Ultrasonic imaging medical report generation method and device and computer equipment

Abstract

The invention relates to an ultrasonic imaging medical report generation method and device and computer equipment. The method comprises the following steps: acquiring an inspection record description standard audio file of ultrasonic image inspection, and performing identification processing on the inspection record description standard audio file to obtain an inspection record text; obtaining ultrasonic examination doctor's advice information of ultrasonic image examination, and performing ultrasonic image medical report template matching processing according to the ultrasonic examination doctor's advice information to obtain a target ultrasonic image medical report template; and obtaining a target cue word, and inputting the examination record text, the target ultrasonic image medical report template and the target cue word into the large language model to obtain an ultrasonic image medical report of the ultrasonic image examination. By adopting the method, the efficiency and the accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus and computer equipment for generating ultrasound imaging medical reports. Background Technology

[0002] Ultrasound imaging is a primary source of information for current clinical diagnosis, and providing accurate, clear, and timely ultrasound imaging reports is a crucial requirement for ultrasound imaging diagnosis. Currently, in most hospitals, the ultrasound imaging physician verbally describes the findings and possible ultrasound impressions in real time, and a data entry clerk immediately writes the ultrasound report based on the physician's description using a pre-defined template.

[0003] However, current methods for generating ultrasound imaging medical reports suffer from problems such as low efficiency or accuracy due to the heavy workload of ultrasound departments, the tight timeframe for report issuance, the need for real-time personalized reports for different lesions, and the presence of numerous values ​​of varying orders of magnitude in reports for different examination sites. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, computer equipment, and storage medium for generating ultrasound imaging medical reports that can improve efficiency and accuracy in addressing the aforementioned technical problems.

[0005] Firstly, a method for generating ultrasound imaging medical reports is provided, the method comprising: Obtain the standard audio file describing the ultrasound imaging examination record, and then process the standard audio file to obtain the examination record text. The ultrasound examination order information is obtained, and the target ultrasound imaging medical report template is obtained by matching the ultrasound examination order information with the ultrasound imaging medical report template. The target prompt words are obtained by inputting the examination record text, the target ultrasound imaging medical report template, and the target prompt words into the large language model to obtain the ultrasound imaging medical report of the ultrasound imaging examination.

[0006] In one embodiment, obtaining the standard audio file describing the examination record of an ultrasound imaging examination includes: real-time downloading of the initial audio files and audio sequence files of each examination record description uploaded by the doctor's user terminal of the ultrasound imaging examination; wherein, the initial audio files describing the examination record description adopt the TS stream file format; the audio sequence file is used to record the playback order of each initial audio file describing the examination record description; and generating the standard audio file describing the examination record description based on each initial audio file describing the examination record description and the audio sequence file; wherein, the standard audio file describing the examination record description adopts the WAV file format.

[0007] In one embodiment, generating a standard audio file for inspection record description based on each initial audio file describing the inspection record and an audio sequence file includes: inputting each initial audio file describing the inspection record into a human voice detection model for audio filtering processing to obtain a corresponding filtered audio file describing the inspection record; wherein the audio filtering processing includes noise audio filtering processing and silence audio filtering processing; updating each initial audio file describing the inspection record to a filtered audio file describing the corresponding initial audio file describing the inspection record; performing audio splicing processing based on each initial audio file describing the inspection record and an audio sequence file to obtain a spliced ​​audio file; and performing audio format processing on the spliced ​​audio file to obtain the standard audio file describing the inspection record.

[0008] In one embodiment, the inspection record description standard audio file includes inspection record description standard audio sub-files of preset duration; the inspection record text is obtained after the inspection record description standard audio file is recognized, including: inputting each inspection record description standard audio sub-file into the optimal automatic speech recognition model for recognition processing to obtain the corresponding inspection record sub-text; and performing text concatenation processing on the corresponding inspection record sub-text according to the time order of each inspection record description standard audio sub-file to obtain the inspection record text.

[0009] In one embodiment, the method further includes: acquiring a general audio dataset and each initial candidate automatic speech recognition model, and inputting the general audio dataset into each initial candidate automatic speech recognition model to evaluate the word error rate and obtain a first word error rate for each initial candidate automatic speech recognition model; filtering the initial candidate automatic speech recognition models according to the first word error rate threshold and each first word error rate to obtain corresponding intermediate candidate automatic speech recognition models; acquiring a general medical audio dataset, and inputting the general medical audio dataset into each intermediate candidate automatic speech recognition model to evaluate the word error rate and obtain a second word error rate for each intermediate candidate automatic speech recognition model; filtering the intermediate candidate automatic speech recognition models according to the second word error rate threshold and each second word error rate to obtain corresponding candidate automatic speech recognition models; acquiring a historical ultrasound examination record description audio dataset, and inputting the historical ultrasound examination record description audio dataset into each candidate automatic speech recognition model to evaluate the word error rate and obtain a third word error rate for each candidate automatic speech recognition model; and determining the candidate automatic speech recognition model corresponding to the minimum value among the third word error rates as the optimal automatic speech recognition model.

[0010] In one embodiment, the process of obtaining a target ultrasound imaging medical report template after matching ultrasound examination medical report templates based on ultrasound examination medical order information includes: matching ultrasound examination categories based on ultrasound examination medical order information to obtain a target ultrasound examination category; querying the ultrasound imaging medical report template library based on the target ultrasound examination category to obtain a default ultrasound imaging medical report template; in response to performing a template confirmation operation on the default ultrasound imaging medical report template, determining the default ultrasound imaging medical report template as the target ultrasound imaging medical report template; and in response to performing a template replacement operation on the default ultrasound imaging medical report template, determining the target ultrasound imaging medical report template based on the result of the template replacement operation.

[0011] In one embodiment, the method further includes: inputting an ultrasound imaging medical report into a vector embedding model to obtain ultrasound imaging medical report word vectors; calculating a corresponding first target cosine similarity value based on the ultrasound imaging medical report word vectors and the word vectors of each text block in the ultrasound medical knowledge base; wherein, the first target cosine similarity value is the cosine similarity value between the corresponding word vector and the ultrasound imaging medical report word vector; determining each first target cosine similarity value greater than a pre-set threshold as a corresponding second target cosine similarity value; and sequentially concatenating the text blocks corresponding to each second target cosine similarity value to obtain medical auxiliary diagnosis and treatment text for ultrasound imaging examination.

[0012] Secondly, an ultrasound imaging medical report generation device is provided, which includes an audio file acquisition module, an ultrasound imaging medical report template matching module, and an ultrasound imaging medical report generation module.

[0013] The system includes: an audio file acquisition module for acquiring standard audio files describing ultrasound imaging examination records, and an examination record text for processing; an ultrasound imaging medical report template matching module for acquiring ultrasound examination medical orders, and a target ultrasound imaging medical report template for matching the medical orders; and an ultrasound imaging medical report generation module for acquiring target prompt words, and an ultrasound imaging medical report for generating ultrasound imaging examinations by inputting the examination record text, the target ultrasound imaging medical report template, and the target prompt words into a large language model.

[0014] Thirdly, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described in the above method embodiments.

[0015] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of any of the methods described in the above method embodiments.

