A lung ultrasound image detection system

Abstract

An embodiment of the present specification provides a lung ultrasound image detection system, which includes: an acquisition module, used to acquire an ultrasound image video of an object; a judgment module, used to process an image sequence based on a trained first prediction model to determine a first judgment result; wherein the image sequence is composed of multiple frames of images in the ultrasound image video in the order of shooting time; the first judgment result includes whether multiple symbols related to pneumothorax exist, and the position of the existing symbols in at least one frame of the multiple frames; the multiple symbols include at least a lung point sign and a lung sliding sign.

Description

Technical Field

[0001] The present invention relates to the field of medical image processing, and in particular to a lung ultrasound image detection system. Background Art

[0002] Among traumatic deaths, about 20% are mainly caused by chest trauma. Therefore, early detection and timely and correct treatment of pneumothorax are very important. At present, chest ultrasound imaging is mainly performed by ultrasound experts manually or based on lesion detection of a single ultrasound image, and clinical applications need to further improve sensitivity and specificity.

[0003] Therefore, it is necessary to establish a popular, easy-to-implement, and effective lung ultrasound image detection system to achieve real-time detection and identification of dynamic ultrasound pneumothorax that meets clinical needs in clinical medical applications. Summary of the invention

[0004] One of the embodiments of the present specification provides a lung ultrasound image detection system, which includes: an acquisition module, used to acquire an ultrasound image video of an object; a judgment module, used to process an image sequence based on a trained first prediction model to determine a first judgment result; wherein the image sequence is composed of multiple frames of images in the ultrasound image video in the order of shooting time; the first judgment result includes whether multiple symbols related to pneumothorax exist, and the position of the existing symbols in at least one frame of the multiple frames; the multiple symbols include at least a lung point sign and a lung sliding sign.

[0005] One of the embodiments of the present specification provides a lung ultrasound image detection device, which includes a processor and a memory; the memory is used to store instructions, and when the instructions are executed by the processor, the device causes the device to implement the functions corresponding to the lung ultrasound image detection system.

[0006] One of the embodiments of the present specification provides a computer-readable storage medium, wherein the storage medium stores computer instructions. When a computer reads the computer instructions in the storage medium, the computer runs the functions corresponding to the lung ultrasound image detection system. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] This specification will be further described in the form of exemplary embodiments, which will be described in detail by the accompanying drawings. These embodiments are not restrictive, and in these embodiments, the same number represents the same structure, wherein:

[0008] Figure 1 is a schematic diagram of an application scenario of a lung ultrasound image detection system according to some embodiments of this specification;

[0009] Figure 2 is an exemplary flow chart of a lung ultrasound image detection system according to some embodiments of this specification;

[0010] Figure 3 is an exemplary flow chart of a lung ultrasound image detection system according to some embodiments of this specification;

[0011] Figure 4 is a schematic diagram of a static detection model according to some embodiments of this specification;

[0012] Figure 5 is a schematic diagram of a dynamic detection model according to some embodiments of this specification;

[0013] Figure 6 is a schematic diagram of a second prediction model according to some embodiments of this specification;

[0014] Figure 7 is a schematic diagram of a third model according to some embodiments of the present specification;

[0015] Figure 8 It is a schematic diagram of the model training method shown in some embodiments of this specification. DETAILED DESCRIPTION

[0016] In order to more clearly illustrate the technical solutions of the embodiments of this specification, the following is a brief introduction to the drawings required for the description of the embodiments. Obviously, the drawings described below are only some examples or embodiments of this specification. For ordinary technicians in this field, this specification can also be applied to other similar scenarios based on these drawings without creative work. Unless it is obvious from the language environment or otherwise explained, the same reference numerals in the figures represent the same structure or operation.

[0017] It should be understood that the "system", "device", "unit" and / or "module" used herein are a method for distinguishing different components, elements, parts, portions or assemblies at different levels. However, if other words can achieve the same purpose, the words can be replaced by other expressions.

[0018] As shown in this specification and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "comprise" and "include" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.

[0019] Flowcharts are used in this specification to illustrate the operations performed by the system according to the embodiments of this specification. It should be understood that the preceding or following operations are not necessarily performed precisely in order. Instead, the steps may be processed in reverse order or simultaneously. At the same time, other operations may also be added to these processes, or one or more operations may be removed from these processes.

[0020] The embodiments of the present application relate to a lung ultrasound image detection system. The lung ultrasound image detection system can be applied to the detection process of clinical, rescue, wartime treatment, accident site first aid, rescue, etc. In some embodiments, the lung ultrasound image detection system can also be applied to other fields. For example, in some embodiments, the lung ultrasound image detection system can be applied to detection or big data analysis in the fields of security and protection. Through the lung ultrasound image detection system, one or more functions can be achieved: automatic detection and identification of pneumothorax, real-time dynamic analysis, non-pause detection, bedside ultrasound examination of patients who are not suitable for movement, first aid on-site inspection, etc. The lung ultrasound image detection system can achieve one or more beneficial effects such as improving detection efficiency and diagnostic accuracy, reducing dependence on doctors' experience level and detection equipment, and winning precious time for on-site first aid.

[0021] Figure 1 It is a schematic diagram of an application scenario of a lung ultrasound image detection system according to some embodiments of this specification.

[0022] like Figure 1 As shown, the lung ultrasound image detection system 100 may include a server 110 , a processor 112 , a terminal 120 , an ultrasound detection device 130 , a storage device 140 , and a network 150 .

[0023] In some embodiments, the server 110 can be used to process information and / or data related to the lung ultrasound image detection system 100, for example, to obtain ultrasound data and determine the detection results. In some embodiments, the server 110 can be a single server or a server group. The server group can be centralized or distributed (for example, the server 110 can be a distributed system). In some embodiments, the server 110 can be local or remote. For example, the server 110 can access information and / or data stored in the terminal 120, the ultrasound detection device 130, and the storage device 140 via the network 150. For another example, the server 110 can be directly connected to the terminal 120, the ultrasound detection device 130 and / or the storage device 140 to access the stored information and / or data. In some embodiments, the server 110 can be implemented on a cloud platform or provided in a virtual manner. By way of example only, the cloud platform can include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof.

