Ultrasonic diagnostic apparatus, performance management method, and storage medium

Abstract

The invention provides an ultrasonic diagnostic apparatus, a performance management method, and a storage medium. The present invention addresses the problem of prompting an examiner to take appropriate measures by notifying the examiner of a decline in performance of an image analysis model. The recording unit (36) records the analysis operation of the image analysis model (28), and records whether or not the analysis result is adopted. A log (33) is thus generated. The calculation unit (38) calculates, on the basis of the log (33), a usage rate as a score indicating the performance of the image analysis model (28). When the usage rate is less than the threshold value, the generation unit (40) generates a reference image for notifying the examiner of the situation.

Description

Technical Field

[0001] The present invention relates to an ultrasonic diagnostic device and a performance management method, and in particular to a technology for managing the performance of an image analysis model. Background Art

[0002] Ultrasonic diagnostic apparatuses are used in ultrasonic examinations of subjects. Ultrasonic diagnostic apparatuses are medical devices that generate and display ultrasonic images using reception signals obtained by transmitting and receiving ultrasonic waves.

[0003] Recently, ultrasonic diagnostic devices with image analysis models generated through machine learning have become popular. The image analysis model is composed of, for example, a CNN (Convolutional Neural Network) that has been machine-learned. The image analysis model is, for example, a model that recognizes tissue cross-sections based on a tomographic image, or a model that measures tissue images contained in a tomographic image.

[0004] Patent Documents 1 and 2 disclose an image analysis model for identifying tissue cross sections based on tomographic images. Patent Documents 1 and 2 do not disclose a technique for managing temporal changes in the performance of the image analysis model. In addition, in this specification, the performance degradation of the image analysis model includes the performance degradation caused by the degradation of the quality of the ultrasonic image.

[0005] Patent Document 1: Japanese Patent No. 6423540

[0006] Patent Document 2: Japanese Patent Application Publication No. 2020-204970

[0007] The performance of the image analysis model depends on the quality of machine learning and varies according to the quality of the input ultrasound image. In short, when it is confirmed that the performance of the image analysis model has deteriorated, it is preferable to notify the examiner (user such as a physician or an examination technician) of such a situation and encourage the examiner to take appropriate measures. Summary of the invention

[0008] An object of the present invention is to make an inspector aware of the fact that the performance of an image analysis model has deteriorated. Alternatively, an object of the present invention is to prompt an inspector to take appropriate measures when the performance of an image analysis model has deteriorated.

[0009] The ultrasonic diagnostic device involved in the present invention is characterized in that it includes: an analysis unit, which has an image analysis model generated through machine learning, and analyzes multiple ultrasonic images in sequence, and outputs multiple analysis results in sequence; a recording unit, which records multiple analysis actions of the image analysis model, and records whether the multiple analysis results are adopted or not, thereby generating a log including a first record column composed of multiple analysis action records and a second record column composed of multiple adoption or not records; a calculation unit, which calculates a score representing the performance of the image analysis model based on the first record column and the second record column; and a generation unit, which generates reference information provided to the examiner based on the score, wherein the reference information includes at least one of information indicating a decline in performance of the image analysis model and information prompting a judgment on whether to adopt the current analysis result.

[0010] The performance management method involved in the present invention is characterized in that it includes the following steps: using an image analysis model generated by machine learning to analyze multiple ultrasonic images in sequence, thereby generating multiple analysis results in sequence; recording multiple analysis actions of the image analysis model, and recording whether the multiple analysis results are adopted or not, thereby generating a log including a first record column consisting of multiple analysis action records and a second record column consisting of multiple adoption or not records; based on the first record column and the second record column, calculating a score representing the performance of the image analysis model; and generating reference information provided to the inspector based on the score, the reference information including at least one of information indicating a decline in performance of the image analysis model and information prompting a judgment on whether to adopt the current analysis result.

[0011] Effects of the Invention

[0012] According to the present invention, when the performance of the image analysis model is degraded, the inspector can be made aware of such a situation. Alternatively, according to the present invention, when the performance of the image analysis model is degraded, the inspector is urged to take appropriate measures. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] Figure 1 This is a block diagram showing an ultrasonic diagnostic apparatus according to an embodiment.

