Artificial Intelligence Q&A System Based on Radiotherapy Process

Abstract

The present application discloses an artificial intelligence question answering system based on the radiotherapy process, belonging to the field of medical equipment. An artificial intelligence question answering system based on the radiotherapy process includes: a server and a plurality of information terminals, the information terminals are signal-connected to the server, and the information terminals are used for inputting questions and outputting answers corresponding to the questions; the server includes: an input module, a large model module, a word vector conversion module, an answer module, a classification module, a matching module, and an output module. In the technical solution provided by the present application: the patient inputs a question to the server through the information terminal, and the server will automatically match the answer to the question, so as to be able to answer the patient's question in time and guide the patient to complete the radiotherapy work in time.

Description

Technical Field

[0001] This application relates to the technical field of medical systems. Specifically, it relates to an artificial intelligence question-answering system based on the radiotherapy process. Background Art

[0002] Radiotherapy uses high-energy rays (such as X-rays, γ-rays or proton beams) to precisely irradiate the tumor area. When a hospital conducts radiotherapy on a patient, it generally goes through the following steps:

[0003] Position fixation: Ensure that the patient maintains a consistent position during the treatment. CT positioning: Determine the specific location and scope of the tumor through CT scanning. Target volume delineation: Mark the tumor area to be irradiated and the surrounding normal tissues based on the CT images. Radiotherapy plan design: Develop a detailed radiotherapy plan, including the irradiation dose, number of sessions, and method. Review and confirmation of the radiotherapy plan: Evaluate and confirm the treatment plan by a professional team. Radiotherapy plan verification: Ensure the accuracy of the treatment plan through simulation or actual testing. Treatment execution: Implement the treatment according to the designed radiotherapy plan. In addition to the above steps, there are also some payment links that the patient needs to go through during the treatment process. The entire radiotherapy process has a long cycle and involves multiple steps.

[0004] When receiving radiotherapy, patients usually complete various treatment tasks in sequence according to the instruction information provided by the hospital, the information provided by the concierge, and the guidance of medical staff. However, in practice, many patients, due to their concerns about the condition and unfamiliarity with the treatment process, will ask a large number of questions to medical staff. These questions cover many aspects, including specific treatment conditions, postoperative expectations, process arrangements, and payment matters. For example: When setting up the radiotherapy plan, a patient may ask the doctor, "Will this radiotherapy plan kill too many normal cells and thus affect my subsequent quality of life?" After completing the CT positioning, the patient may ask, "What are the next treatment steps? Which stage has the current radiotherapy plan reached?"

[0005] These questions are not only numerous and diverse in type, but also involve the personalized needs and process details of patients. If medical staff fail to answer these questions in a timely manner, it may lead to the patient being unable to complete the treatment smoothly; on the contrary, if all questions need to be answered by medical staff one by one, it will take up a lot of their time. Currently, hospitals usually use some simple question-answering models to reply to patients' questions. However, such systems often have the following deficiencies in practice:

[0006] Lack of timeliness: When the patient's question-asking method or wording does not match the preset database, the system may not be able to accurately find the corresponding answer.

[0007] Lack of context review ability: Existing question-and-answer models usually cannot combine the previous conversation history to provide more accurate answers;

[0008] Lack of semantic understanding ability: Unable to understand the implicit semantics in the question information and only match the corresponding answers by keyword matching, resulting in very low accuracy.

[0009] In summary, there is an urgent need for a medical information interaction system that can answer patients' various information needs in a timely and accurate manner. Such a system not only needs to have an efficient question-and-answer function, but also should be able to adapt to patients' diverse questioning methods and review the previous conversation content when necessary to provide more comprehensive services. Summary of the Invention

[0010] This part of the application is used to briefly introduce the concepts, which will be described in detail in the following detailed implementation part. This part of the application is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to limit the scope of the claimed technical solution.

[0011] As the first aspect of this application, to solve the technical problem of inaccurate answers of the question-and-answer model, this application provides an artificial intelligence question-and-answer system based on the radiotherapy process, including: a server and a number of information terminals, the information terminals are signal-connected to the server, and the information terminals are used to input questions and output answers corresponding to the questions;

[0012] The server includes:

[0013] An input module for inputting questions;

[0014] A large model module, on which a trained language large model is deployed and an initial answer is generated according to the input question;

[0015] A word vector conversion module for converting the initial answer into an answer vector;

[0016] An answer module, on which two or more answer question banks are deployed;

[0017] A classification module for corresponding the answer vector to the answer question bank with the highest similarity based on the similarity between the answer vector and each answer question bank;

[0018] A matching module for matching the answer vector with the corresponding answer question bank and screening out the answer information corresponding to the question from the answer question bank;

[0019] An output module for sending the answer information to the terminal device.