[0024] In some embodiments, the server 110 may include a processor 112. The processor 112 may process information and / or data related to the lung ultrasound image detection system 100 to perform one or more functions described in the present application. For example, the processor 112 may acquire an ultrasound image video of an object and determine a detection result based on the ultrasound image video. In some embodiments, the processor 112 may include one or more processing engines (e.g., a single-chip processing engine or a multi-chip processing engine). By way of example only, the processor 112 may include a central processing unit (CPU), an application-specific integrated circuit (ASIC), an application-specific instruction set processor (ASIP), a graphics processing unit (GPU), a physical processing unit (PPU), a digital signal processor (DSP), a field programmable gate array (FPGA), a programmable logic device (PLD), a controller, a microcontroller unit, a reduced instruction set computer (RISC), a microprocessor, etc., or any combination thereof.

[0025] Terminal 120 refers to one or more terminal devices or software used by the user. In some embodiments, the terminal 120 may be used by one or more users, and the users may include testing department personnel, emergency personnel, technical experts, etc. In some embodiments, the test results may be presented to the user through the terminal 120. In some embodiments, a request for retesting may be obtained through the terminal 120. The reasons and methods for retesting are described in detail in the following sections. Figure 4 In some embodiments, the terminal 120 may be one of the mobile device 120-1, tablet computer 120-2, laptop computer 120-3, desktop computer 120-4, or any combination thereof, which have input and / or output functions. In some embodiments, the mobile device 120-1 may include a mobile phone, a smart phone, a personal digital assistant (PDA), a navigation device, a handheld terminal (POS), or any combination thereof. In some embodiments, the desktop computer 120-4 may be a vehicle-mounted computer, a vehicle-mounted TV, or the like.

[0026] The ultrasonic detection device 130 refers to a device that uses the propagation law of ultrasonic waves in a medium to obtain the detection image / video of the object, such as an ultrasonic pulse echo imaging device, an ultrasonic echo Doppler imaging device, an ultrasonic electronic endoscope, an ultrasonic Doppler blood flow analysis device, an ultrasonic human tissue measurement device, etc. The object is the object to be detected, such as a patient or a wounded person. In some embodiments, the object can undergo lung ultrasound examination in any position, such as supine position, lateral position, prone position, semi-recumbent position or sitting position. In some embodiments, the scanning mode of the ultrasonic detection device 130 may include A-ultrasound, B-ultrasound, M-ultrasound and D-ultrasound. In some embodiments, the ultrasonic detection device 130 can be set in a medical place or facility, such as a ward, a delivery room, an examination room, an operating room, a rescue room, an ambulance, etc. In some embodiments, the ultrasonic detection device 130 can be set in other places, such as a marathon venue, an extreme sports venue, a racing venue, a disaster relief site, etc. In some embodiments, the ultrasonic detection device 130 can transmit the ultrasonic image video to the server 110 via the network 150.

[0027] The storage device 140 may be used to store data and / or instructions related to the lung ultrasound image detection system 100. In some embodiments, the storage device 140 may store data obtained / acquired from the terminal 120 and / or the ultrasound detection device 130. In some embodiments, the storage device 140 may store historical data, case data, training samples, etc. In some embodiments, the storage device 140 may store data and / or instructions used by the server 110 to execute or use to complete the exemplary methods described in this application. In some embodiments, the storage device 140 may include a large capacity memory, a removable memory, a volatile read-write memory, a read-only memory (ROM), etc. or any combination thereof. Exemplary large capacity memory may include a magnetic disk, an optical disk, a solid-state disk, etc. Exemplary removable memory may include a flash drive, a floppy disk, an optical disk, a memory card, a compressed disk, a magnetic tape, etc. Exemplary volatile read-write memory may include a random access memory (RAM). Exemplary RAM may include dynamic random access memory (DRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), static random access memory (SRAM), thyristor random access memory (T-RAM) and zero capacitance random access memory (Z-RAM), etc. Exemplary read-only memory may include model read-only memory (MROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), compact disk read-only memory (CD-ROM) and digital versatile disk read-only memory, etc. In some embodiments, the storage device 140 may be implemented on a cloud platform. By way of example only, the cloud platform may include a private cloud, a public cloud, a hybrid cloud, a community cloud, a distributed cloud, an internal cloud, a multi-layer cloud, etc. or any combination thereof. In some embodiments, the storage device 140 may be connected to a network 150 to communicate with one or more components (e.g., server 110, terminal 120, ultrasound detection device 130) of the lung ultrasound image detection system 100. One or more components of the lung ultrasound image detection system 100 may access data or instructions stored in the storage device 140 via the network 150. In some embodiments, the storage device 140 may be directly connected to or communicate with one or more components (e.g., the server 110, the terminal 120, the ultrasound detection device 130) of the lung ultrasound image detection system 100. In some embodiments, the storage device 140 may be part of the server 110. In some embodiments, the storage device 140 may be a separate memory.

[0028] The network 150 can facilitate the exchange of information and / or data. In some embodiments, one or more components of the lung ultrasound image detection system 100 (e.g., the server 110, the ultrasound detection device 130, the storage device 140) can send information and / or data to other components of the lung ultrasound image detection system 100 via the network 150. For example, the ultrasound detection device 130 can send ultrasound image videos to the server 110 via the network 150. In some embodiments, the network 150 can be a wired network or a wireless network, etc., or any combination thereof. As just an example, the network 150 may include a cable network, a wired network, an optical fiber network, a telecommunication network, an intranet, the Internet, a local area network (LAN), a wide area network (WAN), a wireless local area network (WLAN), a metropolitan area network (MAN), a public switched telephone network (PSTN), a Bluetooth network, a ZigBee network, a near field communication (NFC) network, a global system for mobile communications (GSM) network, a code division multiple access (CDMA) network, a time division multiple access (TDMA) network, a general packet radio service (GPRS) network, an enhanced data rate for GSM evolution (EDGE) network, a wideband code division multiple access (WCDMA) network, a high speed downlink packet access (HSDPA) network, a long term evolution (LTE) network, a user datagram protocol (UDP) network, a transmission control protocol / Internet protocol (TCP / IP) network, a short message service (SMS) network, a wireless application protocol (WAP) network, an ultra-wideband (UWB) network, infrared, etc. or any combination thereof. In some embodiments, the lung ultrasound image detection system 100 may include one or more network access points. For example, base stations and / or wireless access points 150 - 1 , 150 - 2 , . . . , one or more components of the lung ultrasound image detection system 100 may be connected to the network 150 to exchange data and / or information.

[0029] It should be noted that the lung ultrasound image detection system 100 is provided for illustrative purposes only and is not intended to limit the scope of the present application. For those of ordinary skill in the art, various modifications or variations can be made according to the description of the present application. For example, the lung ultrasound image detection system 100 may also include a database. For another example, the lung ultrasound image detection system 100 may implement similar or different functions on other devices. However, these changes and modifications will not deviate from the scope of the present application.