[0016] The aforementioned method, apparatus, computer equipment, and storage medium for generating ultrasound imaging medical reports acquire a standard audio file describing the examination record of an ultrasound imaging examination, and obtain the examination record text after recognizing and processing the standard audio file. Then, it acquires the ultrasound examination medical order information, performs ultrasound imaging medical report template matching processing based on the ultrasound examination medical order information, and obtains the target ultrasound imaging medical report template. Next, it acquires target prompt words, and inputs the examination record text, the target ultrasound imaging medical report template, and the target prompt words into a large language model to obtain the ultrasound imaging medical report, thus improving the efficiency and accuracy of generating ultrasound imaging medical reports. Attached Figure Description

[0017] Figure 1 This is a diagram illustrating the application environment of an ultrasound imaging medical report generation method in one embodiment. Figure 2 This is a schematic diagram of the first process of a method for generating ultrasound imaging medical reports in one embodiment; Figure 3 This is a schematic diagram of the process for obtaining a standard audio file describing the examination record of an ultrasound imaging examination in one embodiment. Figure 4 This is a schematic diagram of the process for generating a standard audio file for inspection record descriptions based on the initial audio file and the audio sequence file for each inspection record description in one embodiment. Figure 5 This is a flowchart illustrating the process of obtaining inspection record text after recognizing and processing a standard audio file describing the inspection record in one embodiment. Figure 6 This is a flowchart illustrating the process of obtaining a target ultrasound imaging medical report template after matching ultrasound imaging medical report templates based on ultrasound examination medical order information in one embodiment. Figure 7 This is a schematic diagram of the second process of an ultrasound imaging medical report generation method in one embodiment; Figure 8 This is a structural block diagram of an ultrasound imaging medical report generation device in one embodiment; Figure 9 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0019] To facilitate understanding of this application, a more complete description will be provided below with reference to the accompanying drawings, which illustrate embodiments of the present application. However, the present application can be implemented in many different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided so that the disclosure of this application will be thorough and complete.

[0020] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0021] It is understood that the terms "first," "second," etc., used herein may be used to describe various elements, but these elements are not limited by these terms. These terms are only used to distinguish one element from another. For example, without departing from the scope of this application, a first resistor may be referred to as a second resistor, and similarly, a second resistor may be referred to as a first resistor. Both the first resistor and the second resistor are resistors, but they are not the same resistor.

[0022] It is understood that the term "connection" in the following embodiments should be understood as "electrical connection," "communication connection," etc., if the connected circuits, modules, units, etc., have electrical signal or data transmission with each other.

[0023] When used herein, the singular forms of “a,” “an,” and “the” may also include the plural forms unless the context clearly indicates otherwise. It should also be understood that the terms “comprising,” “including,” or “having,” etc., specify the presence of the stated feature, whole, step, operation, component, part, or combination thereof, but do not preclude the possibility of the presence or addition of one or more other features, wholes, steps, operations, components, parts, or combinations thereof.

[0024] The ultrasound imaging medical report generation method provided in this application can be applied to, for example... Figure 1 In the application environment shown, the doctor's user terminal 102 communicates with the server 104 via a network. The doctor's user terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server 104 can be a standalone server or a server cluster consisting of multiple servers.

[0025] In one embodiment, such as Figure 2 As shown, a method for generating ultrasound imaging medical reports is provided, which can be applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 201 to 203.

[0026] Step 201: Obtain the standard audio file describing the ultrasound imaging examination record, and obtain the examination record text after recognizing and processing the standard audio file describing the examination record record.

[0027] The examination record text is used to record the content verbally described by the doctor during the ultrasound imaging examination. Specifically, server 104 obtains the standard audio file of the ultrasound imaging examination examination record description, and obtains the examination record text after recognizing and processing the standard audio file of the examination record description.

[0028] In one embodiment, such as Figure 3 As shown, the standard audio file describing the examination record of the ultrasound imaging examination is obtained, including steps 301 to 302.

[0029] Step 301: Download in real time the initial audio files and audio sequence files describing each examination record uploaded by the doctor's user terminal for ultrasound imaging examinations.

[0030] The initial audio files describing the examination records are in TS stream format; the audio sequence file records the playback order of each initial audio file describing the examination records. Specifically, server 104 downloads in real time the initial audio files describing the examination records and the audio sequence file uploaded by the doctor's user terminal during the ultrasound imaging examination.

[0031] In a specific example, when the physician conducting the ultrasound imaging examination clicks the record button on the physician's user terminal 102, real-time recording of the physician's verbal descriptions during the ultrasound examination is initiated, generating initial audio files and sequential audio files for each examination record description. Then, the initial audio files for each examination record description are automatically uploaded to the server 104. The duration of the initial audio files for the examination record descriptions is a first target duration; the first target duration can be, but is not limited to, 10 seconds. The sequential audio files can be, but are not limited to, M3U8 audio sequential metadata files. The above is only a specific example; in actual applications, it can be flexibly set according to user needs, and no restrictions are imposed here.

[0032] Step 302: Generate standard audio files for each inspection record description based on the initial audio files and audio sequence files for each inspection record description.

[0033] The standard audio file describing the inspection record uses the WAV file format. Specifically, server 104 generates the standard audio file describing the inspection record based on the initial audio file and the audio sequence file for each inspection record description. This effectively reduces audio transmission latency and improves real-time performance by downloading the initial audio file and the audio sequence file in real time. At the same time, by splicing the initial audio files describing the inspection record description based on the audio sequence file to generate the standard audio file describing the inspection record description, the accuracy of the audio file is ensured.

[0034] In one embodiment, such as Figure 4 As shown, a standard audio file for the inspection record description is generated based on the initial audio file and the audio sequence file for each inspection record description, including steps 401 to 404.

[0035] Step 401: Input the initial audio files describing each inspection record into the human voice detection model for audio filtering processing to obtain the corresponding filtered audio files describing the inspection records.

[0036] The audio filtering process includes noise audio filtering and silence audio filtering. Specifically, server 104 inputs the initial audio files describing each inspection record into the human voice detection model for audio filtering, resulting in the corresponding filtered audio files describing the inspection records. This ensures that the filtered audio files describing the inspection records have discarded silence and noise audio, improving the accuracy of subsequent speech recognition and avoiding the waste of speech recognition resources.

[0037] In a specific example, the voice detection model can be, but is not limited to, the Silero VAD model. Setting the detection threshold of the Silero VAD model to 0.5 can filter out silent and noisy audio, keeping only the audio files containing human voices. This is just a specific example; in actual applications, the settings can be flexibly adjusted according to user needs, and no restrictions are imposed here.

[0038] Step 402: Update the initial audio file of each inspection record description to the corresponding filtered audio file of the initial audio file of the inspection record description.

[0039] Step 403: Based on the descriptions of the initial audio file and the audio sequence file in each inspection record, perform audio splicing processing to obtain the spliced ​​audio file.

[0040] Step 404: Perform audio format processing on the spliced ​​audio file to obtain the standard audio file for generating the inspection record description.

[0041] Specifically, server 104 updates each initial audio file describing the inspection record to the corresponding filtered audio file describing the inspection record. Then, it performs audio concatenation processing based on each initial audio file describing the inspection record and the audio sequence file to obtain a concatenated audio file. Next, it processes the concatenated audio file to obtain a standard audio file describing the inspection record, thereby ensuring that the filtered audio file describing the inspection record has discarded silent and noisy audio, improving the accuracy and efficiency of subsequent speech recognition, and avoiding the waste of speech recognition resources.