[0020] In the technical solution provided by this application: The patient inputs questions to the server through the information terminal, and the server will automatically match the answers to the questions, so as to be able to answer the patient's questions in a timely manner and guide the patient to complete radiotherapy work in a timely manner. At the same time, in this solution, a trained large language model is first used for processing to generate an initial answer. Because the large language model has strong information processing capabilities and can understand different sentence patterns, different emotional information, and sentences with some typos, the generated initial answer itself is the conventional answer to the patient's corresponding question. Then, after converting the initial answer into an answer vector, it is matched with the answer question bank, and the final answer to the question will be more accurate. In this way, the solution provided by this application has, on the one hand, the good understanding ability of the large language model, and on the other hand, does not require specific training of the large language model, and can better adapt to the work in the hospital at a low training cost.

[0021] Further, the large language model is a locally deployed DeepSeek model.

[0022] The DeepSeek model is a large language model that can be locally deployed with relatively small computing resources. This model can well understand the semantic information of different users, accurately understand the intentions of users from different questions, and obtain accurate initial answers. Therefore, only by providing information related to the initial answer can the needs of users be met.

[0023] When converting the initial answer into a vector representation, because the lengths of the sentences are different, the vector lengths of the final sentences may be inconsistent, resulting in information redundancy and low matching accuracy. For this reason, this application provides the following technical solution:

[0024] Further, the word vector conversion module includes:

[0025] Tokenizer: Divide the initial answer into several words;

[0026] Word vector calculator: Convert each word into a word vector to generate a word vector matrix of all words;

[0027] Word vector pooling layer: Pool the word vector matrix to generate a pooled vector;

[0028] Vector output layer: Normalize the pooled vector to generate an answer vector.

[0029] In the technical solution provided by this application, after generating the word vector matrix, the word vector matrix will be pooled. Therefore, whether it is a long sentence or a short sentence, its dimension can be reasonably reduced, and the accuracy is higher when performing similarity matching subsequently.

[0030] When converting words into word vectors, the conventional approach is to substitute words with numbers, which results in the loss of the position information and semantic characteristics of the words. To address this issue, the present application provides the following technical solutions:

[0031] Further, the calculation formula for the word vector in the word vector calculator is: ; where i represents the index of the word, E i represents the word vector of the i-th word in the initial answer, represents the position vector of the i-th word in the initial answer, represents the frequency vector of the i-th word in the initial answer, represents the part-of-speech vector of the i-th word in the initial answer, λ1 represents the position weight, λ2 represents the frequency weight, λ3 represents the part-of-speech weight, and λ1 + λ2 + λ3 = 1.

[0032] In the technical solutions provided by the present application, the word vector is generated by weighting the three-dimensional information of the position, frequency, and part-of-speech of the word. Thus, the finally obtained word vector can accurately represent the position characteristics and semantic roles of the word in the sentence. Furthermore, when matching the sentence similarity, it can better capture the correlation information between words and improve the accuracy of the similarity matching.

[0033] Currently, the method for generating the word frequency vector usually requires pre-collecting a large-scale sentence sample, performing word segmentation processing, and then counting the occurrence frequency of each word to generate the corresponding label value for each word. In practice, the label values of high-frequency words are often relatively close. Although it can make the labels show a certain correlation, the labels generated in this way lack semantic orientation and are difficult to accurately distinguish the specific meanings of high-frequency words in different contexts. Especially in texts in different fields, the ability to distinguish high-frequency words is even more insufficient. To address the above issues, the present application provides the following technical solutions:

[0034] Further, each answer question bank is preset with a dictionary constructed according to the word frequency;

[0035] The combined vector generated by the i-th word in the initial answer traversing each dictionary is the frequency vector of the i-th word.

[0036] In the technical solutions provided by the present application, each answer question bank is set with a dictionary. When generating the frequency vector, it is necessary to traverse each dictionary. Therefore, the finally generated frequency vector can contain the frequency information of the word in different answer question banks, and thus accurately distinguish the meanings of high-frequency words in different fields.

[0037] The problems of patients generally have high pertinence and directivity. If the problems of patients are directly matched with the initial answers and then with the answers, it will result in a large amount of matching, which is not conducive to quickly generating answers.

[0038] Further, there are more than two deployed answer question banks, including a process answer question bank, a treatment expectation answer question bank, and a nursing advice answer question bank;

[0039] Among them, the process answer question bank stores the answer information of all process precautions and the specific steps to be executed in each stage;

[0040] The treatment expectation answer question bank stores all answer information related to treatment;

[0041] The nursing advice answer question bank stores all answer information related to radiotherapy nursing.

[0042] In the technical solution provided by this application, the possible questions of patients are divided into three categories: one is related to the process, one is related to treatment, and the other is related to nursing. The word usage and sentence construction of the answers to these three categories are different. Therefore, classification matching can improve accuracy, and at the same time, the directional information in the patient's questions can be used to quickly narrow down the scope of the question bank and improve the matching efficiency.

