A method, system and electronic device for capturing thyroid ultrasound abnormal regions

Abstract

The present application relates to a kind of thyroid ultrasound abnormal region capture method, system and electronic equipment, the method includes: obtaining the detection frame of the same abnormal region in the thyroid ultrasound image of current frame and the previous N frame of current frame, constitute N+1 detection frame;N+1 detection frame is subjected to spatial coordinate attention mechanism processing, obtain the detection frame to be measured;The detection frame to be measured is input into long short-term memory network model, obtains the parameter information of the detection frame to be measured;According to the parameter information of the detection frame to be measured, the parameter information of the detection frame in the thyroid ultrasound image of current frame and the previous 1 frame of current frame, obtain the optimized detection frame of abnormal region in the thyroid ultrasound image of current frame, as abnormal region detection result.The present application is combined with the mobility of thyroid nodule, based on spatial coordinate attention mechanism and long short-term memory network model, the detection frame of abnormal region of continuous multiple frames is remembered and analyzed, realizes the capture of thyroid ultrasound abnormal region.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of image processing and recognition, in particular to a thyroid ultrasound abnormal region capturing method and system and electronic equipment. BACKGROUND

[0002] Thyroid nodule refers to a lump in the thyroid gland, which is a common clinical condition and can be caused by various factors. There are many thyroid diseases in clinical practice, such as thyroid degeneration, inflammation, autoimmune, and neoplasm, which can all manifest as nodules. Thyroid nodules can occur singly or in multiple, and the incidence of multiple nodules is higher than that of single nodules, but the incidence of thyroid cancer is higher in single nodules.

[0003] The existing ultrasonic diagnosis method is to use ultrasonic waves to scan the thyroid of the patient, and can obtain multiple frames of thyroid ultrasound images. Then the doctor determines the abnormal area by manually judging the thyroid ultrasound images with the naked eye. The existing diagnosis method relying on the naked eye requires high experience of the doctor, and the misdiagnosis rate is also high.

[0004] How to use image recognition technology to automatically capture the abnormal area in the thyroid ultrasound image and obtain the detection box information of the abnormal area to feed back to the doctor for diagnosis to reduce the misdiagnosis rate has become a technical problem to be solved. SUMMARY

[0005] Therefore, the present application provides a thyroid ultrasound abnormal region capturing method, system and electronic equipment to automatically capture the abnormal area in the thyroid ultrasound image using image recognition technology.

[0006] To achieve the above-mentioned purpose, the present application provides the following solutions:

[0007] A thyroid ultrasound abnormal region capturing method, the method comprising the following steps:

[0008] Obtain the detection box of the same abnormal area in the thyroid ultrasound image of the current frame and the previous N frames of the current frame, and form N+1 detection boxes;

[0009] Perform spatial coordinate attention mechanism processing on the N+1 detection boxes to obtain a to-be-tested detection box;

[0010] Input the to-be-tested detection box into a long short-term memory network model to obtain parameter information of the to-be-tested detection box; the parameter information includes position and area;

[0011] According to the parameter information of the to-be-tested detection box and the parameter information of the detection box in the thyroid ultrasound image of the current frame and the previous 1 frame of the current frame, obtain an optimized detection box of the abnormal area in the thyroid ultrasound image of the current frame as an abnormal area detection result.

[0012] Optionally, the detection boxes of the same abnormal region in the thyroid ultrasound images of the current frame and the previous N frames of the current frame are obtained, and N+1 detection boxes are formed, specifically including:

[0013] An initial detection box of the abnormal region in the thyroid ultrasound image of the current frame is obtained by using an ultrasound capture algorithm.

[0014] The initial detection box is modified to obtain a detection box of the abnormal region in the thyroid ultrasound image of the current frame as the N+1th detection box.

[0015] The detection box with the maximum intersection over union with the N+1th detection box is selected from all the detection boxes in the Nth frame of the thyroid ultrasound image before the current frame as the N+1-nth detection box; n = 1, 2,..., N.

[0016] Optionally, the initial detection box is modified to obtain a detection box of the abnormal region in the thyroid ultrasound image of the current frame as the N+1th detection box, specifically including:

[0017] It is judged whether there is a state of a large box containing a small box in all the initial detection boxes, and a judgment result is obtained.

