Cancer pain level assessment method, system, device and medium based on personalized representation space

Abstract

The application belongs to the technical field of cancer pain evaluation, and discloses a cancer pain grade evaluation method, system, device and medium based on a personalized representation space, which comprises the following steps: constructing a pre-training large language model, and obtaining video tokens and physiological tokens by using a linear projection unit; constructing patient-specific instruction pairs according to patient-specific data, and performing personalized parameter fine-tuning and structure selection on the pre-training large language model by using QLoRA technology; obtaining a personalized high-dimensional cancer pain representation space according to a personalized embedding vector; calculating the relative pain index of the patient according to the personalized pain data and the personalized high-dimensional cancer pain representation space; and updating the adapter of the pre-training large language model and the relative pain index according to the latest steady physiological data. The parameter isolation fine-tuning of 'one person one model' is realized, the model is no longer dominated by the group data distribution, and the unique pain expression mode of special tumor patients such as the elderly, complex underlying diseases and facial muscle stiffness can be accurately captured.

Description

[0001] This invention relates to the field of cancer pain assessment technology, and in particular to a method, system, device and medium for assessing cancer pain levels based on a personalized representation space. Background Technology

[0002] Cancer pain is the most common accompanying symptom for cancer patients during disease progression, treatment, and palliative care. Effective assessment and standardized management of pain directly affect patients' quality of life, treatment adherence, psychological state, and overall prognosis. According to the World Health Organization, tens of millions of cancer patients are diagnosed globally each year, and more than 70% of these patients with advanced-stage cancer experience varying degrees of cancer pain. Nearly half of these patients experience moderate to severe or even excruciating cancer pain, and some even experience recurrent, sudden bursts of pain. In my country, with the improvement of cancer diagnosis and treatment, the refinement of palliative care systems, and the comprehensive advancement of standardized cancer pain treatment, the clinical requirements for the accuracy, real-time nature, and continuity of cancer pain assessment are continuously increasing. Traditional assessment models are no longer adequate to meet the evolving needs of refined analgesia and full-cycle pain management in oncology.

[0003] From the perspective of the overall development of the industry, intelligent pain assessment technology (including cancer pain) has undergone three generations of iterations: the first generation is the subjective scale assessment technology (such as the NRS numerical rating scale) that has been used in clinical practice for many years; the second generation is the single-modal physiological signal objective monitoring technology based on biosensors and traditional machine learning; and the third generation is the general multimodal fusion assessment technology that combines deep learning algorithms. These three generations of technology have gradually upgraded from subjective human judgment to objective data collection and from single information dimension to multi-source data fusion. However, due to the inherent defects of static model architecture, the existing mainstream solutions often adopt a "one-size-fits-all" general model when facing a large group of cancer patients with huge individual differences. This ignores the differentiated modeling of prior information such as patient age, disease type, and basic physical signs, resulting in poor model generalization ability and high misjudgment rate. Summary of the Invention

[0004] The purpose of this application is to propose a method, system, device, and medium for assessing cancer pain levels based on a personalized representation space to solve the problems of inability to perform personalized adaptation, high misjudgment rate, and poor generalization ability in the prior art.

[0005] Firstly, a method for assessing cancer pain levels based on a personalized representation space includes:

[0006] Step 100: Construct a pre-trained large language model, and set a learnable linear projection unit in the pre-trained large language model. Collect the patient's facial video stream and skin conductance response signal, encode and map the facial video stream and skin conductance response signal respectively, and use the linear projection unit to obtain video token and physiological token.

[0007] Step 200: Obtain patient-specific data, construct patient-specific instruction pairs based on the patient-specific data, and use QLoRA technology to fine-tune the personalized parameters and select the structure of the pre-trained large language model to obtain personalized embedding vectors. Obtain personalized high-dimensional cancer pain representation space based on the personalized embedding vectors.

[0008] Step 300: In the early stage of hospitalization, collect the patient's individual pain data, and calculate the patient's relative pain index based on the individual pain data and the individual high-dimensional cancer pain representation space;

[0009] Step 400: Obtain the latest homeostatic physiological data of the patient's condition at the new stage in real time, and update the adapter of the pre-trained large language model and the relative pain index based on the latest homeostatic physiological data.

[0010] Furthermore, step 100 includes:

[0011] Step 110: Encode the frame sequence of the facial video stream using the video coding unit of a pre-trained large language model to obtain a high-dimensional feature vector of the video.

[0012] Step 120: Using the linear projection unit, map the high-dimensional feature vector of the video to the hidden unit dimension to obtain the video token;

[0013] Step 130: The skin conductance response signal is segmented and its features are extracted using the physiological coding unit of the pre-trained large language model to obtain a physiological feature vector.

[0014] Step 140: Map the physiological feature vector to the hidden unit dimension using the linear projection unit to obtain the physiological token.

[0015] Furthermore, step 200 includes:

[0016] Step 210: Obtain the patient-specific data, and use the patient-specific data to train the pre-trained large language model to obtain the first personalized LoRA adapter;

[0017] Step 220: Connect the first personalized LoRA adapter in parallel next to the attention unit of the pre-trained large language model, and freeze the parameters of the pre-trained large language model to fine-tune the personalized parameters.

[0018] Furthermore, after step 220, step 200 also includes:

[0019] Step 230: Use a multimodal fusion hidden output unit as the output unit of the pre-trained large language model;

[0020] Step 240: The output of the multimodal fusion hidden output unit is subjected to attention-weighted pooling to obtain a personalized embedding vector. The set of the personalized embedding vectors is used to obtain the personalized high-dimensional cancer pain representation space.