[0030] In some embodiments, the lung ultrasound image detection system 100 may include an acquisition module, a judgment module, an update module, and a determination module.

[0031] The acquisition module can be used to acquire the ultrasonic image video of the object.

[0032] The judgment module can be used to process the image sequence based on the trained first prediction model to determine a first judgment result; wherein the image sequence is composed of multiple frames of images in the ultrasound image video in the order of shooting time; the first judgment result may include whether multiple symbols related to pneumothorax exist, and the position of the existing symbols in one or more frames of the multiple frames; the one or more symbols include at least a lung point sign and a lung sliding sign. In some embodiments, the judgment module can also be used to determine a second judgment result of whether the ultrasound image video needs to be re-tested based on preset rules and the first judgment result, wherein the preset rules are related to the pneumothorax judgment; if the second judgment result is yes, based on the symbol gold standard level of the one or more symbols, one or more target symbols that need to be re-tested are determined from the one or more symbols.

[0033] The update module can be used to obtain one or more second prediction models corresponding to one or more target symbols, respectively. For any one of the one or more second prediction models, the second prediction model can be used to predict whether the corresponding target symbol exists; based on the second prediction model, the target image sequence is processed to determine a third judgment result of whether the target symbol exists; based on the third judgment result, the first judgment result is updated. In some embodiments, the update module can also be used for each of the one or more symbols, if the first judgment result includes the existence of the above-mentioned symbol, to obtain at least one frame or multiple frames of images in which the above-mentioned symbol exists, and to construct a target image sequence in chronological order based on the one or more frames of images in which the symbol exists; if the first judgment result includes the non-existence of the above-mentioned symbol, the image sequence composed of multiple frames of images in the ultrasound image video is used as the target image sequence.

[0034] The determination module can be used to process the first judgment result, and one or more symbolic characteristic values ​​and sensitivity values ​​based on the third model to determine the probability of pneumothorax of the object.

[0035] It should be noted that the above description of the candidate display, lung ultrasound image detection system and its modules is only for convenience of description and cannot limit this specification to the scope of the embodiments. It is understandable that for those skilled in the art, after understanding the principle of the system, it is possible to arbitrarily combine the modules or form a subsystem to connect with other modules without deviating from this principle. In some embodiments, Figure 1 The judgment module and update module disclosed in the specification can be different modules in a system, or one module can realize the functions of two or more modules. For example, each module can share a storage module, or each module can have its own storage module. Such variations are all within the protection scope of this specification.

[0036] Figure 2is an exemplary flow chart of a lung ultrasound image detection system according to some embodiments of the present specification.

[0037] Step 210: Acquire an ultrasonic video of the object.

[0038] For a description of the object, see Figure 1 .

[0039] In some embodiments, the acquisition module can acquire the ultrasonic image video collected by the ultrasonic detection device 130 in real time (eg, every 0.02 seconds, 0.01 seconds, etc.).

[0040] In some embodiments, the acquisition module may periodically (eg, every 10 seconds, 20 seconds, etc.) acquire the ultrasonic image video collected by the ultrasonic detection device 130 .

[0041] Step 220: Process the image sequence based on the trained first prediction model to determine a first judgment result.

[0042] An image sequence is a sequence of multiple frames of images in an ultrasound video in the order in which they are taken.

[0043] The first judgment result includes whether multiple symbols related to pneumothorax exist, and the position of the existing symbols in at least one frame of the multiple frames. For example, a certain ultrasound video has a lung point sign but no lung sliding sign, and the lung point sign is located in the lower left corner of the 87th frame of the ultrasound video.

[0044] There are multiple signs associated with pneumothorax including at least the lung point sign and the lung sliding sign.

[0045] The lung point sign is seen in the transition zone between normal lung tissue and pathological lung tissue such as pneumothorax, and is manifested by the alternation of the beach sign and the stratosphere sign with respiratory movement. The lung point is the main sign for diagnosing pneumothorax, with a specificity of 100% and a sensitivity of 79% for diagnosing occult pneumothorax. The location of the lung point sign can also be used to assess the size of the pneumothorax. If it is seen in front of the chest wall, it indicates that the lesion area is small; if it is seen on the side, back, or posterior axillary line of the chest, the area is larger. At the same time, it should be noted that the "false lung point sign" is also synchronized with respiratory movement, but there is no stratosphere sign, which can be seen in patients with pulmonary contusion.

[0046] Lung sliding is an important sign for identifying pneumothorax. It is manifested as a high-echo scintillation produced by the relative movement of the visceral pleura and the parietal pleura during respiratory movement. In patients with pneumothorax, air exists between the two layers of pleura, which affects the visualization of the pleura. Therefore, lung sliding disappears and a "barcode sign" appears, also known as the "stratosphere sign". The disappearance of lung sliding has a sensitivity of 95.3%, a specificity of 91.1%, and a negative predictive value of 99.2% to 100% (P<0.001) for diagnosing pneumothorax, indicating that the presence of lung sliding can effectively exclude pneumothorax. However, the disappearance of lung sliding is not completely equal to pneumothorax. Some critically ill patients may also have disappeared lung sliding if they have acute respiratory distress syndrome, pleural adhesions, large areas of atelectasis, lung contusions, etc. The positive predictive value is 87% in the general population, which drops to 56% in critically ill patients and only 27% in patients with respiratory failure. Therefore, it should be combined with other signs when making clinical diagnosis.

[0047] In some embodiments, the plurality of signs associated with pneumothorax further include a B-line sign, an A-line sign, and a pulmonary pulse sign.

[0048] The B-line sign is a discrete vertical reverberation artifact that extends from the pleural line to the bottom of the screen. It does not attenuate with distance and moves back and forth synchronously with lung sliding and breathing. Reverberation artifact refers to two or more reflections in the ultrasound propagation path, which often occurs on two smooth acoustic interfaces with large differences in acoustic impedance. The B-line is formed by the reflection of ultrasound on the alveolar air-fluid plane and is also called the "comet tail sign." The B-line is not a direct sign of pneumothorax. It is often used clinically to determine the severity of interstitial lung disease. However, if the B-line is combined with the pleural line, when these two signs appear, pneumothorax can be ruled out to a large extent.