[0014] Figure 2 This is a block diagram showing a configuration example of an image analysis unit.

[0015] Figure 3 This is a diagram showing an example of a log.

[0016] Figure 4 It is a graph showing an adoption rate graph.

[0017] Figure 5 It is a figure which shows a display example.

[0018] Figure 6 It is a flowchart showing the performance management method involved in the implementation method.

[0019] Figure 7 This is a diagram showing a first example of a pop-up window.

[0020] Figure 8 This is a diagram showing a second example of a pop-up window.

[0021] Fig. 9 is a diagram showing another pop-up window.

[0022] Fig.10 It is a diagram showing an image evaluation method.

[0023] Fig.11 This is a diagram showing another example of a log.

[0024] Explanation of symbols

[0025] 26 - image analysis unit, 28 - image analysis model, 33 - log, 36 - recording unit, 38 - calculation unit, 40 - generation unit, 42 - input image evaluation unit, 44 - action monitoring unit, 54 - first record list, 56 - second record list. DETAILED DESCRIPTION

[0026] Hereinafter, embodiments will be described based on the drawings.

[0027] (1) Overview of Implementation

[0028] The ultrasonic diagnostic device involved in the embodiment has an analyzing unit, a recording unit, a computing unit and a generating unit. The analyzing unit has an image analysis model generated by machine learning, analyzes a plurality of ultrasonic images in sequence, and outputs a plurality of analysis results in sequence. The recording unit records a plurality of analysis actions of the image analysis model, and records whether a plurality of analysis results are adopted or not, thereby generating a log including a first record column consisting of a plurality of analysis action records and a second record column consisting of a plurality of adoption or not records. The computing unit calculates a score representing the performance of the image analysis model based on the first record column and the second record column. The generating unit generates reference information provided to the examiner based on the score. The reference information includes at least one of information indicating a decline in the performance of the image analysis model and information prompting a determination as to whether the current analysis result should be adopted.

[0029] According to the above structure, when the performance of the image analysis model changes, reference information is generated and provided to the inspector. By visually recognizing the reference information, the inspector can identify the performance change of the image analysis model. This prompts the inspector to take appropriate measures. For example, prompting the correction or rejection of the analysis results, the operation of the probe to improve the quality of the ultrasound image, etc.

[0030] If the reference information is always provided to the examiner, the reference information may hinder the ultrasonic examination (for example, it is an eyesore). In an embodiment, the generating unit generates the reference information when the score is lower than a set threshold. That is, the reference information is displayed when the score is lower than the threshold.

[0031] The image analysis model is a model for identifying the type of tissue cross-section, a model for measuring tissue images, a model for determining the lesion, a model for analyzing the lesion, and the like. Each analysis action record indicates the fact that the analysis action has been performed. Each adoption or non-adoption record indicates whether the analysis result is adopted or not adopted. On the premise of recording each analysis action, an adopted record is also a non-adopted record, and conversely, a non-adopted record is also an adopted record. Therefore, it is possible to record only the adoption, only the non-adoption, or both the adoption and non-adoption. The score is information indicating the performance of the image analysis model, in other words, information indicating the reliability of the image analysis result. In an embodiment, as described above, the score is calculated based on a log, that is, based on objectively determined past performance.

[0032] In an embodiment, a calculation unit calculates an adoption rate as a fraction based on the number of analysis actions within a certain period determined from the first record column and the number of adoptions within a certain period determined from the second record column. For example, the adoption rate is calculated by dividing the number of adoptions by the number of analysis actions. The adoption rate can also be calculated by subtracting the number of non-adoptions from the number of analysis actions and dividing the value (the number of adoptions) obtained by dividing the number of analysis actions by the number of analysis actions. When the number of non-adoptions is less than the number of adoptions, it is more reasonable to record the non-adoptions. The adoption rate may also be referred to as an effective action rate, an effective usage frequency, etc. In a sense, the adoption rate is information indicating the non-adoption rate.

[0033] In the embodiment, each adoption record is a non-adoption record indicating the correction or rejection of the analysis result by the inspector. The evaluation of the analysis result is usually performed by the inspector. When the analysis result is inappropriate, the inspector corrects the analysis result or rejects the analysis result. By recording such judgments or behaviors of the inspector, a score can be calculated.