[0042] In this embodiment, each initial audio file describing the inspection record is updated to a corresponding filtered audio file describing the inspection record. Then, audio concatenation is performed based on each initial audio file describing the inspection record and the audio sequence file to obtain a concatenated audio file. Next, the concatenated audio file undergoes audio format processing to generate a standard audio file describing the inspection record. This ensures that the filtered audio file describing the inspection record has discarded silent and noisy audio, improving the accuracy and efficiency of subsequent speech recognition while avoiding resource waste in speech recognition.

[0043] In this embodiment, the initial audio files and audio sequence files describing each examination record uploaded by the doctor's user terminal during the ultrasound imaging examination are downloaded in real time. Then, a standard audio file describing the examination record is generated based on the initial audio files and audio sequence files. This effectively reduces audio transmission latency and improves real-time performance by downloading the initial audio files and audio sequence files in real time. At the same time, the standard audio file describing the examination record is generated by splicing the initial audio files describing each examination record based on the audio sequence file, ensuring the accuracy of the audio file.

[0044] In one embodiment, the inspection record description standard audio file includes inspection record description standard audio sub-files of various preset durations.

[0045] In a specific example, the preset duration of the check record describing the standard audio sub-file can be, but is not limited to, 50 seconds. The above is just a specific example, and in actual applications, it can be flexibly set according to user needs. There are no restrictions here.

[0046] Among them, such as Figure 5 As shown, the inspection record text is obtained after the standard audio file describing the inspection record is identified and processed, including steps 501 to 502.

[0047] Step 501: Input the standard audio sub-files describing each inspection record into the optimal automatic speech recognition model for recognition processing to obtain the corresponding inspection record sub-text.

[0048] Specifically, server 104 inputs the standard audio sub-files describing each inspection record into the optimal automatic speech recognition model for recognition processing to obtain the corresponding inspection record sub-text.

[0049] In one embodiment, the method further includes: Obtain a general audio dataset and each initial candidate automatic speech recognition model. Input the general audio dataset into each initial candidate automatic speech recognition model to evaluate the word error rate and obtain the first word error rate of each initial candidate automatic speech recognition model. Based on the first word error rate threshold and the error rate of each first word, the initial candidate automatic speech recognition models are screened to obtain the corresponding intermediate candidate automatic speech recognition models. A general medical audio dataset is obtained, and the general medical audio dataset is input into each intermediate candidate automatic speech recognition model to evaluate the word error rate, and then the second word error rate of each intermediate candidate automatic speech recognition model is obtained. Based on the second word error rate threshold and the error rate of each second word, the intermediate candidate automatic speech recognition models are screened to obtain the corresponding candidate automatic speech recognition models. Obtain the audio dataset describing historical ultrasound examination records, and input the audio dataset describing historical ultrasound examination records into each candidate automatic speech recognition model to evaluate the word error rate and obtain the third word error rate of each candidate automatic speech recognition model; The candidate automatic speech recognition model corresponding to the minimum error rate of each third word is determined as the optimal automatic speech recognition model.

[0050] Specifically, server 104 acquires a general audio dataset and each initial candidate automatic speech recognition model, and inputs the general audio dataset into each initial candidate automatic speech recognition model to evaluate the word error rate, thereby obtaining the first word error rate of each initial candidate automatic speech recognition model; then... The initial candidate automatic speech recognition (ASR) models are screened based on the first word error rate threshold and the error rates of each first word, resulting in intermediate candidate ASR models. Next, a general medical audio dataset is acquired and input into each intermediate candidate ASR model for word error rate evaluation, yielding the second word error rate for each intermediate candidate ASR model. Then, the intermediate candidate ASR models are screened based on the second word error rate threshold and the error rates of each second word, resulting in corresponding candidate ASR models. Simultaneously, a historical ultrasound examination record description audio dataset is acquired and input into each candidate ASR model for word error rate evaluation, yielding the third word error rate for each candidate ASR model. Finally, the candidate ASR model corresponding to the minimum third word error rate is determined as the optimal ASR model, thus selecting the best-performing ASR model and improving the efficiency and accuracy of speech recognition.

[0051] In a specific example, seven open-source ASR models were deployed on server 104, including the whisper-large-v3 open-source ASR model, the qwen2-audio open-source ASR model, the paraformer open-source ASR model, and the baichuan-omni open-source ASR model. Model selection was based on the Word Error Rate (WER). WER is one of the evaluation metrics for ASR models; it calculates the degree of mismatch between the ASR model's output and the reference text. The formula for WER is WER = (S + D + I) / N, where S represents substitution error, D represents deletion error, I represents insertion error, and N represents the total number of words. A lower WER indicates higher system recognition accuracy.

[0052] The general medical audio dataset includes 200 target audio files, which are obtained by inputting corresponding real ultrasound imaging medical reports into a TTS (Text-to-Speech) model. Intermediate candidate automatic speech recognition (ASR) models are selected based on the second-word error rate threshold and the error rates of each second word. The performance of each intermediate candidate ASR model in medical terminology can be evaluated using the second-word error rate to ensure that the candidate ASR models can recognize ultrasound examination terminology. Furthermore, the candidate ASR models include the open-source xunfei, baichaun-omni, paraformer_zh, SenseVoiceSmall, and whisper-large-v3 open-source ASR models.

[0053] The historical ultrasound examination record description audio dataset includes N real historical ultrasound examination record description audios. The historical ultrasound examination record description audio dataset was input into each candidate automatic speech recognition model to evaluate the word error rate, resulting in the third-word error rate for each candidate automatic speech recognition model. Specifically, the third-word error rates for each candidate automatic speech recognition model are as follows: xunfei open-source ASR model: 0.0576; baichaun-omni open-source ASR model: 0.0484; paraformer_zh open-source ASR model: 0.099; SenseVoiceSmall open-source ASR model: 0.085; and whisper-large-v3 open-source ASR model: 0.20. The candidate automatic speech recognition model corresponding to the minimum error rate among all third-word options was determined as the optimal automatic speech recognition model, namely the baichaun-omni open-source ASR model. Version 7b of the baichaun-omni open-source ASR model was deployed on server 104, thus ensuring an accuracy rate of over 95%. The above is only a specific example; in actual applications, settings can be flexibly configured according to user needs, and no restrictions are imposed here.

[0054] In this embodiment, a general audio dataset and each initial candidate automatic speech recognition model are obtained. The general audio dataset is then input into each initial candidate automatic speech recognition model to evaluate the word error rate, resulting in the first word error rate of each initial candidate automatic speech recognition model. Then... The initial candidate automatic speech recognition (ASR) models are screened based on the first word error rate threshold and the error rates of each first word, resulting in intermediate candidate ASR models. Next, a general medical audio dataset is acquired and input into each intermediate candidate ASR model for word error rate evaluation, yielding the second word error rate for each intermediate candidate ASR model. Then, the intermediate candidate ASR models are screened based on the second word error rate threshold and the error rates of each second word, resulting in corresponding candidate ASR models. Simultaneously, a historical ultrasound examination record description audio dataset is acquired and input into each candidate ASR model for word error rate evaluation, yielding the third word error rate for each candidate ASR model. Finally, the candidate ASR model corresponding to the minimum third word error rate is determined as the optimal ASR model, thus selecting the best-performing ASR model and improving the efficiency and accuracy of speech recognition.

[0055] Step 502: Based on the time sequence of each inspection record description standard audio sub-file, perform text concatenation on the corresponding inspection record sub-text to obtain the inspection record text.

[0056] Specifically, server 104 performs text concatenation on the corresponding inspection record sub-texts according to the time sequence of each inspection record description standard audio sub-file to obtain the inspection record text, thereby improving the accuracy and acquisition efficiency of the inspection record text.