[0043] When the classification module classifies the initial answer vector, if the classification is incorrect, it will lead to the generation of an incorrect answer finally, and such an error cannot be corrected by the subsequent matching layer. Since the key information of the initial answer vector mainly depends on the word frequency feature, it may not be possible to achieve a reasonable question bank allocation only relying on the word frequency information.

[0044] To solve this problem, this application provides the following technical solution: Arrange the words in the dictionary corresponding to the process answer question bank in ascending order of word frequency to generate the first-dimensional axis;

[0045] Arrange the words in the dictionary corresponding to the treatment expectation answer question bank in ascending order of word frequency to generate the second-dimensional axis;

[0046] Arrange the words in the dictionary corresponding to the nursing advice answer question bank in ascending order of word frequency to generate the third-dimensional axis;

[0047] Make the first-dimensional axis, the second-dimensional axis, and the third-dimensional axis perpendicular to each other in space to construct a three-dimensional orthogonal coordinate system;

[0048] Convert the frequency vector of the i-th word into the position of the i-th word in the three-dimensional orthogonal coordinate system.

[0049] In the technical solution provided by this application, the frequency information of each word in the three dictionaries is converted into the three axes of the three-dimensional orthogonal coordinate system, and the frequency vector is converted into the position in the three-dimensional orthogonal coordinate system. Therefore, the word frequency information of this word frequency in the three dictionaries can be clearly indicated in the frequency vector, and thus the accuracy of word frequency allocation can be increased.

[0050] In the answers, there are many common words, and the frequencies of these common words are also very high. Using these common words actually cannot distinguish the question bank corresponding to the initial answer. Therefore, relying solely on the frequency of words appearing in different question banks cannot accurately distinguish the question bank corresponding to the initial answer. For this reason, the present application provides the following technical solutions:

[0051] Take the midpoints of the first-dimensional axis, the second-dimensional axis, and the third-dimensional axis as the origin of the three-dimensional orthogonal coordinate system.

[0052] In the technical solution provided by the present application, taking the midpoints of the first-dimensional axis, the second-dimensional axis, and the third-dimensional axis as the origin of the three-dimensional orthogonal coordinate system, in the space of the spatial index coordinate system, the words in the part close to the origin are more likely to exhibit the characteristics of common words. Therefore, it is possible to further distinguish common words and characteristic words by whether the words are close to the origin, increasing the accuracy of discrimination.

[0053] When patients ask questions about the process, most of them want to obtain the next process plan. The pre-trained large language model can only understand the general meaning of the patients' questions, and the initial answers provided lack regionality and timeliness. The questions directly matched from the process answer question bank also lack timeliness. For this reason, the present application provides the following technical solutions:

[0054] Furthermore, the output module further includes:

[0055] A process progress query unit, used to query the current process progress of the patient and obtain the process nodes of the patient;

[0056] An answer information monitoring unit, used to monitor the answer question bank matched with the answer information;

[0057] An answer information output unit. For the answer information that is the information in the process answer question bank, according to the current process node, screen out the information after the process node from the answer information as the answer information.

[0058] In the technical solution provided by the present application, by obtaining the process nodes of the patient, the timeliness of the answer can be increased, and the problem of the answer being too mechanical can be avoided, so that the patient can accurately understand the current situation.

[0059] Furthermore, for the answer information that is the information in the treatment expectation answer question bank, the answer information output unit adds a contact window for the attending doctor of the patient.

[0060] In the technical solution provided by this application, a contact window for doctors can be provided specifically. When patients ask specific questions related to treatment, on the one hand, some standard answers are provided, and on the other hand, a contact channel for the attending doctor is also provided, so that patients can further inquire about the treatment situation of the doctor. In this way, the number and scope of questions that the attending doctor needs to answer are reduced, and on the other hand, it is also more convenient for patients to understand the changes in the condition and the treatment expectations. BRIEF DESCRIPTION OF THE DRAWINGS

[0061] The drawings forming a part of this application are used to provide a further understanding of this application, making other features, purposes, and advantages of this application more obvious. The schematic drawings of the exemplary embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application.

[0062] In addition, throughout the drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic, and the elements and components are not necessarily drawn to scale.

[0063] In the drawings:

[0064] Figure 1 It is a schematic structural diagram of an artificial intelligence question-and-answer system based on the radiotherapy process.

[0065] Figure 2 It is a schematic structural diagram of the server.

[0066] Figure 3 It is a schematic diagram of a three-dimensional orthogonal coordinate system. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0067] The embodiments of this application will be described in more detail below with reference to the drawings. Although some embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand this application. It should be understood that the drawings and embodiments of this application are only for exemplary purposes and are not used to limit the protection scope of this application.