[0034] In the embodiment, each analysis action record includes information indicating the time when the analysis action is executed. Each adoption or rejection record includes information indicating the time when the input analysis result is adopted or rejected. According to this structure, the number of analysis actions and the number of adoptions to be calculated can be accurately determined.

[0035] The ultrasonic diagnostic apparatus according to the embodiment further includes an evaluation unit for determining whether each ultrasonic image input to the analysis unit satisfies a predetermined image quality condition. The number of analysis actions is the number of analysis actions corresponding to the plurality of ultrasonic images satisfying the image quality condition. The number of adoptions is the number of one or more analysis results adopted by the examiner from among the plurality of analysis results corresponding to the plurality of ultrasonic images satisfying the image quality condition.

[0036] According to the above configuration, for example, analysis errors caused by a significant decrease in the quality of an ultrasonic image can be excluded from statistical objects, so that the score can be accurately calculated.

[0037] In an embodiment, the recording unit records the analysis action corresponding to the ultrasonic image that satisfies the image quality condition, and does not record the analysis action corresponding to the ultrasonic image that does not satisfy the image quality condition. In addition, the recording unit records whether the analysis result corresponding to the ultrasonic image that satisfies the image quality condition is adopted, and does not record whether the analysis result corresponding to the ultrasonic image that does not satisfy the image quality condition is adopted.

[0038] According to the above structure, the number of records included in the log can be reduced, or unnecessary records in the log can be avoided. Of course, it is also possible to record all analysis actions and all adoptions while recording whether the image quality condition is satisfied.

[0039] In an embodiment, the first record column includes a plurality of subject identifiers correspondingly associated with a plurality of analysis action records. The second record column includes a plurality of subject identifiers correspondingly associated with a plurality of adoption or non-adoption records. The calculation unit calculates a score for each subject based on the first record column and the second record column. The content of the ultrasonic image changes according to the physique or tissue properties of the subject. According to the above structure, it is possible to appropriately determine whether to generate and display reference information for each subject.

[0040] The performance management method involved in the implementation method includes an analysis process, a recording process, a calculation process and a generation process. In the analysis process, a plurality of ultrasonic images are analyzed in sequence using an image analysis model generated by machine learning, thereby generating a plurality of analysis results in sequence. In the recording process, a plurality of analysis actions of the image analysis model are recorded, and the adoption or non-adoption of a plurality of analysis results is recorded. Thus, a log is generated including a first record column consisting of a plurality of analysis action records and a second record column consisting of a plurality of adoption or non-adoption records. In the calculation process, a score representing the performance of the image analysis model is calculated based on the first record column and the second record column. In the generation process, reference information provided to the inspector is generated based on the score. The reference information includes at least one of information indicating a decline in the performance of the image analysis model and information prompting a determination as to whether to adopt the current analysis result.

[0041] The program for executing the above-mentioned performance management method is installed in an ultrasonic diagnostic apparatus as an information processing apparatus via a network or via a portable storage medium. The ultrasonic diagnostic apparatus has a storage medium for non-temporarily storing the installed program.

[0042] (2) Details of implementation methods

[0043] Figure 1 An ultrasonic diagnostic apparatus according to an embodiment is shown in . This ultrasonic diagnostic apparatus is a medical apparatus installed in a medical institution such as a hospital and used for ultrasonic examination of a subject.

[0044] The ultrasonic probe 10 is a device that transmits ultrasonic waves into a living body and receives reflected waves from the living body. The ultrasonic probe 10 has a transducer array composed of a plurality of transducers. The transducer array forms an ultrasonic beam 12. A beam scanning plane 13 is formed by electronic scanning of the ultrasonic beam 12. The electronic scanning of the ultrasonic beam 12 is repeated according to the transmission frame rate, thereby repeatedly forming the beam scanning plane 13. Figure 1 In the figure, the r direction is the depth direction and the θ direction is the electronic scanning direction.

[0045] Electronic linear scanning and electronic sector scanning are known as electronic scanning methods for the ultrasonic beam 12. A two-dimensional transducer array may be provided as the transducer array. By two-dimensionally scanning the ultrasonic beam using the two-dimensional transducer array, volume data can be acquired from the three-dimensional space in the living body.