[0057] In this embodiment, the standard audio sub-files describing each inspection record are input into the optimal automatic speech recognition model for recognition processing to obtain the corresponding inspection record sub-text. Then, the corresponding inspection record sub-texts are concatenated according to the time order of each standard audio sub-file describing the inspection record to obtain the inspection record text, thereby improving the accuracy and efficiency of obtaining the inspection record text.

[0058] Step 202: Obtain the ultrasound examination order information, and obtain the target ultrasound examination medical report template by matching the ultrasound examination order information with the ultrasound medical report template.

[0059] Specifically, server 104 obtains ultrasound examination medical order information and performs ultrasound imaging medical report template matching processing based on ultrasound examination medical order information to obtain target ultrasound imaging medical report template.

[0060] In one embodiment, such as Figure 6 As shown, the target ultrasound imaging medical report template is obtained after matching the ultrasound examination medical order information with the ultrasound imaging medical report template, including steps 601 to 604.

[0061] Step 601: Match the ultrasound examination category according to the ultrasound examination medical order information to obtain the target ultrasound examination category.

[0062] Specifically, server 104 performs ultrasound examination category matching based on ultrasound examination medical order information to obtain the target ultrasound examination category.

[0063] In a specific example, the target ultrasound examination category is the subcategory of the corresponding major target ultrasound examination category. The major target ultrasound examination category is obtained by matching from each major ultrasound examination category based on the ultrasound examination medical order information. The subcategory of the target ultrasound examination is obtained by matching from each subcategory of the major target ultrasound examination category based on the ultrasound examination medical order information.

[0064] Specifically, the major categories of ultrasound examinations include "Category 0: Breast", "Category 1: Obstetrics", "Category 2: Other", "Category 3: Gynecology", "Category 4: Cardiac", "Category 5: Early, Mid, and Late Pregnancy and NT", "Category 6: Thyroid", "Category 7: Abdomen + Mesentery", "Category 8: Vascular", "Category 9: Appendix", "Category 10: Scrotum" and "Category 11: Hip Joint".

[0065] Specifically, "Category 4 Heart Disease" includes "Category 0 Coronary Artery Disease", "Category 1 Left Ventricular Systolic Function Impairment", "Category 2 Left Ventricular Diastolic Function Impairment", "Category 3 Pericardial Effusion", "Category 4 Normal" and "Category 5 Valvular Regurgitation".

[0066] Specifically, "Class 0 breast" includes "Class 0 breast type 2 single lesion", "Class 1 breast type 2 multiple lesion", "Class 2 breast type 3 single lesion", "Class 3 breast type 3 multiple lesion", "Class 4 breast type 4 and 5" and "Class 5 normal breast".

[0067] Specifically, under "Class 8 Vascular Vessels" are "Class 0: Normal Arteries in Both Upper Limbs", "Class 1: Normal Veins in Both Upper Limbs (Breast Area)", "Class 2: Multiple Plaques in Arteries of Both Lower Limbs", "Class 3: Normal Arteries in Both Lower Limbs", "Class 4: Deep Vein Thrombosis in Both Lower Limbs", "Class 5: Normal Veins in Both Lower Limbs", "Class 6: Normal Abdominal Aorta", "Class 7: Carotid Intima-Media Thickening", "Class 8: Normal Plaques in the Abdominal and Carotid Arteries", and "Class 9: Normal Carotid Artery". These are just examples; in actual applications, the settings can be flexibly adjusted according to user needs, and no restrictions are imposed here.

[0068] Step 602: Search the ultrasound imaging medical report template library according to the target ultrasound examination category to obtain the default ultrasound imaging medical report template.

[0069] Step 603: In response to the template confirmation operation of the default ultrasound imaging medical report template, the default ultrasound imaging medical report template is determined as the target ultrasound imaging medical report template.

[0070] Step 604: In response to the template replacement operation on the default ultrasound imaging medical report template, determine the target ultrasound imaging medical report template based on the result of the template replacement operation.

[0071] Specifically, server 104 queries the ultrasound imaging medical report template library according to the target ultrasound examination category to obtain a default ultrasound imaging medical report template; then, in response to the default ultrasound imaging medical report template, it performs a template confirmation operation to determine the default ultrasound imaging medical report template as the target ultrasound imaging medical report template; at the same time, in response to the default ultrasound imaging medical report template, it performs a template replacement operation to determine the target ultrasound imaging medical report template based on the result of the template replacement operation, thereby improving the adaptability and efficiency of the target ultrasound imaging medical report template.

[0072] In a specific example, if the target ultrasound examination category is found to be inconsistent with the ultrasound examination category required for the actual examination during an ultrasound imaging examination, the default ultrasound imaging medical report template can be directly replaced. This allows the target ultrasound imaging medical report template to be retrieved from the ultrasound imaging medical report template library based on the ultrasound examination category required for the actual examination. Alternatively, if the default ultrasound imaging medical report template is found to be unsuitable during the ultrasound imaging examination, a template replacement operation can be performed to use a custom ultrasound imaging medical report template as the target ultrasound imaging medical report template. The above are just specific examples; in actual applications, the settings can be flexibly configured according to user needs, and no restrictions are imposed here.

[0073] In this embodiment, ultrasound examination category matching is performed based on ultrasound examination medical order information to obtain the target ultrasound examination category; then, the target ultrasound examination category is queried in the ultrasound imaging medical report template library to obtain the default ultrasound imaging medical report template; subsequently, in response to the template confirmation operation of the default ultrasound imaging medical report template, the default ultrasound imaging medical report template is determined as the target ultrasound imaging medical report template; simultaneously, in response to the template replacement operation of the default ultrasound imaging medical report template, the target ultrasound imaging medical report template is determined based on the result of the template replacement operation, thereby improving the adaptability and efficiency of the target ultrasound imaging medical report template.

[0074] Step 203: Obtain target prompt words. Input the examination record text, the target ultrasound imaging medical report template, and the target prompt words into the large language model to obtain the ultrasound imaging medical report of the ultrasound imaging examination.

[0075] The target prompt words guide the large language model to output an ultrasound imaging report based on the examination record text and the corresponding target ultrasound imaging report template. In other words, there is a one-to-one correspondence between the target prompt words and the target ultrasound imaging report template. Specifically, server 104 obtains the target prompt words, inputs the examination record text, the target ultrasound imaging report template, and the target prompt words into the large language model to obtain the ultrasound imaging report, thus improving the efficiency and accuracy of generating the ultrasound imaging report.

[0076] In a specific example, the large language model can be, but is not limited to, a dense structured large language model. The above is just a specific example, and in actual applications, it can be flexibly set according to user needs, without any restrictions.

[0077] In a specific example, the target prompt word includes: "{Inspection findings: Instructions: You are an ultrasound expert. Please combine the abnormal findings from this ultrasound examination that I have entered with the ultrasound report template I have provided, and merge them. You must fill in the corresponding locations in the ultrasound report template with the abnormal findings according to your medical understanding. Also, delete any initial or contradictory descriptions in the report template. Regenerate a complete ultrasound report. Please note that you should not generate the diagnostic impressions section from the template.