[0018] When the judgment result indicates yes, the areas of the two initial detection boxes in the state of a large box containing a small box are weighted and averaged to obtain the N+1th detection box.

[0019] Optionally, the N+1 detection boxes are subjected to spatial coordinate attention mechanism processing to obtain a to-be-detected detection box, specifically including:

[0020] The N+1 detection boxes are input into a first one-dimensional convolution layer for one-dimensional convolution processing to obtain a first convolution result.

[0021] The N+1 detection boxes are input into a second one-dimensional convolution layer for one-dimensional convolution processing to obtain a second convolution result.

[0022] The N+1 detection boxes are input into a relative position encoder for relative position encoding to obtain an encoding result.

[0023] The second convolution result and the encoding result are point multiplied to obtain a point multiplication result.

[0024] The first convolution result and the point multiplication result are added to obtain an addition result.

[0025] The addition result is sequentially subjected to softmax function calculation and reshape function calculation to obtain a calculation result as the to-be-detected detection box.

[0026] Optionally, the parameter information of the optimized detection frame of the abnormal region in the thyroid ultrasound image of the current frame is obtained according to the parameter information of the to-be-tested detection frame, the parameter information of the detection frame in the thyroid ultrasound image of the current frame and the thyroid ultrasound image of the previous 1 frame of the current frame, as an abnormal region detection result, and specifically includes:

[0027] The parameter information of the optimized detection frame of the abnormal region in the thyroid ultrasound image of the current frame is obtained by weighted average processing of the parameter information of the to-be-tested detection frame, the parameter information of the detection frame in the thyroid ultrasound image of the current frame and the thyroid ultrasound image of the previous 1 frame of the current frame, as an abnormal region detection result.

[0028] A thyroid ultrasound abnormal region capturing system, the system is applied to the method, and the system comprises:

[0029] A detection frame acquisition module is configured to acquire detection frames of the same abnormal region in the thyroid ultrasound image of the current frame and the previous N frames of the current frame, and to form N+1 detection frames.

[0030] A to-be-tested detection frame acquisition module is configured to perform spatial coordinate attention mechanism processing on the N+1 detection frames to obtain a to-be-tested detection frame.

[0031] A parameter information acquisition module is configured to input the to-be-tested detection frame into a long short-term memory network model to obtain parameter information of the to-be-tested detection frame, wherein the parameter information comprises a position and an area.

[0032] A detection frame optimization module is configured to obtain the parameter information of the optimized detection frame of the abnormal region in the thyroid ultrasound image of the current frame according to the parameter information of the to-be-tested detection frame, the parameter information of the detection frame in the thyroid ultrasound image of the current frame and the thyroid ultrasound image of the previous 1 frame of the current frame, as an abnormal region detection result.

[0033] An electronic device comprises a memory, a processor and a computer program stored in the memory and executable on the processor, and the processor implements the method when executing the computer program.

[0034] A computer readable storage medium, the storage medium stores a computer program, and the computer program is executed to implement the method.

[0035] According to the embodiments of the present application, the following technical effects are provided:

[0036] This invention discloses a method, system, and electronic device for capturing abnormal regions in thyroid ultrasound. The method includes: acquiring detection boxes of the same abnormal region in the current frame and N frames preceding the current frame of thyroid ultrasound images, forming N+1 detection boxes; processing the N+1 detection boxes using a spatial coordinate attention mechanism to obtain a detection box to be tested; inputting the detection box to be tested into a long short-term memory network model to obtain parameter information of the detection box to be tested; the parameter information includes position and area; and obtaining an optimized detection box of the abnormal region in the current frame of thyroid ultrasound image based on the parameter information of the detection box to be tested and the parameter information of the detection boxes in the current frame and the frame preceding the current frame, as the abnormal region detection result. This invention combines the mobility of thyroid nodules and uses a spatial coordinate attention mechanism and a long short-term memory network model to memorize and analyze the detection boxes of abnormal regions in multiple consecutive frames, thus realizing the capture of abnormal regions in thyroid ultrasound. Attached Figure Description

[0037] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0038] Figure 1 This is a flowchart of a method for capturing abnormal areas of thyroid ultrasound provided in an embodiment of the present invention;

[0039] Figure 2 A schematic diagram illustrating the principle of the thyroid ultrasound abnormality region capture method provided in this embodiment of the invention;

[0040] Figure 3 Example images of thyroid ultrasound before and after optimization provided in embodiments of the present invention;

[0041] Figure 4 A schematic diagram of the spatial coordinate attention mechanism provided in an embodiment of the present invention;

[0042] Figure 5 This is a schematic diagram of the structure of a long short-term memory network model provided in an embodiment of the present invention. Detailed Implementation

[0043] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0044] The application aims to provide a thyroid ultrasound abnormal region capturing method, system and electronic device, which realizes automatic capture of abnormal regions in thyroid ultrasound images by using image recognition technology.