[0046] The transmission circuit 14 is a circuit that functions as a transmission beamformer, and outputs a plurality of transmission signals in parallel to the transducer array during transmission, thereby forming a transmission beam through the action of the transducer array.

[0047] The receiving circuit 16 is a circuit that functions as a receiving beam former. When receiving, it applies phasing and addition to a plurality of receiving signals output in parallel from the transducer array, thereby generating receiving beam data. As the electronic scanning is repeated, a receiving frame data sequence is output from the receiving circuit 16. Each receiving frame data is composed of a plurality of receiving beam data arranged in the electronic scanning direction. Each receiving beam data is composed of a plurality of echo data arranged in the depth direction.

[0048] The received frame data sequence is sent to the image forming unit 18 via a data processing unit (not shown). The data processing unit is a module that applies a plurality of processes to each received beam data. The plurality of processes include logarithmic transformation, filter processing, and the like.

[0049] The image forming unit 18 has a digital scan converter (DSC). The DSC has a coordinate conversion function, a pixel interpolation function, etc. The DSC generates a display frame data column based on the received frame data column. The display frame data column is equivalent to a tomographic image as a dynamic image. Each display frame data constituting the display frame data column is equivalent to a tomographic image as a still image.

[0050] In the illustrated configuration example, after the display frame data sequence is temporarily stored in the cine memory 20, the display frame data sequence read therefrom is sent to the display 24 via the display processing unit 22. The tomographic image is displayed as a dynamic image or as a still image on the display 24. The display processing unit 22 has an image synthesis function, a color processing function, and the like.

[0051] The cine memory 20 has a ring buffer structure. The cine memory 20 is composed of, for example, a semiconductor memory. The image forming unit 18 and the display processing unit 22 are each composed of, for example, a processor. The display 24 is composed of a liquid crystal display, an organic EL device, etc. In the image forming unit 18, ultrasonic images other than tomographic images can also be formed. For example, blood flow images, elasticity images, etc.

[0052] The image analysis unit 26 has an image analysis model 28. The image analysis model 28 is a model that has been machine-learned, and is composed of, for example, CNN. The image analysis model 28 involved in the embodiment is a model that recognizes a tissue cross section based on a tomographic image. As the image analysis model 28, a model that measures tissues based on tomographic images, a model that detects lesions included in tomographic images, etc. can be used.

[0053] In the embodiment, a display frame data column is transmitted from the cine memory 20 to the image analysis unit 26. The image analysis model 28 performs image analysis on each display frame data constituting the display frame data column (i.e., for each tomographic image), and outputs the analysis result for each display frame data. The analysis result column corresponding to the display frame data column is output from the image analysis unit 26 to the display processing unit 22 and the information processing unit 30. The analysis result column is sent to the display 24 via the display processing unit 22, and the analysis result column is displayed on the display 24. Alternatively, in the information processing unit 30, a prescribed process is performed based on the analysis result column. The image analysis unit 26 is composed of, for example, a processor. The received frame data may also be analyzed in the image analysis unit 26.

[0054] For example, the information processing unit 30 is composed of a CPU that executes a program. The information processing unit 30 functions as a control unit that controls the actions of the various elements that constitute the ultrasonic diagnostic apparatus. In the figure, multiple blocks are used to describe multiple functions performed in the information processing unit 30. Specifically, the information processing unit 30 functions as a recording unit 36, a computing unit 38, and a generating unit 40. This will be described in detail later.

[0055] The information processing unit 30 is connected to an operation panel 34. The operation panel 34 is an input device including a trackball, a plurality of switches, a plurality of knobs, etc. The operation panel 34 allows the inspector to input whether to adopt the analysis result output by the image analysis model or to correct the analysis result.

[0056] The information processing unit 30 is connected to a storage unit 32. The storage unit 32 is composed of a semiconductor memory, a hard disk, etc. In the embodiment, a log 33 for managing the temporal change of the performance of the image analysis model 28 is constructed on the storage unit 32.