[0049] The A-line sign is a series of bright lines parallel to the pleura formed by multiple reflections of ultrasound waves from the pleura. It is a reverberation artifact that assists in the diagnosis of pneumothorax. It is a high-echo horizontal line that is parallel to the pleura, equally spaced, and whose intensity decreases successively. The spacing between each A-line represents the distance between the parietal pleura and the skin surface. A-lines often appear when lung density is normal, and can also be seen in patients with pneumothorax. If lung sliding disappears in the presence of A-lines, the sensitivity and specificity of diagnosing occult pneumothorax are as high as 95% and 94%.

[0050] The pulmonary pulse sign, also known as lung pulsation, is another form of movement of the pleural line. In some cases, when lung sliding disappears, the pleural line can be seen to move vertically, consistent with the heart rhythm. This is due to insufficient inflation of the lungs, which causes the heart to pulsate and cause the pleural line to vibrate. A diagnostic test uses the presence of the pulmonary pulse as a criterion for excluding pneumothorax. Therefore, one of the characteristics of a pneumothorax patient is the disappearance of lung sliding and the pulmonary pulse, which is mainly because the air in the pleural cavity affects the horizontal and vertical movement of the parietal pleura.

[0051] The first prediction model is a multi-classification model for determining whether multiple target symbols exist. The input of the first prediction model is an image sequence of an object. As described above, the output of the first prediction model (first judgment result) includes whether multiple symbols related to pneumothorax exist, and the position of the existing symbols in at least one frame of the multiple frames. In some embodiments, the judgment module may input the image sequence into the trained first prediction model to obtain a first judgment result for the image sequence. For example, the image sequence of an object is input into the trained first prediction model, and the results show that the lung point sign and the A-line sign exist, but the lung sliding sign, the B-line sign, and the lung pulse sign do not exist. The lung point sign is located in the lower left corner of the 87th to 89th frames of the ultrasound image video, and the A-line sign is located in the middle of the 103th to 121th frames of the ultrasound image video.

[0052] In some embodiments, the first prediction model can be constructed based on the static detection model and the dynamic detection model. Figure 4 , the construction of dynamic detection model is detailed in Figure 5 For details on the sample construction and training methods of static detection models and dynamic detection models, see Figure 8 .

[0053] The first prediction model is initialized based on the static detection model, which is conducive to rapid model convergence and improves training efficiency. The first prediction model is based on symbolic real-time detection of the dynamic detection model, combining multi-frame ultrasound image information, avoiding the incompleteness of single image information and reducing false detection.

[0054] Step 230, based on a preset rule and the first judgment result, determine whether the second judgment result of the ultrasound image video needs to be re-detected, wherein the preset rule is related to pneumothorax judgment.

[0055] The preset rules are rules related to pneumothorax judgment.

[0056] In some embodiments, the preset rule may include that contradictory symbols do not exist simultaneously in the first judgment result. Contradictory symbols are multiple symbols indicating opposite diagnostic results. For example, the lung point sign is the main sign for diagnosing pneumothorax, and the presence of the lung sliding sign can effectively exclude pneumothorax. Therefore, the lung point sign and the lung sliding sign are contradictory symbols. If the lung point sign and the lung sliding sign exist simultaneously in the first judgment result of a segment of ultrasound image video, it is considered that the contradictory symbols coexist in the first judgment result, which does not meet the preset rule.

[0057] In some embodiments, conflicting symbols may include the following:

[0058]

[0059]

[0060] Table 1

[0061] As shown in Table 1, if the judgment result of whether multiple related symbols exist is consistent with any row in the table, it is considered that a contradictory symbol exists.

[0062] In some embodiments, the judgment module can determine the second judgment result of whether the ultrasound image video needs to be re-tested based on preset rules and the first judgment result. For example, the judgment module can determine the second judgment result of whether the ultrasound image video needs to be re-tested based on whether there are contradictory symbols in the ultrasound image video. Specifically, if there are contradictory symbols, the second judgment result is yes, otherwise, the second judgment result is no. It can be understood that since the specificity and sensitivity of multiple related symbols do not reach 100%, even if the test results are completely accurate, contradictory symbols may appear, but inaccurate detection is still the primary reason for the appearance of contradictory symbols. Therefore, in order to ensure the accuracy of the final test results, once a contradictory symbol appears, a more accurate model is adopted (the second prediction model, see Figure 3-Figure 5 )Retest.

[0063] Further judging the first judgment result output by the first prediction model to determine whether re-testing is needed is helpful to ensure the accuracy of the test.

[0064] If the second determination result is yes, step 240 is executed.

[0065] Step 240 , determining at least one target symbol that needs to be re-detected from the plurality of symbols based on the symbol gold standard levels of the plurality of symbols.

[0066] The gold standard refers to the most reliable method for diagnosing diseases currently recognized by the clinical medical community. The gold standard level refers to the reliability of the method. For example, the first-level gold standard has the highest reliability, the second-level is the second, and so on. The symbolic gold standard level of multiple symbols refers to the ranking of the reliability of multiple symbols for diagnosing pneumothorax. For example, the first-level gold standard is: lung point sign; the second-level gold standard is: lung sliding sign; the third-level gold standard is: B line sign; the fourth-level gold standard is: lung pulse sign; the fifth-level gold standard is: A line sign.

[0067] In some embodiments, the gold standard level between related symbols can be determined based on clustering analysis of ultrasound images of a large number of pneumothorax cases using a clustering model, as well as case information of the patient. For example, if the number of cases containing lung point signs is greater than the number of cases not containing lung sliding signs, the gold standard level of lung point signs is higher.

[0068] In some embodiments, the gold standard level between related symbols can be determined based on the specificity and / or sensitivity of multiple related symbols recognized by the medical community. For example, the first-level gold standard is the lung point sign, the second-level gold standard is the lung sliding sign, the third-level gold standard is the B-line sign, the fourth-level gold standard is the lung pulse sign, and the fifth-level gold standard is the A-line sign.

[0069] In some embodiments, the judgment module may re-test the symbol with the highest gold standard level among the contradictory symbols. In some embodiments, the judgment module may re-test the symbols with the highest gold standard level among all the adjacent symbols. For example, if the first-level gold standard lung point sign and the second-level gold standard lung sliding sign exist at the same time, and the fourth-level gold standard lung pulse sign and the fifth-level gold standard A-line sign exist at the same time, then the first-level gold standard lung point sign and the fourth-level gold standard lung pulse sign are re-tested. The re-testing method is described in detail in Figure 3 .