[0045] In order to make the above-mentioned purposes, features and advantages of the application more obvious and easy to understand, the application will be further described in detail below in combination with the drawings and specific embodiments.

[0046] Embodiment 1

[0047] As shown in Figure 1 and 2 , the application embodiment 1 provides a thyroid ultrasound abnormal region capturing method, which comprises the following steps:

[0048] Step 101, obtaining the detection box of the same abnormal region in the thyroid ultrasound images of the current frame and the previous N frames of the current frame, to form N+1 detection boxes.

[0049] Whenever the user uses the thyroid nodule ultrasound positioning software to newly create a scan and start running, the application starts to receive the thyroid ultrasound image and detection box information of each frame output by the software backend target detection thyroid nodule algorithm as the input of the application. First, the application will correct the detection box of the current frame. The correction basis is that the thyroid ultrasound image shows the anatomical information of a cross section of human organs or tissues, and the detection box of the same category will not exist in the case of large box containing small box. For example Figure 3 As shown in the figure, when the case of large box containing small box under the same category occurs, the weighted average of the areas of the two detection boxes is calculated to obtain the corrected detection box of the current frame. The formula is as follows:

[0050]

[0051] Subsequently, the corrected detection box of the current frame is compared with the cached memory. By calculating the intersection over union (IoU, calculation method as follows, that is, the area ratio of intersection and union), the cached record with the largest intersection over union is selected as the memory to be inherited and updated by the detection box of the current frame, because there are multiple nodule cases, it is necessary to determine which nodule the detection box of the current frame identifies.

[0052] Based on the above principle, step 101 of the embodiment 1 of the present application comprises: acquiring an initial detection box of an abnormal area in the thyroid ultrasound image of the current frame by using an ultrasound capture algorithm; correcting the initial detection box to obtain a detection box of the abnormal area in the thyroid ultrasound image of the current frame as the N+1th detection box; selecting a detection box with the maximum intersection-over-union with the N+1th detection box from all detection boxes in the Nth frame before the current frame as the N+1-Nth detection box; n=1, 2,..., N.

[0053] In the method, the initial detection boxes are corrected to obtain the detection box of the abnormal area in the thyroid ultrasound image of the current frame as the N+1th detection box, and the method specifically comprises: judging whether there is a state of a large box containing a small box in all the initial detection boxes to obtain a judgment result; when the judgment result indicates yes, performing weighted average processing on the areas of the two initial detection boxes in the state of the large box containing the small box to obtain the N+1th detection box.

[0054] In step 102, the N+1 detection boxes are subjected to spatial coordinate attention mechanism processing to obtain a to-be-detected detection box.

[0055] Since the LSTM (Long Short-Term Memory, long short-term memory network model) here analyzes spatial coordinate information instead of time sequence information, the detection box coordinates are subjected to one-dimensional convolution (Conv1D for short) and relative position encoding (RPE for short) when input, so as to realize spatial coordinate attention mechanism and better analyze the relative spatial relationship of the coordinates.

[0056] Based on the above principle, as shown in Figure 4 The step 102 of the present application specifically comprises: inputting the N+1 detection boxes into a first one-dimensional convolution layer Conv1D_1 to perform one-dimensional convolution processing and obtain a first convolution result; inputting the N+1 detection boxes into a second one-dimensional convolution layer Conv1D_2 to perform one-dimensional convolution processing and obtain a second convolution result; inputting the N+1 detection boxes into a relative position encoder RPE to perform relative position encoding and obtain an encoding result; performing point multiplication on the second convolution result and the encoding result to obtain a point multiplication result; adding the first convolution result and the point multiplication result to obtain an addition result; sequentially performing softmax function calculation and reshape function calculation on the addition result to obtain a calculation result as the to-be-detected detection box x t .