[0057] The recording unit 36 ​​records the fact of the analysis action in the log 33 for each analysis action of the image analysis model 28. Furthermore, the recording unit 36 ​​records the fact of the adoption or rejection of the analysis result in the log 33 for each analysis action of the image analysis model 28. As described later, the log 33 includes a first record list and a second record list. The first record list is composed of a plurality of analysis action records arranged in chronological order. The second record list is composed of a plurality of adoption or rejection records arranged in chronological order. In the embodiment, each adoption or rejection record is actually a non-adoption record.

[0058] The calculation unit 38 calculates the adoption rate as a score indicating the performance of the image analysis model. The calculation unit 38 determines the number of analysis actions within a certain period from the current time based on the first record list, and determines the number of non-adoptions within the same certain period based on the second record list. On this basis, the calculation unit 38 calculates the adoption rate based on the number of analysis actions and the number of non-adoptions. The adoption rate is an evaluation value reflecting the performance of the image analysis model.

[0059] When the adoption rate as a score is lower than the set threshold value, the generator 40 generates a reference image as reference information. The reference image is sent to the display 24 via the display processor 22. As described later, the reference image is displayed as a pop-up window on the display 24. The reference image includes first information for notifying the inspector of the decrease in the adoption rate and second information for prompting the inspector to determine whether to adopt the analysis result. Each information is text, graphics, marks, etc.

[0060] When the adoption rate is greater than the threshold, the reference image is not displayed. In addition, the analysis result is always corrected or rejected. The information processing unit 30 can also function as the image analysis unit 26 and the display processing unit.

[0061] Figure 2 2 shows a configuration example of the image analysis unit 26. The image analysis unit 26 has an image analysis model 28 generated by machine learning. When a tomographic image is given as an input image 46 to the image analysis model 28, an analysis result 48 is output from the image analysis model 28. The analysis result 48 is, for example, a cross-section identification result.

[0062] In addition to the image analysis model 28, the image analysis unit 26 also has an input image evaluation unit 42 and an action monitoring unit 44. The action monitoring unit 44 is a module that monitors the analysis action of the image analysis model and reports the fact to the information processing unit each time the analysis action is performed. As shown by symbol 50, the analysis result can be transmitted to the information processing unit by the action monitoring unit 44. As shown by symbol 52, the image analysis model can be relearned as needed. For example, in the case where the performance of the image analysis model is reduced, it can be determined whether the image analysis model should be relearned.

[0063] The input image evaluation unit 42 is a module for evaluating the quality of each input image 46. Specifically, the input image evaluation unit 42 determines whether each input image 46 satisfies the image quality condition. The determination result is sent to the information processing unit. Only when the input image 46 satisfies the image quality condition, the analysis operation is recorded in the log, and whether it is adopted or not is recorded in the log. When all or part of a plurality of pre-set conditions are satisfied, or when at least one pre-set condition is satisfied, it can be determined that the input image 46 satisfies the image quality condition.

[0064] Figure 3 An example of a log is shown. The log 33 is composed of a first record list 54 and a second record list 56. The log 33 is updated at any time by the recording unit.

[0065] The first record list 54 is composed of a plurality of analysis action records arranged in chronological order. In the embodiment, one analysis action record is created for each analysis action of the image analysis model. Each analysis action record is actually information for determining the time when the analysis action is performed, and includes information indicating the year, month, and day and information indicating the hour, minute, and second. A plurality of analysis actions may also be recorded at a constant sampling interval.

[0066] The second record list 56 is composed of a plurality of non-adopted records 58, 60, 62 arranged in chronological order. Each non-adopted record 58, 60, 62 is information for determining the time of correction or rejection of the input analysis result. It contains information indicating the year, month, and day and information indicating the hour, minute, and second. A plurality of non-adopted records may also be recorded at a constant sampling interval.

[0067] The calculation unit calculates the non-adoption rate based on the log 33. Specifically, the calculation unit sets a time range W1, W2, W3 from the present to a certain period in the past at each calculation time T1, T2, T3. Next, the calculation unit counts the number of analysis action records contained in each time range W1, W2, W3 based on the first record list 54. The statistical result is the number of analysis actions N.

[0068] On the other hand, the calculation unit counts the number of non-adopted records contained in each time range W1, W2, and W3 based on the second record list 56. The statistical result is the non-adopted number N1. The calculation unit calculates (N-N1) / N at each calculation time T1, T2, and T3, thereby calculating the adoption rate α (reference symbol 64). The adoption rate curve graph is generated by the multiple adoption rates calculated at the multiple calculation times T1, T2, and T3.