[0078] Specific requirements are as follows: 1. Format Requirements: The generated report must strictly adhere to the fixed format, generating only the content of the images seen. The title "Examination Findings" should not be included, even if the examination content contains ultrasound conclusions; no explanations or elaborations should be added to the generated report. The final report should be returned in text format. 2. Generation Principles: Prioritize Retaining Examination Content: If content mentioned in the examination content is not described in the template, it should be retained in the generated report. 3. Template Filling: If the template contains content that needs to be filled (such as selection options or values), and the examination content mentions relevant information, it should be filled into the template. 4. Marking Conflicts: If the template uses " / " to indicate a need for explicit selective description by the user, such as anteverted / retroverted uterus, but the ultrasound examination content does not explicitly mention this, it should be marked with 【】 in the generated report, such as: Uterus

anteverted / retroverted

x

[0079] Please generate an ultrasound report according to the instructions above. Understandably, the examination record text included: "Description of ultrasound abnormality: "Hyperechoic nodule in the right lobe of the liver, approximately 10x10mm in size, regular in shape, with clear borders. An isoechoic mass, approximately 2mm in diameter, is visible on the gallbladder wall."

[0080] Understandably, the target ultrasound imaging medical report template includes the following: "Imaging findings: The liver has a normal outline and shape, a smooth and regular capsule, and homogeneous parenchymal echoes. No obvious abnormalities were observed in the intrahepatic ductal structures. CDFI: Blood flow in the portal vein, hepatic vein, and hepatic artery is normal. The gallbladder is of normal size, with a smooth wall and good acoustic transmission within the lumen; no obvious abnormalities were observed. The common bile duct is not dilated. The pancreas has a normal shape and homogeneous echoes. The main pancreatic duct is not dilated. The spleen has normal thickness and shape, homogeneous parenchymal echoes, and no abnormal blood flow. Both kidneys are in normal position, size, and shape; the capsules are smooth and intact, the corticomedullary boundary is clear, and the collecting system is not separated. CDFI: Blood flow signal distribution in both kidneys is normal. No obvious dilation was observed in either ureter."

[0081] Specifically, the contents of an ultrasound imaging medical report include: "The liver has a normal outline and shape, with a smooth and regular capsule. A hyperechoic nodule, approximately 10x10mm in size, is visible in the right lobe of the liver. It is regular in shape, with clear borders, and the remaining parenchyma has homogeneous echogenicity. No obvious abnormalities were observed in the intrahepatic ductal structures. Color Doppler flow imaging (CDFI) showed normal blood flow in the portal vein, hepatic vein, and hepatic artery. The gallbladder is of normal size with a smooth wall. An isoechoic nodule, approximately 2mm in diameter, is visible on the gallbladder wall. The lumen has good sound transmission, and no obvious abnormal echoes were observed. No obvious dilation of the common bile duct was observed. The pancreas has a normal shape and homogeneous echogenicity. No obvious dilation of the main pancreatic duct was observed. The spleen has normal thickness, homogeneous parenchymal echogenicity, and no obvious abnormalities in blood flow. Both kidneys are of normal size and shape, with smooth and intact capsules, clear corticomedullary boundaries, and no obvious separation of the collecting system. Color Doppler flow imaging (CDFI) showed normal blood flow signal distribution in both kidneys. No obvious dilation of both ureters was observed." The above is only a specific example and can be flexibly configured according to user needs in actual applications. No restrictions are imposed here.

[0082] In a specific example, an ultrasound imaging medical report includes information on the findings and ultrasound impressions.

[0083] Exemplary ultrasound imaging medical reports may include findings such as "the liver has a normal outline and shape, a smooth and regular capsule, and homogeneous parenchymal echoes."

[0084] No obvious abnormalities were found in the intrahepatic ductal structures. Color Doppler flow imaging (CDFI) showed normal blood flow in the portal vein, hepatic vein, and hepatic artery.

[0085] The gallbladder is of normal size, with a smooth wall and good sound transmission within the lumen; no obvious abnormalities were observed. No dilation of the common bile duct was observed.

[0086] The pancreas is of normal shape and has homogeneous echogenicity. No dilation of the main pancreatic duct was observed.

[0087] The spleen was of normal thickness and shape, with uniform parenchymal echogenicity, and no abnormalities were observed in blood flow.

[0088] Both kidneys are normal in position, size and shape, with smooth and intact capsules, clear cortical-medullary boundaries, and no separation of the collecting system. A fluid-filled cyst is visible in the right kidney, measuring approximately 7x5mm, and in the left kidney, a fluid-filled cyst is visible, measuring approximately 31x24mm. Both are regular in shape, with clear boundaries and good sound transmission. CDFI: Blood flow signal distribution is normal in both kidneys.

[0089] No obvious dilation was observed in either ureter.

[0090] An example ultrasound impression in a medical report might include "Bilateral renal cysts; no obvious abnormalities found in the liver, gallbladder, spleen, or pancreas." This is merely an example; in practical applications, settings can be flexibly adjusted according to user needs, and no limitations are imposed here.

[0091] In the above-mentioned method for generating ultrasound imaging medical reports, a standard audio file describing the examination record of an ultrasound imaging examination is obtained, and the examination record text is obtained after recognition processing of the standard audio file describing the examination record. Then, the ultrasound examination medical order information is obtained, and the ultrasound imaging medical report template is obtained after matching the ultrasound examination medical order information with the ultrasound imaging medical report template. Next, the target prompt words are obtained, and the examination record text, the target ultrasound imaging medical report template, and the target prompt words are input into a large language model to obtain the ultrasound imaging medical report of the ultrasound imaging examination, which improves the efficiency and accuracy of generating ultrasound imaging medical reports.

[0092] In one embodiment, such as Figure 7 As shown, the method further includes steps 701 to 704.

[0093] Step 701: Input the ultrasound imaging medical report into the vector embedding model to obtain the word vector of the ultrasound imaging medical report.

[0094] Step 702: Calculate the corresponding first target cosine similarity value based on the word vectors of the ultrasound imaging medical report and the word vectors of each text block in the ultrasound medical knowledge base.

[0095] Step 703: Determine the first target cosine similarity value of each first target cosine similarity value that is greater than the cosine similarity value threshold as the corresponding second target cosine similarity value.

[0096] Step 704: Concatenate the text blocks corresponding to the cosine similarity values ​​of each second target in sequence to obtain the medical auxiliary diagnosis and treatment text of ultrasound imaging examination.

[0097] Among them, the first target cosine similarity value is the cosine similarity value between the corresponding word vector and the word vector of the ultrasound imaging medical report.

[0098] Specifically, server 104 inputs the ultrasound imaging medical report into the vector embedding model to obtain the word vector of the ultrasound imaging medical report; then, it calculates the corresponding first target cosine similarity value based on the word vector of the ultrasound imaging medical report and the word vector of each text block in the ultrasound medical knowledge base; next, it determines the first target cosine similarity value as the corresponding second target cosine similarity value for each first target cosine similarity value that is greater than the cosine similarity value threshold; then, it concatenates the text blocks corresponding to each second target cosine similarity value in sequence to obtain the medical auxiliary diagnosis text of ultrasound imaging examination, which improves the convenience and clinical applicability of generating ultrasound imaging medical reports.

[0099] In one specific example, the method also includes the step of constructing an ultrasound medical knowledge base.