[0057] Step 103: Input the detection box to be tested into the long short-term memory network model to obtain the parameter information of the detection box to be tested; the parameter information includes position and area.

[0058] The structure of the long short-term memory network model is as follows: Figure 5 As shown, a single LSTM layer consists of a single loop structure. The number of times a single loop structure needs to be updated depends on the dimension of the input data and the number of loops, rather than being composed of multiple single loop structures connected together. In other words, the total number of parameters of the current LSTM layer only needs to be calculated for one loop unit, rather than the total number of multiple consecutive units.

[0059] A single cyclic structure, or LSTM cell, includes an input gate, a forget gate, an output gate, and a cell state.

[0060] Input gate: determines how much of the network's input data at the current moment needs to be saved to the cell state.

[0061] Forget gate: determines how much of the cell state from the previous time step needs to be retained in the current time step.

[0062] Output gate: Controls how much of the current cell state needs to be output to the current output value.

[0063] The output h of the previous time step t-1 And the current data input (the detection box to be tested x) t f is obtained through the forget gate. t (f t The process of outputting the forget gate is as follows:

[0064] f t =σ(W f ·[h t-1 x t ]+b f );

[0065] Among them, W f and b f These are the two parameters of the forget gate, and σ(·) is the sigmoid function.

[0066] The output h of the previous time step t-1 And the current data input (the detection box to be tested x) t ), i is obtained through the input gate t And obtain the current temporary state through the cell state. The process is as follows:

[0067]

[0068] Among them, W i and b i The two parameters W are the input gate.C and b C These are two parameters representing the cell state.

[0069] The previous single-loop structure's cell state C t-1 The output f of the forget gate t The output i of the input gate t and the current temporary state Get the current state C of a single loop structure. t The process is as follows:

[0070]

[0071] Applying the output h from the previous time step t-1 And the current data input (the detection box to be tested x) t ), and obtain o through the output gate. t (o t The process of outputting the output gate, and the cell state C combined with the current single loop structure. t and ot The final output h is obtained. t The process is as follows:

[0072]

[0073] Among them, W o and b o These are the two parameters of the output gate.

[0074] Step 104: Based on the parameter information of the detection box to be tested, the parameter information of the detection boxes in the current frame and the previous frame of the thyroid ultrasound image, obtain the optimized detection box of the abnormal region in the current frame of the thyroid ultrasound image, and use it as the abnormal region detection result.

[0075] Optionally, the step of obtaining optimized detection box parameter information for the abnormal region in the current frame of the thyroid ultrasound image based on the parameter information of the detection box to be tested, the parameter information of the detection boxes in the current frame and the previous frame of the thyroid ultrasound image, and using this as the abnormal region detection result, specifically includes:

[0076] The parameter information of the detection box to be tested, the parameter information of the detection boxes in the current frame and the previous frame of the thyroid ultrasound image are subjected to weighted average processing to obtain the optimized parameter information of the detection box of the abnormal region in the current frame of the thyroid ultrasound image, which is used as the abnormal region detection result.

[0077] The application obtains an optimized detection frame by performing a weighted average on the output of the LSTM, the detection frame of the current frame and the detection frame of the previous frame. The cache memory of the nodules is updated according to the center point, and the displacement during the memory cache is calculated to track the target while suppressing the memory whose displacement amplitude exceeds the limit. According to the research results and the guidance of clinical medical experts, the thyroid nodule is an isolated individual, and other tissue structures are continuous, which will deform and move in a dynamic picture. Therefore, false positive suppression can be achieved according to the displacement amplitude.

[0078] The detection frame monitored by the displacement is first compared with the automatic record so far and the record is updated. According to the guidance of the clinician and the actual treatment habit, the algorithm retains the detection frame with the largest cross-sectional area and the corresponding image.

[0079] Then, whether to output the current result is selected according to the number of frames in which the detection frame of the nodule continuously appears. Since the application corrects the detection frame of the current frame according to the LSTM and the detection frame coordinates of the previous frame, the application sets a stable period of 3 frames, that is, the detection frame of the current frame is the 3rd frame and above that continuously appears, so as to start outputting the result to ensure the best detection effect.