[0068] In addition, it should also be noted that for the convenience of description, only the parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in this application and the features in the embodiments can be combined with each other.

[0069] This application will be described in detail below with reference to the drawings and in combination with the embodiments.

[0070] Refer to Figure 1, the artificial intelligence Q&A system based on the radiotherapy process includes a server and information terminals. Multiple information terminals are deployed, all of which are communication devices, covering types such as mobile phones, laptops, and computers. The server, as the back-end processing center of the information terminals, can be deployed in the cloud network. The information terminals are connected to the server through a wireless network.

[0071] The information terminals are mainly used for question input and output of corresponding answers. The specific process is as follows: After the user submits a question through the information terminal, the question is transmitted to the server. After the server completes the question processing, it generates a corresponding answer and then transmits the answer back to the information terminal. Finally, the information terminal displays the answer to the user.

[0072] For example, if the user asks, "Do I need to drink yogurt before radiotherapy?" The server will send the generated answer to the information terminal, and the information terminal will output the following text content to the user: "If there is no requirement for fasting, you can drink yogurt." This content is the corresponding answer to this question.

[0073] It can be seen that the artificial intelligence Q&A system based on the radiotherapy process can achieve remote automatic answering of patients' questions, effectively promote the radiotherapy process, alleviate patients' resistance to radiotherapy, and has a psychological counseling function.

[0074] The core of this application is to ensure that the server can accurately generate question answers. Existing technical solutions usually adopt a question-answer library matching mechanism, that is, select the reply with the highest matching degree with the pre-stored answers. However, this matching method has insufficient intelligence, and the matching process is mechanical and rigid, making it difficult to accurately understand the deep meaning of the user's question, resulting in a large deviation between the generated answer and the actual needs of the user.

[0075] If we want to enhance the question parsing ability by constructing a language understanding model to generate answers that meet the user's needs, we need to face the problem of model training costs. In practice, the stronger the semantic parsing ability of the language understanding model, the greater the training investment required. If we want to achieve strong semantic parsing ability in the radiotherapy system, we must conduct model training on a large number of professional materials in the radiotherapy field, which will lead to a significant increase in the cost of model construction and training. To address the above problems, this application proposes the following technical solutions:

[0076] Reference Figure 2 , the server includes: an input module, a large model module, a word vector conversion module, an answer module, a classification module, a matching module, and an output module. Among them, the input module is connected to the large model module, the large model module is connected to the word vector conversion module, the classification module is connected to the word vector conversion module, the classification module is connected to the answer module, the matching module is connected to the classification module, and the output module is connected to the matching module.

[0077] An input module for inputting questions; a large model module deployed with a trained language large model, which generates an initial answer based on the input question; a word vector conversion module that converts the initial answer into an answer vector; an answer module deployed with more than two answer question banks; a classification module that corresponds the answer vector to the answer question bank with the highest similarity based on the similarity between the answer vector and each answer question bank; a matching module that matches the answer vector with the corresponding answer question bank and filters out the answer information for the corresponding question from the answer question bank; an output module that sends the answer information to the terminal device. Among them, the language large model is a locally deployed DeepSeek model.

[0078] The core of this application is to parse the user's question through the pre-trained language large model in the large model module and generate a preliminary answer. This preliminary answer is generated based on a deep understanding of the user's question and has strong pertinence. Compared with the original question, it contains higher-dimensional information supplementation. Using such a preliminary answer for subsequent matching can significantly improve the accuracy of the answer information. This solution only needs to deploy lightweight hardware on the server side and configure pre-trained general large models such as the DeepSeek model or Wenxin Yiyan, without additional training. By comparing the initial answer with the question bank information, the required reply can be generated.

[0079] The input module is connected to the terminal device through a signal and is mainly responsible for receiving questions. The questions referred to here are process inquiries submitted by users through information terminals. For example, after a patient logs in to the specified APP and submits a question, the data is uploaded to the server and received by the input module.

[0080] After receiving the question, the input module transfers it to the large model module, which generates an initial answer based on the pre-trained language large model. Although this large model has not been retrained for a specific domain, it can still understand the core semantics of the question and give a general reply to the question.

[0081] Example: The user asks, "What should be done next after the target area is contoured?" Initial answer: "After the target area is contoured, it is necessary to design beam parameters and optimize the dose distribution through the treatment planning system (TPS) to ensure accurate coverage of the tumor area and protect normal tissues. Then, plan verification and patient position reset are carried out, and finally, treatment is implemented." It can be seen that although the initial answer outlines the subsequent process, it does not contain specific execution details (such as the guidance of the payment department), and the process description is relatively general and does not reflect the hospital's personalized process. However, compared with the original question, the initial answer has provided a more accurate information framework. Using it as a matching benchmark can significantly improve the accuracy of the final answer. At the same time, the solution entrusts the most critical semantic understanding task to the pre-trained large model, which only needs to be deployed without training, greatly reducing the development cost.