[0069] In the above calculation formula, (N-N1) is equivalent to the number of adoptions. Assuming that each analysis action is recorded, the records not adopted are equal to the records adopted. Generally speaking, since the number of adoptions is greater than the number of non-adoptions, it is more reasonable to record non-adoptions. The above time range is equivalent to the statistical period. The calculation time range or the period of non-adoption rate can be set arbitrarily. For example, the time range can be set to 24 hours, or the period can be set to 1 hour. The period of recording analysis actions can also be set to 1 second.

[0070] Figure 4 An example of an adoption rate graph is shown. The horizontal axis is the time axis, and the vertical axis represents the adoption rate. A plurality of adoption rates are calculated at a plurality of calculation times. The plurality of adoption rates are plotted as a plurality of points 70, thereby generating an adoption rate graph 68.

[0071] A threshold value α1 is set for the adoption rate graph 68. When the calculated adoption rate (reference symbol 70a) is lower than the threshold value α1, a reference image (pop-up window) is displayed to inform the inspector of the situation. Figure 4 In the example shown, the reference image is displayed along with the analysis result after t4, as indicated by reference symbol 72. The reference image may be always displayed after t4. When the calculated adoption rate (reference symbol 70b) exceeds the threshold value α1, the reference image is not displayed.

[0072] It is also possible to generate an approximate curve 76 based on the plurality of points 70 constituting the adoption rate graph 68, and display the reference image (reference symbol 80) after the time when the approximate curve 76 falls below the threshold value α1. In this case, the display of the reference image may be terminated when the approximate curve 82 exceeds the threshold value α1. It is also possible to set a plurality of threshold values ​​α1 and α2, and switch the content or mode of the reference image according to the interval to which the calculated adoption rate belongs.

[0073] Figure 5 A display example is shown. The image 84 displayed on the display includes a tomographic image 86. Specifically, the tomographic image 86 is a frozen tomographic image. For example, at the moment when a specific cross section is recognized, the ultrasonic diagnostic device automatically becomes a frozen state. Here, the detection of the lesion is also performed. The lesion is surrounded by a frame 96.

[0074] Window 90 includes text information indicating the identified cross section. Button 88 is operated to start or end image analysis. Button 92 is operated to correct the cross section recognition result. It is equivalent to a button for rejecting the cross section recognition result. When correcting the cross section recognition result, the operation panel or touch screen panel is used.

[0075] In the embodiment, when the adoption rate is less than the threshold value, the reference image 94 is displayed together with the cross-section recognition result. The reference image 94 includes first information indicating that the adoption rate has decreased and second information prompting the determination of whether to adopt the analysis result. By displaying the reference image 94, the inspector can be made aware of the decrease in the adoption rate (or the possibility that the reliability of the analysis result has decreased), and the inspector can be prompted to take necessary measures.

[0076] Figure 6 In the embodiment, a performance management method involved in the embodiment is shown in the form of a flow chart. Figure 1 The method is performed in the ultrasonic diagnostic apparatus shown.

[0077] In S10, the input image is evaluated. If the input image satisfies the prescribed image quality condition, it is determined that the input image is a suitable image, and S12 described later is allowed to be executed, and S16 described later is allowed to be executed. On the other hand, if the input image does not satisfy the prescribed image quality condition, in S11, the record of analysis execution is restricted, and the record of correction (not adopted) is restricted. In this case, the image analysis itself may also be restricted.

[0078] In S12, the fact that the analysis was performed is recorded in the log 33. Specifically, an analysis action record is added to the log 33. In S14, the inspector corrects the analysis result. Based on such behavior, in S16, the fact that the correction (not adopted) is recorded in the log 33. Specifically, a non-adoption record is added to the log 33.

[0079] In S18, the log is referred to and the adoption rate α is calculated based on the log. In the embodiment, S18 is executed at regular intervals based on the time information generated by the timer 100. When the adoption rate α is greater than the threshold α1, the analysis result is displayed in S20. In this case, the reference image is not displayed. On the other hand, when the adoption rate α is less than the threshold α1, the analysis result is displayed in S22, and the reference image is displayed at the same time. As shown by symbol 102, the analysis result is corrected in S14 by the inspector who observes the reference image as needed.