[0080] According to whether the current scan is finished, if not, the next frame of image and detection frame is continuously received; if yes, the automatic record image and detection frame information are saved, and the next scan is started.

[0081] Embodiment 2

[0082] Embodiment 2 of the application provides a thyroid ultrasound abnormal region capturing system, which is applied to the method of embodiment 1. The system comprises a detection frame acquisition module, which is used to acquire detection frames of the same abnormal region in thyroid ultrasound images of a current frame and N previous frames of the current frame, to form N+1 detection frames; a to-be-detected detection frame acquisition module, which is used to perform spatial coordinate attention mechanism processing on the N+1 detection frames to obtain a to-be-detected detection frame; a parameter information acquisition module, which is used to input the to-be-detected detection frame into a long short-term memory network model to obtain parameter information of the to-be-detected detection frame; the parameter information comprises a position and an area; and a detection frame optimization module, which is used to obtain parameter information of an optimized detection frame of the abnormal region in the thyroid ultrasound image of the current frame according to the parameter information of the to-be-detected detection frame and parameter information of a detection frame in the thyroid ultrasound image of the current frame and the previous frame, as an abnormal region detection result.

[0083] The specific way in which the functions of each module in the system provided in the embodiment are implemented is the same as the specific steps of the method of embodiment 1, and will not be repeated here.

[0084] Embodiment 3

[0085] Embodiment 3 of the present application provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method of embodiment 1 when executing the computer program. The exemplary electronic device can be integrated into an existing thyroid ultrasound image acquisition system, and can identify and record abnormal areas of real-time acquired thyroid ultrasound images, thereby realizing automatic tracking and automatic picture leaving of the target (abnormal area).

[0086] In addition, the computer program in the memory described above is implemented in the form of a software functional unit and sold or used as an independent product, and can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the part that contributes to the prior art or part of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium, and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory, a random access memory, a magnetic disk or an optical disk, and various media that can store program codes.

[0087] Embodiment 4

[0088] Embodiment 4 of the present application provides a computer readable storage medium, wherein the storage medium stores a computer program, and the computer program is executed to implement the method described above.

[0089] Advantages of the present application

[0090] 1) The present application can integrate clinical medical knowledge into the target detection algorithm of the thyroid ultrasound abnormal area capture software, optimize the output detection frame, and realize false positive suppression;

[0091] 2) Since the reading of dynamic ultrasound images is based on the continuity of the picture, and the thyroid nodule ultrasound capture algorithm can only analyze each frame of image independently, and does not have the ability to remember and analyze continuous data, the present application optimizes the current frame result through the LSTM network and the previous frame picture, and the thyroid ultrasound abnormal area capture software has the ability to remember and analyze continuous data;

[0092] 3) The LSTM network is originally used to analyze time series data rather than spatial coordinate information. The present application innovatively uses one-dimensional convolution and relative position coding to modify the LSTM network, so that it has the ability to process continuous spatial coordinates;

[0093] 4) The function of automatic image keeping can greatly reduce the difficulty and workload of the user in simultaneously operating the ultrasonic machine and the thyroid ultrasonic abnormal region capturing software;

[0094] 5) The program of the present application runs very fast, and the running time is ≤0.001 second, which does not affect the software running or user experience;

[0095] The various embodiments are described in a progressive manner in the specification, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between various embodiments can be mutually referred to. For the system disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0096] The principles and implementation manners of the present application are described by applying specific examples herein, and the above embodiment description is only used to help understand the control method and the core idea of the present application; meanwhile, for the general technical personnel in the field, the specific implementation manner and application range can be changed according to the idea of the present application. In conclusion, the content of the specification should not be understood as the limitation of the present application.

Claims

1. A method for capturing abnormal areas of the thyroid gland using ultrasound, characterized in that, The method comprises the following steps: Obtaining the detection boxes of the same abnormal area in the thyroid ultrasound images of the current frame and the previous N frames of the current frame, and composing N+1 detection boxes; Performing spatial coordinate attention mechanism processing on the N+1 detection boxes to obtain a to-be-detected detection box; Inputting the to-be-detected detection box into a long short-term memory network model to obtain parameter information of the to-be-detected detection box; the parameter information comprises a position and an area; According to the parameter information of the to-be-detected detection box, the parameter information of the detection box in the thyroid ultrasound images of the current frame and the previous 1 frame of the current frame, obtaining an optimized detection box of the abnormal area in the thyroid ultrasound images of the current frame as an abnormal area detection result, specifically comprising: Performing weighted average processing on the parameter information of the to-be-detected detection box and the parameter information of the detection box in the thyroid ultrasound images of the current frame and the previous 1 frame of the current frame to obtain the parameter information of the optimized detection box of the abnormal area in the thyroid ultrasound images of the current frame as the abnormal area detection result.