[0082] The word vector conversion module is responsible for converting the initial answer into a vector representation to improve classification accuracy and then optimize the subsequent answer matching efficiency. The answer module deploys three independent question banks: the process answer question bank, the treatment expectation answer question bank, and the nursing advice answer question bank. Among them, the process question bank stores the precautions and specific operation steps for each link; the treatment question bank covers all treatment-related information; the nursing question bank contains radiotherapy nursing guidance content. The information crossover between the three question banks is low and the correlation is weak. Classifying and matching according to the type of the initial answer can significantly improve the matching efficiency. Since the computer cannot directly process text, it is necessary to convert the initial answer into a vector form through the word vector conversion module and then perform classification and matching.

[0083] To this end, the present application adopts the following solution to realize the vectorization of the initial answer:

[0084] The word vector conversion module includes:

[0085] Tokenizer: Divide the initial answer into several words.

[0086] Specifically, the initial answer is a piece of text, which is directly regarded as a sentence in this solution. In practice, the length of the initial answer can be controlled within a reasonable range, such as 300 words, by restricting the output length of the language model. After obtaining the initial answer, it is necessary to divide the initialized answer into several words, and the index of the word is i. Dividing the text into words is a common technical means and will not be elaborated here. After dividing the sentence into words, the word vectors are obtained, and then all the word vectors are combined (concatenated) together to form the vector of the whole sentence, that is, the answer vector.

[0087] Word vector calculator: Convert each word into a word vector and generate a word vector matrix for all words.

[0088] The word vector calculator is used to convert each word into a word vector, and then combine all the word vectors together to obtain the word vector matrix H.

[0089] Specifically, the calculation formula of the word vector in the word vector calculator is: ; where i represents the index of the word, E i represents the word vector of the i-th word in the initial answer, represents the position vector of the i-th word in the initial answer, represents the frequency vector of the i-th word in the initial answer, represents the part-of-speech vector of the i-th word in the initial answer, λ1 represents the position weight, λ2 represents the frequency weight, and λ3 represents the part-of-speech weight.

[0090] The generation method is as follows. Replace the position of the i-th word with the corresponding value. For example, if "sugar" is the third word in the sentence, the position vector of "sugar" is "3". Here, the position vector is used to indicate the position of the word in the sentence, thus including the position information of the word and adding more semantic information to the subsequent answer matching process.

[0091] The generation method is as follows. According to the pre-marked content, divide the words into nouns, verbs, adjectives, and pronouns, and the types of parts of speech are preset. In this solution, the words are only divided into the above 5 types.

[0092] The pre-marking is to count the words that appear in all the answers and then assign a specific label to each word. For example, "diabetes" is defined as "noun", "rapid" is defined as "adjective", etc. To avoid the possibility of a single word having multiple parts of speech, in this solution, the part-of-speech frequency of the words in all the answer question banks is used as vector information.

[0093] Specifically: , a1 is the proportion of the i-th word belonging to nouns among the number of times the i-th word appears in all the answer question banks, a2 is the proportion of the i-th word belonging to verbs among the number of times the i-th word appears in all the answer question banks, a3 is the proportion of the i-th word belonging to adjectives among the number of times the i-th word appears in all the answer question banks, a4 is the proportion of the i-th word belonging to pronouns among the number of times the i-th word appears in all the answer question banks. Therefore, the part-of-speech vector basically indicates the general role that the word plays in a sentence.

[0094] Both the position vector and the part-of-speech vector in the above solution are to make the word vectors of the words contain more context information. And for the most crucial information of the word itself, the word needs to be vectorized. In the current solution, the Word2Vec model is used for conversion. The core idea of this model is that words with the same word frequency are semantically similar. Therefore, convert the words into their frequencies in the samples and use the word frequencies to represent the words.

[0095] The accuracy of this solution is not high. In fact, it will confuse the meanings of words with similar frequencies and cannot accurately distinguish the nature of the words. Refer to Figure 3 , for this reason, this application provides the following technical solution: Each answer question bank is preset with a dictionary constructed according to word frequency; the combined vector generated by traversing each dictionary for the i-th word in the initial answer is the frequency vector of the i-th word. The essence of this solution is to generate a frequency vector by traversing the frequencies of the i-th word in each dictionary. This frequency vector contains the position information of the word in 3 question banks.