[0100] The steps involved in constructing an ultrasound medical knowledge base include: Obtain a second preset number of ultrasound medical book texts; wherein the ultrasound medical book texts are in PDF format; The texts of various ultrasound medical books are converted to their corresponding target medical book texts; the target medical book texts are in MD format. Based on the preset content length and the table of contents of each target medical book text, the corresponding target medical book text is divided into text chunks to obtain each text chunk of the target medical book text. Each text block corresponding to the text of each target medical book is input into the vector embedding model to obtain the word vectors of each text block corresponding to the text of each target medical book. An ultrasound medical knowledge base is constructed based on the word vectors of each text block corresponding to the text of each target medical book.

[0101] It is understandable that the ultrasound medical knowledge base can be, but is not limited to, a .pkl file. Vector embedding models can be, but are not limited to, embedding models.

[0102] Specifically, the storage information of an illustrative text block includes: "content: '## Real-time ultrasound examination is used, with probes that can be transabdominal or transvaginal. An appropriate probe frequency should be selected. Real-time ultrasound assesses fetal viability by observing fetal heartbeats and activity. Note that probe frequency selection should consider both ultrasound penetration and image resolution. For transabdominal examinations, a probe frequency of 3-5MHz is generally chosen to achieve the necessary penetration while obtaining high-resolution ultrasound images...'; 'metadata': {'First-level heading': 'ULTRASONOGRAPHY', 'Second-level heading': 'Part Three: Equipment Specifications'; 'block_id': 26; 'span_id': 1; 'source_file': 'Obstetrics and Gynecology Ultrasound Medicine'};".

[0103] The fields in the aforementioned text block include: `content` is the actual content of the current block, which is also the text that needs to be vectorized later; `source_file` is the book to which the current text block belongs; `block_id` indicates the sequential code of the current text block within the book; `span_id` indicates the sequential code after segmentation exceeding the length under a `block_id`. `span_id` and `block_id` are used to reconstruct paragraphs and chapters during subsequent retrieval. `source_file` is used to display the source book for auxiliary diagnosis.

[0104] The above are just specific examples. In actual applications, the settings should be flexibly adjusted according to user needs, and no restrictions are imposed here.

[0105] In a specific example, this applies when the target ultrasound examination category is abdominal ultrasound.

[0106] The specific content of the examination record text includes: "The liver outline and shape are normal, the capsule is regular and smooth, and the parenchyma has uniform echo. The gallbladder is of normal size, the wall is relatively smooth, and a slightly strong echo of about 11x7mm is detected near the base, with relatively clear borders and posterior acoustic shadowing..."

[0107] The target ultrasound imaging medical report template is an abdominal ultrasound examination template; After inputting the examination record text, the target ultrasound imaging medical report template, and the target prompt words into the large language model, the ultrasound imaging medical report for the ultrasound examination is obtained. The specific content of the ultrasound imaging medical report includes: "Examination findings: The liver has a normal outline and shape, a regular and smooth capsule, and homogeneous parenchymal echo. The gallbladder is of normal size, with a relatively smooth wall. Near the base, a slightly hyperechoic mass measuring approximately 11x7mm was detected, with relatively clear borders and posterior acoustic shadowing."

[0108] Diagnostic impression: Mild fatty liver. Gallstones suspected.

[0109] The specific content of medical auxiliary diagnostic texts for ultrasound imaging examinations includes: "Most likely diagnosis" Gallstones: Ultrasound showed a slightly strong echo of about 11×7mm near the bottom of the gallbladder, with relatively clear borders and posterior acoustic shadowing. This is a typical manifestation of gallstones.

[0110] Mild fatty liver: The liver parenchyma shows uniformly increased echogenicity, consistent with the ultrasound findings of mild fatty liver.

[0111] Differential diagnosis Gallbladder polyps: - They typically appear as medium to high echogenicity attached to the gallbladder wall, with no posterior acoustic shadowing, and do not move with changes in body position. - In this case, the lesion has posterior acoustic shadowing, further supporting the diagnosis of gallstones.

[0112] Gallbladder adenomyomatosis: - Characterized by localized or diffuse thickening of the gallbladder wall, possibly accompanied by small cystic anechoic areas (Rogner-Arc). - In this case, the gallbladder wall is smooth, which does not support this diagnosis.

[0113] Gallbladder sludge or concentrated bile: - May appear as hypoechoic or anechoic deposits within the gallbladder, without posterior acoustic shadowing, and shifts with body position. - In this case, it is hyperechoic with acoustic shadowing, which does not support the presence of gallbladder sludge.

[0114] Precautions Complications related to gallstones: - Watch for signs of cholecystitis (such as thickened gallbladder wall, surrounding fluid accumulation, and a positive Murphy's sign). - Be alert for stone impaction or common bile duct stones (observation of bile duct dilation is necessary).

[0115] Fatty liver assessment: - It is necessary to combine liver function, blood glucose, blood lipid and other tests to assess metabolic abnormalities.

[0116] Technical factors: - Multi-section scanning to confirm the location and number of stones, and changing body position if necessary to observe stone mobility.

[0117] Follow-up recommendations Gallstones: - Asymptomatic individuals: It is recommended to have a follow-up ultrasound examination every 6-12 months to observe changes in the stones. - If symptoms such as right upper quadrant pain or fever occur, it is necessary to seek medical attention promptly to rule out cholecystitis. - If the stones enlarge (>1cm) or are accompanied by gallbladder wall thickening, it is recommended to have a surgical consultation to assess surgical indications.

[0118] Fatty liver: - Lifestyle intervention (dietary control, exercise) is recommended. Follow up with ultrasound and liver function tests after 6 months. - Individuals with concurrent metabolic abnormalities need regular monitoring of blood glucose and blood lipids. The above are just specific examples; in actual application, settings can be flexibly adjusted according to user needs, and no restrictions are imposed here.

[0119] In a specific example, the target ultrasound examination category is thyroid ultrasound examination.

[0120] The specific content of the examination record text includes: "The right lobe of the thyroid gland measures approximately 25x22x65mm, the left lobe measures approximately 26x19x64mm, and the isthmus is approximately 6mm thick. The thyroid gland is full and enlarged, with slight thickening of the isthmus and diffuse coarsening of internal echoes, showing a network-like change..."

[0121] The target ultrasound imaging medical report template is the thyroid ultrasound examination template selected by the system.

[0122] After inputting the examination record text, the target ultrasound imaging medical report template, and the target prompt words into the large language model, the ultrasound imaging medical report for the ultrasound examination is obtained. The specific content of the ultrasound imaging medical report includes: Findings during the inspection: The right lobe of the thyroid gland measures approximately 25x22x65mm, the left lobe measures approximately 26x19x64mm, and the isthmus is approximately 6mm thick.

[0123] The thyroid gland is full and enlarged, with slight thickening of the isthmus and diffuse coarsening of internal echoes, showing a network-like appearance.

[0124] Diagnostic Impression: The sonographic findings of diffuse thyroid lesions suggest Hashimoto's thyroiditis; further thyroid function tests are recommended.

[0125] The specific content of medical auxiliary diagnostic texts for ultrasound imaging examinations includes: Most likely diagnosis: Hashimoto's thyroiditis (chronic lymphocytic thyroiditis). - Basis: Ultrasound shows diffuse enlargement of the thyroid gland, diffuse thickening of internal echoes with a network-like appearance, and thickening of the isthmus, all of which are typical ultrasound manifestations of Hashimoto's thyroiditis.

[0126] Differential diagnosis: Graves' disease (toxic diffuse goiter) - Ultrasound findings: Diffuse enlargement of the thyroid gland, with decreased and heterogeneous echogenicity and abundant blood flow signals ("fire sign"). - Clinical differentiation: Requires consideration of thyroid function (e.g., decreased TSH, increased FT3 / FT4) and antibodies (TRAb positive).