2. The thyroid ultrasound abnormal region capturing method of claim 1, wherein, The obtaining of the detection boxes of the same abnormal area in the thyroid ultrasound images of the current frame and the previous N frames of the current frame, and the composition of N+1 detection boxes specifically comprises: Obtaining an initial detection box of the abnormal area in the thyroid ultrasound images of the current frame by using an ultrasound capture algorithm; Performing correction on the initial detection box to obtain a detection box of the abnormal area in the thyroid ultrasound images of the current frame as the N+1th detection box; Selecting a detection box with the maximum intersection over union with the N+1th detection box from all the detection boxes in the Nth frame of the thyroid ultrasound images before the current frame as the N+1-nth detection box; n=1, 2, …, N.

3. The thyroid ultrasound abnormal region capturing method according to claim 2, characterized by, The correction of the initial detection box to obtain a detection box of the abnormal area in the thyroid ultrasound images of the current frame as the N+1th detection box specifically comprises: Judging whether there is a state of a large box containing a small box in all the initial detection boxes to obtain a judgment result; When the judgment result indicates yes, performing weighted average processing on the areas of the two initial detection boxes in the state of a large box containing a small box to obtain the N+1th detection box.

4. The thyroid ultrasound abnormal region capturing method of claim 1, wherein, The spatial coordinate attention mechanism processing on the N+1 detection boxes to obtain a to-be-detected detection box specifically comprises: Inputting the N+1 detection boxes into a first one-dimensional convolution layer to perform one-dimensional convolution processing to obtain a first convolution result; Inputting the N+1 detection boxes into a second one-dimensional convolution layer to perform one-dimensional convolution processing to obtain a second convolution result; Inputting the N+1 detection boxes into a relative position encoder to perform relative position encoding to obtain an encoding result; Point-multiplying the second convolution result and the encoding result to obtain a point-multiplication result; Adding the first convolution result and the point-multiplication result to obtain an addition result; Performing softmax function calculation and reshape function calculation on the addition result in sequence to obtain a calculation result as the to-be-detected detection box.

5. A thyroid ultrasound abnormal region capturing system characterized by, The system is applied to the method of any one of claims 1-4, and the system comprises: The detection frame acquisition module is configured to acquire detection frames of the same abnormal region in the thyroid ultrasound images of the current frame and the previous N frames of the current frame, thereby composing N+1 detection frames; The to-be-tested detection frame acquisition module is configured to perform spatial coordinate attention mechanism processing on the N+1 detection frames, thereby obtaining a to-be-tested detection frame; The parameter information acquisition module is configured to input the to-be-tested detection frame into a long short-term memory network model, thereby obtaining parameter information of the to-be-tested detection frame; the parameter information includes a position and an area; The detection frame optimization module is configured to obtain, according to the parameter information of the to-be-tested detection frame and the parameter information of the detection frame in the thyroid ultrasound images of the current frame and the previous 1 frame of the current frame, parameter information of an optimized detection frame of the abnormal region in the thyroid ultrasound images of the current frame as an abnormal region detection result, and specifically includes: The detection frame optimization module is configured to obtain, according to the parameter information of the to-be-tested detection frame and the parameter information of the detection frame in the thyroid ultrasound images of the current frame and the previous 1 frame of the current frame, parameter information of an optimized detection frame of the abnormal region in the thyroid ultrasound images of the current frame as an abnormal region detection result, and specifically includes:

6. An electronic device, comprising: The computer program is stored on the storage medium and is executable on the processor, and the processor implements the method according to any one of claims 1 to 4 when executing the computer program.

7. A computer-readable storage medium, characterized in that, The computer program is stored on the storage medium and is executable on the processor, and the processor implements the method according to any one of claims 1 to 4 when executing the computer program.

Images