[0070] In the event of contradictory symbols, retesting the symbol with a higher gold standard level is beneficial to ensuring the detection accuracy of the higher-level symbol and the reliability of the test results.

[0071] It should be noted that the above description of the process 200 is only for example and illustration, and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to the process 200 under the guidance of this specification. However, these modifications and changes are still within the scope of this specification.

[0072] Figure 3 is an exemplary flow chart of a lung ultrasound image detection system according to some embodiments of the present specification.

[0073] Step 310: Obtain at least one second prediction model corresponding to the at least one target symbol. For any one of the at least one second prediction model, the second prediction model is used to predict whether the corresponding target symbol exists.

[0074] The second prediction model is a model for determining whether a specific target symbol exists, for example, a model for determining whether a lung point symbol exists. The second prediction model only determines the corresponding specific target symbol. For example, the second prediction model for the lung point only determines whether a lung point symbol exists, and for another example, the second prediction model for the A line only determines whether an A line symbol exists. It can be understood that, unlike the first prediction model, each second prediction model is a binary classification model. The structure of the second prediction model is detailed in Figure 6 , the sample construction and training method of the second prediction model are detailed in Figure 8 .

[0075] It is worth noting that the second prediction model introduces the attention mechanism after the LSTM layer, the construction of the training set and test set of the second prediction model is more targeted at the corresponding target symbol, and the parameters and network layers of the second prediction model are more complex than those of the first prediction model. Therefore, the second prediction model has a higher accuracy in judging specific symbols than the first prediction model.

[0076] In some embodiments, the update module may obtain at least one trained second prediction model from the storage device 140. In some embodiments, the update module may obtain at least one trained second prediction model through an interface, including but not limited to a program interface, a data interface, a transmission interface, and the like.

[0077] In some embodiments, the update module can determine the target image sequence based on the ultrasound image video and the first judgment result. As mentioned above, the first judgment result includes whether multiple symbols related to pneumothorax exist. In some embodiments, the update module can determine the target image sequence to be used when re-detecting the symbol based on whether the symbol in the first judgment result exists. For each symbol, if the symbol exists, step 330 is executed; if the symbol does not exist, step 320 is executed. Taking the first judgment result in which the lung point symbol, B-line symbol, and A-line symbol exist, but the lung sliding symbol and lung pulse sign do not exist as an example, when the lung point symbol, B-line symbol, and A-line symbol are re-detected, step 330 is executed to determine the target image sequence, and when the lung sliding symbol and lung pulse sign are re-detected, step 320 is executed to determine the target image sequence.

[0078] Step 320: Use an image sequence consisting of multiple frames of images in the ultrasound image video as the target image sequence.

[0079] In some embodiments, multiple frames of images in the ultrasound image video may be preprocessed, for example, to unify the resolution and image angle.

[0080] Step 330: Acquire at least one frame of image containing the symbol, and construct the target image sequence in chronological order based on the at least one frame of image containing the symbol.

[0081] As mentioned above, the first judgment result includes whether multiple symbols related to pneumothorax exist, and the position of the existing symbols in at least one frame of the multiple frames. In some embodiments, the update module can obtain at least one frame of the image in which a specific symbol exists based on the ultrasound image video and the first judgment result. For example, at least one frame of the image in which the symbol exists is determined by image retrieval or other methods.

[0082] In some embodiments, the update module can construct a target image sequence based on the at least one frame of image in chronological order. Specifically, if there are related symbols in multiple image frames in the ultrasound image video, the sequence formed by the multiple image frames in chronological order is used as the target image sequence. If there is only one image frame in the ultrasound image video with related symbols, the target image sequence can be determined according to a preset rule, for example, the at least one frame of image is used as the middle frame segment, and the 10 frames before and after the middle frame segment in the ultrasound image video are used as the target image sequence. For another example, the middle frame of the at least one frame of image is taken, and the middle frame and the 50 frames before and after the middle frame segment in the ultrasound image video are used as the target image sequence.

[0083] Again, taking the first judgment result that there are lung point symbols, B-line symbols, and A-line symbols, but no lung sliding symbols and lung pulse signs as an example, there are lung point symbols in frames 101-150 of the ultrasound image video, there are B-line symbols in frames 177-198 of the ultrasound image video, and there are A-line symbols in frames 205-210 and 225-240 of the ultrasound image video. When the lung point symbol is re-detected, frames 91-160 can be used as the target image sequence, or frames 76-176 can be used as the target image sequence. When the B-line symbol is re-detected, frames 167-208 can be used as the target image sequence, or frames 138-238 can be used as the target image sequence. When the A-line symbol is re-detected, frames 205-210 and 225-240 can be spliced ​​as the target image sequence.

[0084] Constructing the target image sequence in chronological order based on at least one frame of image containing the symbol can reduce unnecessary detection, reduce the amount of calculation and improve efficiency.

[0085] Step 340: Process the target image sequence based on the second prediction model to determine a third judgment result of whether the target symbol exists.

[0086] In some embodiments, the updating module may input the target image sequence of a symbol obtained in step 320 or step 330 into a second prediction model corresponding to the symbol to determine whether the symbol exists.

[0087] It can be understood that, since different models (a first prediction model and a second prediction model for a specific symbol) are used, the first judgment result and the third judgment result may be the same or different.

[0088] Step 350: Update the first judgment result based on the third judgment result.

[0089] In some embodiments, the update module may update the first judgment result by overwriting, rewriting, appending, etc.

[0090] It should be noted that the above description of process 300 is only for example and explanation, and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to process 300 under the guidance of this specification. However, these modifications and changes are still within the scope of this specification. For example, step 330 can be omitted, and it is not necessary to determine whether the first judgment result includes the existence of the symbol, and directly re-detect the image sequence composed of multiple frames of images in the ultrasound image video. For another example, step 310 can be performed after step 330.

[0091] Figure 4 It is a schematic diagram of a static detection model according to some embodiments of this specification.

[0092] In some embodiments, the input of the static detection model 400 is an ultrasound image, for example, an image frame of an ultrasound video captured by the ultrasound detection device 130, an ultrasound image stored in the storage device 140, etc. The output is the position of the relevant symbol in the image.