[0096] In practice, this solution will confuse the frequency information of the words in each question bank. For this reason, this application provides the following technical solution:

[0097] Arrange the words in the dictionary corresponding to the process answer question bank from low to high according to word frequency to generate the first dimension axis;

[0098] Arrange the words in the dictionary corresponding to the treatment expectation answer question bank from low to high according to word frequency to generate the second dimension axis;

[0099] Arrange the words in the dictionary corresponding to the nursing advice answer question bank from low to high according to word frequency to generate the third dimension axis;

[0100] The first dimension axis, the second dimension axis and the third dimension axis are perpendicular to each other in space to construct a three-dimensional orthogonal coordinate system;

[0101] Specifically, after the process answer question bank, treatment expectation answer question bank, and nursing suggestion answer question bank are constructed, the samples therein are collected to establish the corresponding first dimension axis, second dimension axis, and third dimension axis.

[0102] The construction methods of the first dimension axis, the second dimension axis, and the third dimension axis are the same. The following only provides the steps for generating the first dimension axis.

[0103] S1: Get all the texts in the process answer question bank and divide all the texts into words;

[0104] When dividing words, you need to remove punctuation, special symbols, HTML tags, unify capitalization, fonts, etc.

[0105] In addition, it is also necessary to filter out some modal particles and stop words, such as "的" and "是", which have little information content.

[0106] S2: Count the number of times all words appear, calculate the frequency of each word, and generate a word list;

[0107] S3: Arrange the words from low to high frequency to get the first dimension axis. So each word becomes a position on the first dimension axis.

[0108] Thus, according to the above scheme, we can get the first dimension axis, the second dimension axis, and the third dimension axis. Most of the words in these three dimensional axes overlap, but the positions of the words on different axes are different. In order to further characterize the positional relationship of the words on the three axes, the first dimension axis, the second dimension axis, and the third dimension axis are perpendicular to each other in space to construct a three-dimensional orthogonal coordinate system.

[0109] In practice, in order to make the distribution of common words and feature words more regular, in this solution, the midpoints of the first-dimensional axis, the second-dimensional axis, and the third-dimensional axis are used as the origin of the three-dimensional orthogonal coordinate system. In this way, common words will be distributed near the origin of the three-dimensional rectangular coordinate system, while feature words will be close to the corresponding axis endpoints. Here, common words refer to words that frequently appear in all three answer question banks (such as "radiotherapy"), and feature words refer to words that frequently appear only in the corresponding answer question bank (such as "CT positioning" in the process question bank).

[0110] Through the above processing, each word corresponds to a set of coordinate values in the three-dimensional orthogonal coordinate system, and the frequency vector in this solution is defined as the vector representation of these coordinate values.

[0111] ;

[0112] Among them, g represents the unit vector along the positive direction of the first-dimensional axis, y represents the unit vector along the positive direction of the second-dimensional axis, k represents the unit vector along the positive direction of the third-dimensional axis, and (x, y, z) represents the coordinates of the i-th word in the three-dimensional orthogonal coordinate system.

[0113] After obtaining the word vectors E of all words i , all the word vectors are concatenated to obtain a word vector matrix H. The dimension of H is m×d, where d is the dimension of the word vector and m is the length of the initial answer. It can be foreseen that the dimension of the word vector matrix H is too high and needs to be reduced. In this solution, the word vector matrix H is reduced in dimension.

[0114] Word vector pooling layer: Pool the word vector matrix to generate a pooled vector.

[0115] The pooling process of the word vector pooling layer includes the following steps:

[0116] Step 1: Perform principal component analysis on the word vector matrix H to reduce the word vector matrix to different dimensions u to obtain a data matrix Z u .

[0117] Principal component analysis (PCA) is an unsupervised dimensionality reduction method. The core idea is to project the original high-dimensional data onto a low-dimensional space through an orthogonal transformation, retaining the direction of the largest variance (i.e., the main features of the data). The PCA algorithm is a prior art, and the specific implementation method will not be elaborated here. After using principal component analysis on the word vector matrix H, the word vector matrix H can be divided into multiple components, and then the word vector matrix H is reduced in dimension based on the components. Therefore, in practice, the word vector matrix H can be reduced to the required dimension.

[0118] Step 2: Based on the Gaussian distribution assumption, calculate the log-likelihood function InL(u): ; ; ; where n represents the number of samples of the word vectors, b represents the traversal index of the samples, u represents the reduced dimension of the word vector matrix H, and Z u is the data matrix after reducing the word vector matrix H to dimension u, and Z b represents the b-th row of Z u , and β u represents the mean vector of the data matrix Z u . represents the covariance matrix of the data matrix Z u . represents the inverse matrix of the covariance matrix, tr() represents the trace of the matrix, and S u represents an intermediate variable.

[0119] Step 3: Calculate the number of parameters P; ;

[0120] Step 4: Calculate the AIC value;

[0121] AIC(u) = 2P - 2InL(u).

[0122] Step 5: Traverse all candidate dimensions u to find the dimension u that minimizes AIC.