[0127] Subacute thyroiditis - Ultrasound findings: Patchy hypoechoic areas in the thyroid gland with indistinct borders, significant tenderness, and no blood flow signal. - Clinical differentiation: Often presents with a history of neck pain and fever, and elevated erythrocyte sedimentation rate (ESR).

[0128] Simple goiter - Ultrasound findings: The thyroid gland is symmetrically enlarged with homogeneous echogenicity and no reticular pattern. - Clinical differentiation: More common in iodine-deficient areas; thyroid function is usually normal.

[0129] Precautions for ultrasound examination Key observation points: - Blood flow signals within the thyroid gland (increased blood flow may be present in the early stages of Hashimoto's thyroiditis, decreasing in later stages). - Whether nodules are present (the risk of malignant transformation of nodules is slightly increased in patients with Hashimoto's thyroiditis).

[0130] Additional assessment: - Cervical lymph nodes (Hashimoto's thyroiditis can be accompanied by reactive lymphadenopathy; metastatic lymphadenopathy needs to be ruled out). ---## Follow-up recommendations Laboratory tests: - Complete thyroid function tests (TSH, FT3, FT4) and antibody tests (TPOAb, TgAb) to determine if there is hyperthyroidism or hypothyroidism.

[0131] Ultrasound follow-up: - If there are no symptoms and the function is normal, have an ultrasound follow-up every 1-2 years; if nodules or functional abnormalities are found, shorten the follow-up interval.

[0132] 3. Clinical Management: - If hypothyroidism (elevated TSH) occurs, levothyroxine replacement therapy is required; if hyperthyroidism symptoms are significant, symptomatic treatment can be administered. The above is only a specific example; in actual application, it can be flexibly set according to user needs, and no restrictions are imposed here.

[0133] In this embodiment, the ultrasound imaging medical report is input into a vector embedding model to obtain the word vector of the ultrasound imaging medical report. Then, based on the word vector of the ultrasound imaging medical report and the word vector of each text block in the ultrasound medical knowledge base, the corresponding first target cosine similarity value is calculated. Next, the first target cosine similarity value that is greater than the cosine similarity value threshold is determined as the corresponding second target cosine similarity value. Then, the text blocks corresponding to each second target cosine similarity value are concatenated in sequence to obtain the medical auxiliary diagnosis text of ultrasound imaging examination, which improves the convenience and clinical applicability of generating ultrasound imaging medical reports.

[0134] It should be understood that, although Figures 2-7 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figures 2-7 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0135] Secondly, such as Figure 8 As shown, an ultrasound imaging medical report generation device is provided, which includes an audio file acquisition module 810, an ultrasound imaging medical report template matching module 820, and an ultrasound imaging medical report generation module 830.

[0136] The audio file acquisition module 810 is used to acquire the standard audio file describing the examination record of ultrasound imaging examination, and to obtain the examination record text after recognizing and processing the standard audio file describing the examination record; the ultrasound imaging medical report template matching module 820 is used to acquire the ultrasound examination medical order information, and to obtain the target ultrasound imaging medical report template after performing ultrasound examination medical report template matching processing based on the ultrasound examination medical order information; the ultrasound imaging medical report generation module 830 is used to acquire the target prompt word, and to obtain the ultrasound imaging medical report of ultrasound imaging examination after inputting the examination record text, the target ultrasound imaging medical report template and the target prompt word into the large language model.

[0137] In one embodiment, the audio file acquisition module 810 includes an audio file acquisition unit.

[0138] The audio file acquisition unit is used to download the initial audio files and audio sequence files of each examination record description uploaded by the doctor's user terminal for ultrasound imaging examination in real time. The initial audio files of the examination record description adopt the TS stream file format. The audio sequence file is used to record the playback order of each initial audio file of the examination record description. The audio file acquisition unit is used to generate standard audio files of the examination record description based on each initial audio file and audio sequence file of the examination record description. The standard audio files of the examination record description adopt the WAV file format.

[0139] In one embodiment, the audio file acquisition unit includes an audio processor.

[0140] The audio processor is used to input the initial audio files of each inspection record description into the human voice detection model for audio filtering processing, thereby obtaining the corresponding filtered audio files of the inspection record description. The audio filtering processing includes noise audio filtering and silence audio filtering. The audio processor is used to update each initial audio file of the inspection record description with the corresponding filtered audio file of the initial audio file of the inspection record description. The audio processor is used to perform audio splicing processing based on each initial audio file of the inspection record description and the audio sequence file, thereby obtaining the spliced ​​audio file. The audio processor is used to process the audio format of the spliced ​​audio file to obtain the standard audio file for generating the inspection record description.

[0141] In one embodiment, the audio file acquisition module 810 includes an audio recognition unit.

[0142] The audio recognition unit is used to input the standard audio sub-files describing each inspection record into the optimal automatic speech recognition model for recognition processing to obtain the corresponding inspection record sub-text; the audio recognition unit is used to perform text concatenation processing on the corresponding inspection record sub-texts according to the time sequence of each standard audio sub-file describing the inspection record to obtain the inspection record text.

[0143] In one embodiment, the device further includes an optimal automatic speech recognition model selection module.

[0144] The optimal automatic speech recognition model selection module is used to acquire a general audio dataset and each initial candidate automatic speech recognition model. It then inputs the general audio dataset into each initial candidate automatic speech recognition model to evaluate the word error rate (BER) and obtain the first BER of each initial candidate model. The optimal automatic speech recognition model selection module further selects models based on the BER threshold and the BER of each initial candidate model to obtain corresponding intermediate candidate automatic speech recognition models. Finally, the optimal automatic speech recognition model selection module acquires a general medical audio dataset and inputs it into each intermediate candidate automatic speech recognition model to evaluate the BER. The second-word error rate of each intermediate candidate automatic speech recognition model; the optimal automatic speech recognition model selection module is used to select intermediate candidate automatic speech recognition models based on the second-word error rate threshold and each second-word error rate to obtain the corresponding candidate automatic speech recognition model; the optimal automatic speech recognition model selection module is used to obtain the historical ultrasound examination record description audio dataset, and input the historical ultrasound examination record description audio dataset into each candidate automatic speech recognition model to evaluate the word error rate and obtain the third-word error rate of each candidate automatic speech recognition model; the optimal automatic speech recognition model selection module is used to determine the candidate automatic speech recognition model corresponding to the minimum value among the third-word error rates as the optimal automatic speech recognition model.

[0145] In one embodiment, the ultrasound imaging medical report template matching module 820 includes an ultrasound imaging medical report template matching unit.

[0146] The ultrasound imaging medical report template matching unit is used to match the ultrasound examination category based on the ultrasound examination order information to obtain the target ultrasound examination category; the ultrasound imaging medical report template matching unit is used to query the ultrasound imaging medical report template library based on the target ultrasound examination category to obtain the default ultrasound imaging medical report template; the ultrasound imaging medical report template matching unit is used to confirm the default ultrasound imaging medical report template in response to the template confirmation operation, and determine the default ultrasound imaging medical report template as the target ultrasound imaging medical report template; the ultrasound imaging medical report template matching unit is used to change the default ultrasound imaging medical report template in response to the template change operation, and determine the target ultrasound imaging medical report template based on the result of the template change operation.

[0147] In one embodiment, the device further includes an automatic diagnostic text generation module.