[0093] In some embodiments, Figure 4 The static detection model is constructed in the manner shown. In some embodiments, EfficientNet-B4 can be selected as the basic network. After the basic network, a 3×3 convolution layer is connected to adjust the number of output feature channels, and then the feature map is upsampled by 2 times through deconvolution, and the number of feature map channels is reduced to 1 / 4 of the original; the first output feature map of the basic network is a convolution layer with a size of 1 / 16 of the original input image, and a skip layer connection is made. After the connection layer, 3×3 convolution is used to adjust the number of output channels to 16, and finally deconvolution is used for 4 times upsampling, and the output probability map is 1 / 4 of the input image size. Finally, the probability map is upsampled by 4 times, and the feature map of the same size as the input image is output; the feature pyramid is used to merge the features obtained from each deconvolution layer, and finally the detection result is output.

[0094] For training of static detection models and construction of training samples, see Figure 8 and its description.

[0095] Convolution and deconvolution enable the static detection model to extract the features of ultrasound images while ensuring that the output feature map is consistent with the input image size. Skip-layer connections can solve the problem of gradient vanishing when the network is deep, and at the same time help the back propagation of gradients and speed up the training process.

[0096] Figure 5 It is a schematic diagram of a dynamic detection model shown in some embodiments of this specification.

[0097] In some embodiments, the input of the dynamic detection model 500 is an image sequence, for example, multiple frames of continuous images in the ultrasound image video collected by the ultrasound detection device 130, multiple frames of images in the ultrasound image video stored in the storage device 140 in the order of shooting time, etc. The output is the position of the relevant symbol in the image.

[0098] In some embodiments, Figure 5 In some embodiments, a plurality of LSTM units can be used to form an LSTM layer in the convolutional layer after the skip layer connection of the static detection model. The number of LSTM units is determined according to the size of the feature map output by the convolutional layer. The specific method is as follows: Assume that the input video frame I at time t t (Video frame t), image size W×H, the skip connection layer outputs m feature maps of size W′×H′, expressed as For each point (I, j), 1≤i≤W′, 1≤j≤H′, represents I t After the skip connection layer, W′×H′ LSTM units are used. is the input of LSTM. Since LSTM retains the information before time t, It combines time and space information, where S represents spatial features and ST represents space-time features. A 3×3 convolutional layer is added after the LSTM layer. The input of the convolutional layer is the output of each frame of the image in the LSTM layer. The number of channels of the convolutional layer output is 16, and the size of the feature map remains unchanged. The structure after the convolutional layer is consistent with the static detection model to build a video dynamic detection model; the dynamic detection model is set to input a 25-frame image sequence each time, save the output feature map of each frame of the image in the LSTM layer, connect the LSTM output feature maps of all frames as the input of the next convolutional layer, and keep the upsampling layer unchanged, and then output the prediction result of each frame.

[0099] For the training of dynamic detection models and the construction of training samples, see Figure 8 and its description.

[0100] Through the computer's continuous observation of the entire ultrasound examination process, there is a lot of information redundancy in time. From the perspective of data fusion, a large amount of data redundancy can offset the noise in the observation, so using ultrasound video for lesion detection and analysis can achieve better results than a single image. Symbolic real-time detection based on the dynamic detection model combines multi-frame ultrasound image information to avoid the incompleteness of a single image and reduce false detection.

[0101] Figure 6 is a schematic diagram of a second prediction model according to some embodiments of the present specification.

[0102] In some embodiments, the input of the second prediction model 600 is a target image sequence, and the method for determining the target image sequence is detailed in steps 310 to 330. The output is whether the target image sequence includes the relevant symbol.

[0103] In some embodiments, Figure 6 The second prediction model is constructed in the manner shown. In some embodiments, an attention mechanism can be introduced after the LSTM layer of the dynamic prediction model. The attention mechanism extracts features from the input target image sequence, finds the intrinsic relationship between the feature information, and calculates the attention probability distribution value assigned to each input. The calculation can give the recognition result by weighted average. The attention mechanism is followed by a fully connected layer, and the binary classification result is output.

[0104] In some embodiments, more network layers may be set for the second prediction model, for example, more convolutional layers.

[0105] The training of the second prediction model and the construction of training samples can be found in Figure 8 and its description.

[0106] By introducing the attention mechanism, setting more network layers and using specialized sample training, the second prediction model can have higher detection accuracy for specific symbols.

[0107] Figure 7 is a schematic diagram of a third model according to some embodiments of the present specification.

[0108] In some embodiments, the input of the third model 700 is an image sequence, for example, multiple frames of continuous images in the ultrasound image video collected by the ultrasound detection device 130, multiple frames of images in the ultrasound image video stored in the storage device 140 in the order of shooting time, etc. The output is the probability that the subject suffers from pneumothorax.

[0109] In some embodiments, Figure 7 The third model is constructed in the manner shown. In some embodiments, EfficientNet-B4 can be selected as the basic network. After the basic network, a fully connected layer is set, and finally the detection result is output.

[0110] For the training of the third model and the construction of training samples, see Figure 8 and its description.

[0111] The trained third model can automatically determine the detection result based on the learned law between the features of the image sequence and the probability of pneumothorax.

[0112] Figure 8 It is a schematic diagram of the model training method shown in some embodiments of this specification.

[0113] like Figure 8 As shown, in some embodiments, the initial model 810 can be trained based on a large number of training samples with labels to update the parameters of the initial model to obtain the trained model 820. The initial model 810 can include an initial static detection model, an initial dynamic detection model, an initial second prediction model, and an initial third model, and the trained model 820 can include a static detection model, a dynamic detection model, a second prediction model, and a third model.

[0114] In some embodiments, the initial static detection model can be trained based on a large number of training samples with labels to update the parameters of the initial static detection model to obtain a static detection model. In some embodiments, the training samples with labels can be input into the initial static detection model, and the parameters of the initial static detection model can be updated through training iterations.

[0115] In some embodiments, the training samples may be ultrasound images and / or ultrasound videos (image frames in the video), and the sample identifiers may be closed figures drawn including relevant symbols, such as rectangles, circles, ellipses, irregular figures, etc. In some embodiments, the training samples of the static detection model may be prepared according to the following steps.

[0116] Step A: Collect static ultrasound pneumothorax image data, based on cases. The data sources are mainly the images accumulated by the hospital over the years for filling out examination reports and newly collected images; the number of cases collected shall not be less than 10,000, and the total number of images shall not be less than 50,000.

[0117] Step B: Collect video data of ultrasonic chest pneumothorax scans, taking cases as units. The data sources are mainly the retained videos of the hospital reports and the newly collected videos as required. The video data cases and static image cases can be the same cases or different cases. The total number of video cases shall not be less than 3,000.