[0123] By using Steps 1 to 4, the AIC values corresponding to each candidate dimension u can be calculated. Therefore, we can calculate the AIC values for all dimensions, and then compare the AIC values for all dimensions, select the dimension with the smallest AIC value from them, and then reduce the word vector matrix to this dimension using principal component analysis.

[0124] In the technical solution provided by this application, when reducing the dimension of the word vector matrix, instead of directly selecting a target dimension for dimension reduction, the best vector dimension is selected according to the AIC value, so that information loss can be avoided while reducing the vector dimension.

[0125] Vector output layer: Normalize the pooled vector to generate an answer vector.

[0126] The answer vector finally output by the vector output layer is the vector representation of the initial answer, and the answer vector implicitly contains the information of each word in the initial answer.

[0127] The classification module will correspond the answer vector to the answer question bank with the highest similarity based on the similarity between the answer vector and each answer question bank. Specifically, the similarity of each answer question bank can convert the text in each answer question bank into 1 average vector using the word vector conversion module; then calculate the similarity between the average vector and the answer vector, and thus screen the corresponding answer question bank according to the similarity level.

[0128] For example, the average vector of the process answer question bank is D1, the average vector of the treatment expectation answer question bank is D2, and the average vector of the nursing advice answer question bank is D3. If the similarity between the answer vector D4 and D1 is the highest, then the answer vector D4 is corresponded to the process answer question bank.

[0129] A matching module that matches the answer vector with the corresponding answer question bank and filters out the answer information for the corresponding question from the answer question bank.

[0130] Both the matching module and the classification module perform question bank selection and answer matching based on similarity. The similarity is the cosine similarity, and the specific calculation method is prior art and will not be elaborated here.

[0131] After matching the corresponding answer information from the answer question bank, it is sent to the corresponding terminal device through the output module. In practice, the current processes of each patient are different, so each patient is to obtain subsequent process information. However, identifying the current process of the patient and then matching the answer information for the subsequent process requires a high answer matching accuracy. For this reason, the present application provides the following technical solutions:

[0132] The output module further includes:

[0133] A process progress query unit for querying the current process progress of the patient and obtaining the process node of the patient;

[0134] An answer information monitoring unit for monitoring the answer question bank for answer information matching;

[0135] An answer information output unit. For the answer information that is the information in the process answer question bank, according to the current process node, the information after the process node is filtered out from the answer information as the answer information.

[0136] In this solution, when dealing with process problems, the complete process information will be summarized first. For example: The complete process is "Position fixation (making a mold in the mold room) → CT positioning (with the mold) → Doctor delineates the treatment target area → Physicist designs the radiotherapy plan → Doctor reviews and confirms the plan → Plan verification: 〔① Doctor takes the patient for pre-radiotherapy positioning (in the CT room or in the two-dimensional simulation positioning room); ② Physicist verifies the radiotherapy plan〕 → Treatment."

[0137] If the answer information monitoring unit monitors that the answer information comes from the process answer question bank, it will query the patient's current process progress to obtain the patient's process node. For example, if the patient's current progress is "CT positioning", the answer information will be output as "The doctor delineates the treatment target area → The physicist designs the radiotherapy plan → The doctor reviews and confirms the plan → Plan verification: 〔① The doctor takes the patient for pre-radiotherapy positioning (in the CT room or in the two-dimensional simulation positioning room); ② The physicist verifies the radiotherapy plan〕→ Treatment." In this way, the difficulty of identifying process-related problems is reduced through this method.

[0138] Furthermore, for the answer information output unit, when the answer information is from the treatment expectation answer question bank, a contact window of the patient's attending doctor is added to the answer information. That is, when the answer information monitoring unit monitors that the patient is asking questions about treatment expectations, it further retrieves the contact window of the attending doctor, enabling the attending doctor to further explain to the patient, which plays a role in calming the patient.

[0139] The above description is only some preferred embodiments of the present application and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present application is not limited to the technical solutions formed by the specific combination of the above technical features, and should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the (but not limited to) technical features with similar functions disclosed in the embodiments of the present application.