[0148] The auxiliary diagnosis and treatment text automatic generation module is used to input the ultrasound imaging medical report into the vector embedding model to obtain the word vector of the ultrasound imaging medical report; the auxiliary diagnosis and treatment text automatic generation module is used to calculate the corresponding first target cosine similarity value based on the word vector of the ultrasound imaging medical report and the word vector of each text block in the ultrasound medical knowledge base; the first target cosine similarity value is the cosine similarity value between the corresponding word vector and the word vector of the ultrasound imaging medical report; the auxiliary diagnosis and treatment text automatic generation module is used to determine the first target cosine similarity value of each first target cosine similarity value that is greater than the cosine similarity value threshold as the corresponding second target cosine similarity value; the auxiliary diagnosis and treatment text automatic generation module is used to concatenate the text blocks corresponding to each second target cosine similarity value in sequence to obtain the medical auxiliary diagnosis and treatment text of the ultrasound imaging examination.

[0149] Specific limitations regarding the ultrasound imaging medical report generation device can be found in the limitations of the ultrasound imaging medical report generation method described above, and will not be repeated here. Each module in the aforementioned ultrasound imaging medical report generation device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device in hardware form, or stored in the memory of a computer device in software form, so that the processor can call and execute the corresponding operations of each module.

[0150] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 9 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computational and control capabilities. The memory includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores the operating system, computer programs, and the database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The database stores examination record text, target ultrasound imaging medical report templates, and target prompts. The network interface communicates with external terminals via a network connection. When executed by the processor, the computer program implements a method for generating ultrasound imaging medical reports.

[0151] Thirdly, a computer device is provided, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to implement the steps of any of the methods described in the above method embodiments.

[0152] Fourthly, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of any of the methods described in the above method embodiments.

[0153] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0154] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0155] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for generating an ultrasound imaging medical report, the method comprising: Obtain the standard audio file describing the ultrasound imaging examination record, and then process the standard audio file to obtain the examination record text. The ultrasound examination order information is obtained, and the target ultrasound medical report template is obtained by matching the ultrasound examination order information with the ultrasound medical report template. The target prompt word is obtained, and the examination record text, the target ultrasound imaging medical report template, and the target prompt word are input into the large language model to obtain the ultrasound imaging medical report of the ultrasound imaging examination.

2. The method according to claim 1, characterized in that, The standard audio file describing the examination record of the ultrasound imaging examination includes: The system downloads in real time the initial audio files and audio sequence files describing each examination record uploaded by the doctor's user terminal during the ultrasound imaging examination; wherein, the initial audio files describing the examination records are in TS stream file format; and the audio sequence files are used to record the playback order of each initial audio file describing the examination records. The standard audio file for the inspection record description is generated based on the initial audio file for each inspection record description and the audio sequence file; wherein the standard audio file for the inspection record description adopts the WAV file format.

3. The method according to claim 2, characterized in that, The step of generating the standard audio file for the inspection record description based on the initial audio file for each inspection record description and the audio sequence file includes: The initial audio files describing each inspection record are input into the human voice detection model for audio filtering processing to obtain the corresponding filtered audio files describing the inspection records; wherein, the audio filtering processing includes noise audio filtering processing and silence audio filtering processing; Update each of the aforementioned inspection record description initial audio files to the corresponding inspection record description filtered audio file of the initial audio file of the inspection record description; Based on the descriptions of the initial audio file and the audio sequence file in each of the inspection records, audio splicing processing is performed to obtain the spliced ​​audio file; The spliced ​​audio file is processed to obtain the standard audio file for generating the inspection record description.

4. The method according to claim 1, characterized in that, The inspection record description standard audio file includes inspection record description standard audio sub-files of preset duration; the inspection record text obtained after recognizing and processing the inspection record description standard audio file includes: Each of the aforementioned inspection record description standard audio sub-files is input into the optimal automatic speech recognition model for recognition processing to obtain the corresponding inspection record sub-text; The corresponding inspection record sub-texts are concatenated according to the time sequence of the standard audio sub-files describing each inspection record to obtain the inspection record text.

5. The method according to claim 4, characterized in that, The method further includes: A general audio dataset and each initial candidate automatic speech recognition model are obtained. The general audio dataset is then input into each of the initial candidate automatic speech recognition models to evaluate the word error rate, and the first word error rate of each initial candidate automatic speech recognition model is obtained. Based on the first word error rate threshold and the error rate of each first word, the initial candidate automatic speech recognition model is screened to obtain the corresponding intermediate candidate automatic speech recognition model. A general medical audio dataset is obtained, and the general medical audio dataset is input into each of the intermediate candidate automatic speech recognition models to evaluate the word error rate, and then the second word error rate of each of the intermediate candidate automatic speech recognition models is obtained. Based on the second word error rate threshold and the error rate of each second word, the intermediate candidate automatic speech recognition models are screened to obtain the corresponding candidate automatic speech recognition models. A dataset of audio descriptions of historical ultrasound examination records is obtained, and the dataset is then input into each of the candidate automatic speech recognition models to evaluate the word error rate, thereby obtaining the third word error rate of each candidate automatic speech recognition model. The candidate automatic speech recognition model corresponding to the minimum error rate of each of the third words is determined as the optimal automatic speech recognition model.

6. The method according to claim 1, characterized in that, The process of matching the ultrasound imaging medical report template with the ultrasound examination medical order information to obtain the target ultrasound imaging medical report template includes: Based on the ultrasound examination medical order information, the ultrasound examination category is matched to obtain the target ultrasound examination category; Based on the target ultrasound examination category, a default ultrasound imaging medical report template is obtained by searching the ultrasound imaging medical report template library. In response to the template confirmation operation of the default ultrasound imaging medical report template, the default ultrasound imaging medical report template is determined as the target ultrasound imaging medical report template. In response to a template replacement operation on the default ultrasound imaging medical report template, the target ultrasound imaging medical report template is determined based on the result of the template replacement operation.

7. The method according to any one of claims 1 to 6, characterized in that, The method further includes: The ultrasound imaging medical report is input into a vector embedding model to obtain the word vectors of the ultrasound imaging medical report; The first target cosine similarity value is calculated based on the word vectors of the ultrasound imaging medical report and the word vectors of each text block in the ultrasound medical knowledge base; wherein, the first target cosine similarity value is the cosine similarity value between the corresponding word vector and the word vector of the ultrasound imaging medical report. Each of the first target cosine similarity values ​​that is greater than the cosine similarity value threshold is determined as the corresponding second target cosine similarity value; The text blocks corresponding to the second target cosine similarity values ​​are sequentially concatenated to obtain the medical auxiliary diagnosis and treatment text of the ultrasound image examination.

8. A device for generating ultrasound imaging medical reports, characterized in that, The device includes: The audio file acquisition module is used to acquire the standard audio file describing the examination record of ultrasound imaging examination, and to obtain the examination record text after recognizing and processing the standard audio file describing the examination record. The ultrasound imaging medical report template matching module is used to obtain the ultrasound examination medical order information of the ultrasound imaging examination, and to obtain the target ultrasound imaging medical report template after performing ultrasound imaging medical report template matching processing based on the ultrasound examination medical order information. The ultrasound imaging medical report generation module is used to obtain target prompt words, and after inputting the examination record text, the target ultrasound imaging medical report template and the target prompt words into the large language model, the ultrasound imaging medical report of the ultrasound imaging examination is obtained.

9. A computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 7.

Images