[0118] Step C: Calibrate the ultrasound image and video, and outline the corresponding lung point and lung sliding area. The outline process is performed manually according to conventional medical image recognition rules. For ultrasound video, it is necessary to outline each frame where the lung point and lung sliding appear, especially when the lung point and lung sliding begin to appear and disappear, and start or end the outline based on the visible lung point and lung sliding in the static image; on this basis, construct a data set for static detection, dynamic detection and pneumothorax recognition of lung point and lung sliding.

[0119] In some embodiments, 80% of the above data set can be randomly selected as a training set, and data augmentation can be performed on the training set; the training data can be iterated several times to obtain a static detection model.

[0120] In some embodiments, the initial dynamic detection model can be trained based on a large number of training samples with labels to update the parameters of the initial dynamic detection model to obtain the dynamic detection model. In some embodiments, the training samples with labels can be input into the initial dynamic detection model, and the parameters of the initial dynamic detection model can be updated through training iterations.

[0121] In some embodiments, the training sample may be an ultrasound video, and the sample identifier may be a closed figure including relevant symbols drawn in a video frame of the ultrasound video, such as a rectangle, a circle, an ellipse, an irregular figure, etc.

[0122] In some embodiments, since the amount of video sample data is relatively large and the training cycle is very long, the number of iterations can be reduced by adjusting the training frequency of each video sample according to the number of sample types and the size of the video. Among all pneumothorax cases, common diseases account for a large proportion, and there are many types of difficult and rare diseases but they account for a small proportion. If the same weight is assigned to each sample during training, that is, the training time is the same, the algorithm will tend to ignore difficult and rare diseases. A clustering method based on image features is used to cluster samples with similar pathologies and small differences in image information into one category. Combined with the number of image frames of each video sample, the training frequency of each sample is adjusted according to the following formula:

[0123] FEQ(A)=Z / sqrt(N(A)×L(A))

[0124] Where FEQ(A) is the training frequency of A, N(A) is the number of samples in the category of A, L(A) is the number of frames of video A, and Z is the normalization coefficient. Since the number of frames of each dynamic video data varies from dozens to hundreds, the number of video frames also needs to be considered during training. All labeled ultrasound video data are randomly divided into training and test sets at a ratio of 8 to 2.

[0125] In some embodiments, the static detection model parameters can be used to initialize the dynamic detection model, the newly added LSTM layer is randomly initialized, the learning rate of all layers before the LSTM layer is set to 0, and only the parameters of the LSTM layer and its subsequent convolutional layer are learned; since the entire network uses a static detection model in the feature extraction part, the information input to the LSTM layer is already a highly abstract feature; the LSTM layer parameters only need to be reasonably configured, and the network converges easily. Each time training, a video is selected according to the sample training frequency, and a continuous 25-frame image sequence in the video is randomly selected. The data enhancement method is used to perform random contrast stretching, brightness adjustment, and pixel value noise perturbation on each channel of the image. The enhanced image sequence is sequentially input into the dynamic detection model, and the random gradient descent method is used to train the network; on the test video, each frame is used as the predicted intermediate frame, and the detection result of the entire video is output. The detection result is evaluated using the average intersection-over-union ratio, and different learning rates and learning rate decay strategies are modified. The result with the highest regional similarity on the test video is selected as the dynamic detection model.

[0126] In some embodiments, the dynamic detection model can also be obtained according to other training methods, for example, a corresponding initial learning rate (for example, 0.1) and a learning rate decay strategy are set for the layer before the LSTM layer, and a dynamic detection model is obtained through joint training based on the ultrasound video with the mark. This application is not limited here.

[0127] In some embodiments, the second prediction model can be trained based on a large number of training samples with labels to update the parameters of the initial second prediction model to obtain the second prediction model. In some embodiments, the training samples with labels can be input into the initial second prediction model, and the parameters of the initial second prediction model can be updated through training iterations.

[0128] In some embodiments, the training sample may be an ultrasound image and / or ultrasound video (image frame in a video). The sample identifier may be a closed figure and number including relevant symbols outlined in the ultrasound video. Specifically, the ultrasound video with relevant symbols is labeled as a closed figure including relevant symbols and number 1, and the ultrasound video without relevant symbols is labeled as only number 0. For example, when constructing a second prediction model sample for training and identifying lung point symbols, an ultrasound video with lung point symbols is labeled as a closed figure and number 1 that circle the lung point symbol area in the image frame, and an ultrasound video without lung point symbols is labeled as number 0. For another example, when constructing a second prediction model sample for training and identifying lung sliding symbols, an ultrasound video with lung sliding symbols is labeled as a closed figure and number 1 that circle the lung sliding symbol area in the image frame, and an ultrasound video without lung sliding symbols is labeled as number 0.

[0129] In some embodiments, all ultrasound video data with labels may be randomly divided into a training set and a test set in a ratio of 8 to 2.

[0130] In some embodiments, the iterative update method uses a gradient descent method using a cross entropy loss function, by calculating the gradient of the loss function and updating the parameters of the second prediction model.

[0131] In some embodiments, the initial third model can be trained based on a large number of training samples with labels to update the parameters of the initial third model to obtain the third model. In some embodiments, the training samples with labels can be input into the initial third model to update the parameters of the initial dynamic detection model through training iterations.

[0132] In some embodiments, the training samples may be ultrasound images and / or ultrasound videos (image frames in a video). The sample identifier may be a number 0-3, including a class label of 0 for the presence of pneumothorax, a class label of 1 for the presence of a high probability of pneumothorax, a class label of 2 for the absence of pneumothorax, and a class label of 3 for the uncertainty of whether there is a pneumothorax. For example, if the lung point and A-line sign are detected in the ultrasound image or video but the lung pulse and lung sliding are not detected, it is determined that there is a pneumothorax; if the lung point and lung sliding are detected, there is a high probability of a pneumothorax; if the lung point is not detected but lung sliding and B-line sign are detected, it is determined that there is no pneumothorax; if the lung point and lung sliding are not detected, it is determined that there is no pneumothorax; if the lung point and lung sliding are not detected, it is uncertain whether there is a pneumothorax and further diagnosis is required.

[0133] In some embodiments, all the pneumothorax data to be identified may be randomly divided into a training set and a test set in a ratio of 8 to 2.

[0134] As described above, the third model may include a base network and a fully connected layer. In some embodiments, the base network of the third model may be initialized using the parameters of the static detection model, and the fully connected layer may be randomly initialized to construct an initial third model.

