index generating unit that generates a new index value indicating the subject's respiratory function based on FVC and FEV1 output from the mathematical model, wherein the multiple types of respiratory values ​​include FVC (Forced Vital Capacity) and FEV1 (Forced expiratory volume (1 second)).

Images