[0080] Figure 7 and Figure 8 A specific example of a reference image is shown. Figure 7 In the first example shown, the reference image is a pop-up window 104 that is displayed in a pop-up manner. The pop-up window 104 is displayed together with the analysis result. The pop-up window 104 includes text information 106 indicating that the adoption rate has decreased and text information 108 prompting a decision on whether to adopt the current analysis result. When approving the current analysis result, a button 110 is operated, and when revising or rejecting the current analysis result, a button 112 is operated.

[0081] exist Figure 8 In the second example shown, the pop-up window 114 includes text information 116 indicating that the adoption rate has decreased and text information 118 prompting the user to determine whether to adopt the current analysis result. When approving the current analysis result, the user operates button 120, and when revising or rejecting the current analysis result, the user operates button 122. When releasing the frozen state and reacquiring images (requesting reanalysis), the user operates button 124.

[0082] In the input image evaluation unit (reference Figure 2 ) If the input image is judged to be inappropriate, it can also be displayed Fig. 9 The pop-up window 126 is shown. The pop-up window 126 is displayed together with a window showing the frozen tomographic image and the analysis result. The pop-up window 126 includes a text message 128 indicating that the analysis failed and a text message 130 requesting image confirmation. To ignore such an alarm, a button 132 is operated. To correct or reject the current analysis result, a button 134 is operated. To release the frozen state and reacquire the image (requesting re-analysis), a button 136 is operated.

[0083] Fig.10The evaluation method of the input image is shown. Reference numeral 138 indicates an image quality condition. In the example shown in the figure, the image quality condition includes a plurality of conditions 140, 142, 144, and 146. In the example shown in the figure, for example, when all the conditions 140, 142, 144, and 146 are satisfied, it is determined that the input image satisfies the image quality condition (reference numeral 148). When any one of the plurality of conditions 140, 142, 144, and 146 is not satisfied, it is determined that the input image does not satisfy the image quality condition (reference numeral 148).

[0084] In the illustrated example, condition 140 requires that the average value of the brightness contained in the input image is greater than threshold value A1. According to condition 140, analysis actions for dark images are excluded from the recording object. Condition 142 requires that the brightness variance contained in the input image is greater than threshold value B1. According to condition 142, analysis actions for blurred images are excluded from the recording object. Condition 144 requires that the SN ratio of the input image is greater than threshold value C1. According to condition 144, analysis actions for images containing a large amount of noise are excluded from the recording object. Condition 146 requires that the proportion of low-brightness pixels in all pixels constituting the input image is greater than threshold value D1. According to condition 146, analysis actions for images containing shadows are excluded from the recording object. For example, when a part of the transmitting and receiving wavefront of the ultrasonic probe is away from the body surface, a shadow is generated in the tomographic image. The above conditions are examples, and the image quality conditions can be set arbitrarily.

[0085] Fig.11 Another example of a log is shown. The log 33A is composed of a first record list 54A and a second record list 56A. The first record list 54A is composed of a plurality of analysis action records 150 arranged in chronological order and a plurality of subject identifiers 152 associated with the plurality of analysis action records 150. Each subject identifier 152 is, for example, a subject code.

[0086] The second record list 56A includes a plurality of non-adopted records 154 and a plurality of subject identifiers 156 associated with the plurality of non-adopted records. Since the first record list 54A includes a plurality of subject identifiers 152, the plurality of subject identifiers 156 can be deleted from the second record list 56A.

[0087] By building log 33A, you can use Figure 3 The method shown calculates the adoption rate as a score for each subject. The content of the input image changes according to the physique and tissue characteristics of the subject. Therefore, if the adoption rate is calculated for each subject, the reference image can be displayed at a more appropriate timing.

[0088] According to the embodiment, when the performance of the image analysis model is degraded, the inspector can be made aware of such a situation, and when the performance of the image analysis model is degraded, the inspector is urged to take appropriate measures.