Claims

1. An artificial intelligence question - answering system based on the radiotherapy process, comprising: A server and several information terminals, where the information terminals are signal - connected to the server, and the information terminals are used to input questions and output answers corresponding to the questions; It is characterized in that The server includes: An input module, used to input questions; A large - model module, on which a trained language large - model is deployed, and an initial answer is generated according to the input question; A word - vector conversion module, which converts the initial answer into an answer vector; An answer module, on which more than two answer question banks are deployed; A classification module, which corresponds the answer vector to the answer question bank with the highest similarity based on the similarity between the answer vector and each answer question bank; A matching module, which matches the answer vector with the corresponding answer question bank and screens out the answer information corresponding to the question from the answer question bank; An output module, which sends the answer information to the terminal device; The word - vector conversion module includes: A tokenizer, which divides the initial answer into several words; A word - vector calculator, which converts each word into a word vector and generates a word - vector matrix of all words; A word - vector pooling layer, which pools the word - vector matrix to generate a pooled vector; A vector output layer, which normalizes the pooled vector to generate an answer vector; The calculation formula of word vectors in the word vector calculator is as follows: where i represents the index of the word, E i represents the word vector of the i-th word in the initial answer, represents the position vector of the i-th word in the initial answer, represents the frequency vector of the i-th word in the initial answer, represents the part-of-speech vector of the i-th word in the initial answer, λ1 represents the position weight, λ2 represents the frequency weight, λ3 represents the part-of-speech weight, and λ1 + λ2 + λ3 = 1; a dictionary constructed according to word frequency is preset for each answer question bank; The combined vector generated by the i - th word in the initial answer traversing each dictionary is the frequency vector of the i - th word; The deployment of more than two answer question banks includes a process answer question bank, a treatment - expectation answer question bank, and a nursing - advice answer question bank; Arrange the words in the dictionary corresponding to the process answer question bank in ascending order of word frequency to generate the first - dimension axis; Arrange the words in the dictionary corresponding to the treatment - expectation answer question bank in ascending order of word frequency to generate the second - dimension axis; Arrange the words in the dictionary corresponding to the nursing - advice answer question bank in ascending order of word frequency to generate the third - dimension axis; Make the first - dimension axis, the second - dimension axis, and the third - dimension axis perpendicular to each other in space to construct a three - dimensional orthogonal coordinate system; Convert the frequency vector of the i - th word into the position of the i - th word in the three - dimensional orthogonal coordinate system; Take the mid - points of the first - dimension axis, the second - dimension axis, and the third - dimension axis as the origin of the three - dimensional orthogonal coordinate system; The generation steps of the first - dimension axis, the second - dimension axis, and the third - dimension axis are as follows: S1: Obtain all the texts in the process answer question bank and divide all the texts into words; S2: Count the number of times each word appears, calculate the frequency of each word, and generate a word list; S3: Arrange the words in ascending order of frequency to obtain the first - dimension axis; So each word becomes a position on the first - dimension axis; The frequency vector is defined as the vector representation of the coordinate value: where g represents the unit vector along the positive direction of the first - dimensional axis, y represents the unit vector along the positive direction of the second - dimensional axis, k represents the unit vector along the positive direction of the third - dimensional axis, and (x, y, z) represents the coordinates of the i - th word in the three - dimensional orthogonal coordinate system; By concatenating all the word vectors, a word - vector matrix H can be obtained; The pooling process includes the following steps: Step 1: Perform principal component analysis on the word - vector matrix H to reduce the word - vector matrix to different dimensions u, obtaining a data matrix Zu; Step 2: Based on the Gaussian - distribution assumption, calculate the log - likelihood function InL(u): ; ; ; Among them, n represents the number of samples of the word vectors, b represents the traversal index of the samples, u represents the reduced dimension of the word vector matrix H, Zu is the data matrix after the word vector matrix H is reduced to the u dimension, Zb represents the b-th row of Zu, and βu represents the mean vector of the data matrix Zu. represents the covariance matrix of the data matrix Zu. represents the inverse matrix of the covariance matrix, tr() represents the trace of the matrix, and Su represents the intermediate variable. Step 3: Calculate the number of parameters P; Step 4: Calculate the AIC value; ; Step 5: Traverse all candidate dimensions u to find the dimension u that minimizes the AIC.

2. The artificial intelligence question-answering system based on the radiotherapy process according to claim 1, characterized in that: The language large - model is a locally - deployed DeepSeek model.

3. The artificial intelligence question - answering system based on the radiotherapy process according to claim 1, characterized in that: The process answer question bank stores all the process precautions and the answer information of the specific steps to be executed in each stage; The treatment - expectation answer question bank stores all the answer information related to treatment; The nursing - advice answer question bank stores all the answer information related to radiotherapy nursing.

4. The artificial intelligence question-answering system based on the radiotherapy process according to any one of claims 1 to 3, characterized in that: The output module further includes: A process - progress query unit, used to query the current process progress of the patient and obtain the process nodes of the patient; Answer information monitoring unit, which is used to monitor the answer question bank that matches the answer information; Answer information output unit. For the answer information that is the information in the process answer question bank, according to the current process node, the information after the process node is screened out from the answer information as the answer information.

5. The artificial intelligence question-answering system based on the radiotherapy process according to claim 4, wherein: Answer information output unit. For the answer information that is the information in the treatment expectation answer question bank, a contact window of the patient's attending doctor is added to the answer information.

Images