[0135] In some embodiments, the output value of the last fully connected layer of the base network can be modified to 2, and the Sigmoid cross entropy loss function is used. The third model is trained on the training set using the stochastic gradient descent method, and different learning rates and learning rate decay strategies are modified. The cross-experiment is repeated, and the ROC curve is predicted and made on the test set. The parameters and model with the highest AUC value are selected as the third model.

[0136] It should be noted that the above description of process 800 is only for example and illustration, and does not limit the scope of application of this specification. For those skilled in the art, various modifications and changes can be made to process 800 under the guidance of this specification. However, these modifications and changes are still within the scope of this specification.

[0137] The basic concepts have been described above. Obviously, for those skilled in the art, the above detailed disclosure is only for example and does not constitute a limitation of this specification. Although not explicitly stated here, those skilled in the art may make various modifications, improvements and corrections to this specification. Such modifications, improvements and corrections are suggested in this specification, so such modifications, improvements and corrections still belong to the spirit and scope of the exemplary embodiments of this specification.

[0138] At the same time, this specification uses specific words to describe the embodiments of this specification. For example, "one embodiment", "an embodiment", and / or "some embodiments" refer to a certain feature, structure or characteristic related to at least one embodiment of this specification. Therefore, it should be emphasized and noted that "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more in different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures or characteristics in one or more embodiments of this specification can be appropriately combined.

[0139] In addition, unless explicitly stated in the claims, the order of the processing elements and sequences described in this specification, the use of alphanumeric characters, or the use of other names are not intended to limit the order of the processes and methods of this specification. Although the above disclosure discusses some invention embodiments that are currently considered useful through various examples, it should be understood that such details are only for illustrative purposes, and the attached claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the essence and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing server or mobile device.

[0140] Similarly, it should be noted that in order to simplify the description disclosed in this specification and thus help understand one or more embodiments of the invention, in the above description of the embodiments of this specification, multiple features are sometimes combined into one embodiment, figure or description thereof. However, this disclosure method does not mean that the features required by the subject matter of this specification are more than the features mentioned in the claims. In fact, the features of the embodiments are less than all the features of the single embodiment disclosed above.

[0141] In some embodiments, numbers describing the number of components and attributes are used. It should be understood that such numbers used in the description of the embodiments are modified by the modifiers "about", "approximately" or "substantially" in some examples. Unless otherwise specified, "about", "approximately" or "substantially" indicate that the numbers are allowed to vary by ±20%. Accordingly, in some embodiments, the numerical parameters used in the specification and claims are approximate values, which may change according to the required features of individual embodiments. In some embodiments, the numerical parameters should take into account the specified significant digits and adopt the general method of retaining digits. Although the numerical domains and parameters used to confirm the breadth of their range in some embodiments of this specification are approximate values, in specific embodiments, the setting of such numerical values ​​is as accurate as possible within the feasible range.

[0142] Each patent, patent application, patent application publication, and other materials, such as articles, books, specifications, publications, documents, etc., cited in this specification are hereby incorporated by reference in their entirety. Except for application history documents that are inconsistent with or conflicting with the contents of this specification, documents that limit the broadest scope of the claims of this specification (currently or later attached to this specification) are also excluded. It should be noted that if the descriptions, definitions, and / or use of terms in the materials attached to this specification are inconsistent or conflicting with the contents described in this specification, the descriptions, definitions, and / or use of terms in this specification shall prevail.

[0143] Finally, it should be understood that the embodiments described in this specification are only used to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, as an example and not a limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly introduced and described in this specification.

Claims

1. A lung ultrasound image detection system, comprising: An acquisition module, used for acquiring an ultrasonic image video of an object; A judgment module is used to process the image sequence based on the trained first prediction model to determine a first judgment result; wherein, The image sequence is composed of multiple frames of images in the ultrasound image video in the order of shooting time; The first judgment result includes whether a plurality of symbols related to pneumothorax exist, and the position of the existing symbols in at least one frame of the plurality of frames of images; The plurality of symbols include at least a lung point sign and a lung sliding sign; A second judgment result of determining whether the ultrasound image video needs to be re-detected based on a preset rule and the first judgment result, wherein the preset rule includes that the contradictory symbols in the first judgment result do not exist at the same time, and the contradictory symbols are multiple symbols indicating opposite diagnostic results; In response to the second determination result being yes, determining at least one target symbol that needs to be re-detected from the plurality of symbols based on the symbol gold standard levels of the plurality of symbols; Update modules for: Acquire at least one second prediction model corresponding to the at least one target symbol, wherein for any one of the at least one second prediction model, the second prediction model is used to predict whether the target symbol corresponding to the second prediction model exists, and the accuracy of the second prediction model is higher than that of the first prediction model; Processing the target image sequence based on the second prediction model to determine a third judgment result of whether the target symbol exists; The first judgment result is updated based on the third judgment result.

2. The system according to claim 1, in order to determine the target image sequence, the updating module is further used for: For each of the plurality of tokens, If the first judgment result includes the existence of the symbol, obtaining at least one frame of image in which the symbol exists, and constructing the target image sequence in chronological order based on the at least one frame of image in which the symbol exists; If the first judgment result includes that the symbol does not exist, an image sequence consisting of multiple frames of images in the ultrasound image video is used as the target image sequence.

3. The system according to claim 1, wherein the first prediction model comprises a sequence model, a first convolutional neural network, and a fully connected layer; The first convolutional neural network processes each image in the image sequence to extract image features of each image in the image sequence; The sequence model processes an image feature sequence formed by image features of each image in the image sequence to extract image sequence features; The fully connected layer processes the image sequence features to determine the first judgment result.

4. According to the system of claim 3, the fully connected layer processes the image sequence features and the image features of each image in the image sequence to determine the first judgment result.

5. The system of claim 3, wherein the first prediction model comprises a second convolutional neural network; The second convolutional neural network processes the image sequence features, extracts features of the image sequence features, and inputs the features into the fully connected layer to determine the first judgment result.

6. The system according to claim 1, further comprising a determination module, wherein the determination module is configured to: The first judgment result, as well as the characteristic values ​​and sensitivity values ​​of the plurality of symbols are processed based on a third model to determine the probability that the subject suffers from pneumothorax.

7. A pneumothorax detection device, comprising a processor and a memory; the memory is used to store instructions, characterized in that: When the instructions are executed by the processor, the device implements the functions corresponding to the lung ultrasound image detection system according to any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that: The storage medium stores computer instructions. When the computer reads the computer instructions in the storage medium, the computer runs the functions corresponding to the lung ultrasound image detection system as described in any one of claims 1 to 6.

Images