Claims

1. An ultrasonic diagnostic device, characterized in that: include: An analysis unit having an image analysis model generated by machine learning, and sequentially analyzing a plurality of ultrasonic images and sequentially outputting a plurality of analysis results; a recording unit that records a plurality of analysis actions of the image analysis model and records whether the plurality of analysis results are adopted or not, thereby generating a log including a first record column consisting of a plurality of analysis action records and a second record column consisting of a plurality of adoption or not records; a calculation unit that calculates a score indicating the performance of the image analysis model based on the first record column and the second record column; and a generating unit, generating reference information to be provided to the inspector based on the score, The reference information includes at least one of information indicating that the performance of the image analysis model has been degraded and information prompting a determination as to whether to adopt a current analysis result.

2. The ultrasonic diagnostic apparatus according to claim 1, wherein: The generating unit generates the reference information when the score is lower than a set threshold.

3. The ultrasonic diagnostic apparatus according to claim 1, wherein: The calculation unit calculates an adoption rate as the score based on the number of analysis operations within a certain period specified from the first record sequence and the number of adoptions within the certain period specified from the second record sequence.

4. The ultrasonic diagnostic apparatus according to claim 1, wherein: Each acceptance / rejection record is a non-acceptance record indicating whether the analysis result has been corrected or rejected by the inspector.

5. The ultrasonic diagnostic apparatus according to claim 1, wherein: Each analysis action record includes information indicating the time when the analysis action was executed. Each adoption / non-adoption record includes information indicating the time when the input analysis result is adopted / non-adopted.

6. The ultrasonic diagnostic apparatus according to claim 3, wherein: include: an evaluation unit, which determines whether each ultrasonic image input to the analysis unit satisfies a predetermined image quality condition, The number of analysis operations is the number of a plurality of analysis operations corresponding to a plurality of ultrasonic images satisfying the image quality condition. The adoption number is the number of one or more analysis results adopted by the examiner among a plurality of analysis results corresponding to a plurality of ultrasonic images satisfying the image quality condition.

7. The ultrasonic diagnostic apparatus according to claim 6, wherein: The recording unit performs the following processing: recording analysis actions corresponding to ultrasonic images that satisfy the image quality condition, and not recording analysis actions corresponding to ultrasonic images that do not satisfy the image quality condition; and Whether the analysis result corresponding to the ultrasonic image that satisfies the image quality condition is adopted or not is recorded, and whether the analysis result corresponding to the ultrasonic image that does not satisfy the image quality condition is adopted or not is not recorded.

8. The ultrasonic diagnostic apparatus according to claim 1, wherein: The first record column includes a plurality of subject identifiers associated with the plurality of analysis action records. The second record column includes a plurality of subject identifiers correspondingly associated with the plurality of adoption / non-adoption records, The calculation unit calculates the score for each subject based on the first record sequence and the second record sequence.

9. A performance management method, characterized in that: Including the following processes: sequentially analyzing a plurality of ultrasound images using an image analysis model generated through machine learning, thereby sequentially generating a plurality of analysis results; Recording a plurality of analysis actions of the image analysis model and recording whether the plurality of analysis results are adopted or not, thereby generating a log including a first record column consisting of a plurality of analysis action records and a second record column consisting of a plurality of adoption or not records; Based on the first record column and the second record column, calculating a score representing the performance of the image analysis model; and generating reference information to be provided to the examiner based on the score, The reference information includes at least one of information indicating that the performance of the image analysis model has been degraded and information prompting a determination as to whether to adopt a current analysis result.

10. A storage medium storing a program for executing a performance management method in an ultrasonic diagnostic apparatus, wherein the program is characterized in that: The performance management method comprises the following steps: sequentially analyzing a plurality of ultrasound images using an image analysis model generated through machine learning, thereby sequentially generating a plurality of analysis results; Recording a plurality of analysis actions of the image analysis model and recording whether the plurality of analysis results are adopted or not, thereby generating a log including a first record column consisting of a plurality of analysis action records and a second record column consisting of a plurality of adoption or not records; Based on the first record column and the second record column, calculating a score representing the performance of the image analysis model; and generating reference information to be provided to the examiner based on the score, The reference information includes at least one of information indicating that the performance of the image analysis model has been degraded and information prompting a determination as to whether to adopt a current analysis result.

Images