[0021] Furthermore, step 300 includes:

[0022] Step 310: Obtain the patient's individual pain data during the initial stage of hospitalization, and determine the pain-free anchor point and risk anchor point in the individual high-dimensional cancer pain representation space using K-means based on the individual pain data;

[0023] Step 320: Real-time detection of patient-specific pain data to obtain real-time personalized pain data, and mapping the real-time personalized pain data into a real-time personalized pain vector;

[0024] Step 330: Calculate the relative pain index based on the pain-free anchor point, risk anchor point, and individual pain vector.

[0025] Furthermore, step 400 includes:

[0026] Step 410: Collect and obtain the latest homeostatic physiological data according to the latest clinical protocol;

[0027] Step 420: Update the pain-free anchor point and risk anchor point using an exponential moving average based on the latest steady-state physiological data to update the relative pain index.

[0028] Furthermore, step 400 also includes:

[0029] Step 430: If the relative pain index changes, start parameter fine-tuning and train to obtain a second personalized LoRA adapter based on the latest homeostatic physiological data;

[0030] Step 440: Perform weighted interpolation on the first personalized LoRA adapter based on the second personalized LoRA adapter to fine-tune the parameters.

[0031] To address the aforementioned technical problems, embodiments of this application also provide a cancer pain level assessment system based on a personalized representation space, employing the cancer pain level assessment method based on a personalized representation space as described in the first aspect, including:

[0032] The first acquisition module is used to pre-train a large language model and set a learnable linear projection unit in the pre-trained large language model. The encoding unit and mapping unit in the pre-trained large language model are used to encode and map the facial video stream and skin electroreactivity signal, and the linear projection unit is used to obtain the video token and physiological token.

[0033] The second acquisition module is used to acquire patient-specific data, construct patient-specific instruction pairs based on the patient-specific data, and use QLoRA technology to fine-tune the individual parameters and select the structure of the pre-trained large language model to obtain a personalized embedding vector. Based on the personalized embedding vector, a personalized high-dimensional cancer pain representation space is obtained.

[0034] The calculation module is used to collect and acquire the patient's individual pain data in the early stage of hospitalization, and calculate the patient's relative pain index based on the individual pain data and the individual high-dimensional cancer pain representation space.

[0035] The update module is used to acquire the latest homeostatic physiological data of the patient's condition at a new stage in real time, and update the adapter of the pre-trained large language model and the relative pain index based on the latest homeostatic physiological data.

[0036] To address the aforementioned technical problems, embodiments of this application also provide a computer device, including a memory and a processor. The memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the cancer pain level assessment method based on personalized representation space as described in the first aspect.

[0037] To address the aforementioned technical problems, embodiments of this application also provide a computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the cancer pain level assessment method based on a personalized representation space as described in the first aspect.

[0038] Compared with existing technologies, the embodiments of this application have the following main technical effects: By constructing exclusive instruction pairs and using QLoRA technology to fine-tune the parameters of the pre-trained large language model, personalized embedding vectors are obtained, achieving isolated parameter fine-tuning for each patient. The model is no longer dominated by the distribution of group data and can accurately capture the unique pain expression patterns of special tumor patients such as the elderly, those with complex underlying diseases, and those with facial muscle stiffness. By utilizing personalized pain data and personalized high-dimensional cancer pain representation space to calculate the relative pain index of patients, the clinical problem of different pain levels despite the same NRS score in traditional methods is avoided, enabling individualized longitudinal comparison. By fine-tuning and updating the parameters of the adapter of the pre-trained large language model based on the latest steady-state physiological data, model optimization is achieved. Attached Figure Description

[0039] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0040] Figure 1 This is a flowchart of an embodiment of a cancer pain level assessment method based on a personalized representation space according to this application;

[0041] Figure 2 yes Figure 1 A flowchart of one specific implementation of step 100;

[0042] Figure 3 yes Figure 1 A flowchart of a specific implementation of step 200;

[0043] Figure 4 yes Figure 3 A flowchart of a specific implementation method following step 220;

[0044] Figure 5 yes Figure 1 A flowchart of a specific implementation method in step 300;

[0045] Figure 6 yes Figure 1 A flowchart of a specific implementation of step 400;

[0046] Figure 7 yes Figure 6 A flowchart of a specific implementation method following step 420;

[0047] Figure 8 This is a schematic diagram of the pre-trained large language model structure of this application;

[0048] Figure 9 This is a schematic diagram of an embodiment of a cancer pain level assessment system based on a personalized representation space according to this application;

[0049] Figure 10 This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation

[0050] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of this invention, and not all of them. Other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are all within the scope of protection of this invention.

[0051] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0052] 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 pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings of this application are intended to cover non-exclusive inclusion.

[0053] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0054] When faced with a large number of cancer patients with significant individual differences, current mainstream solutions often adopt a "one-size-fits-all" general model, ignoring the differentiated modeling based on prior information such as patient age, disease type, and basic physical signs. This results in poor model generalization ability and a high misjudgment rate.

[0055] The purpose of this application is to propose a method, system, device, and medium for assessing cancer pain levels based on a personalized representation space to solve the problems of inability to perform personalized adaptation, high misjudgment rate, and poor generalization ability in the prior art.

[0056] Firstly, a cancer pain level assessment method based on a personalized representation space, such as... Figure 1 , Figure 1 This is a flowchart of an embodiment of a cancer pain level assessment method based on a personalized representation space according to this application; including:

[0057] Step 100: Construct a pre-trained Large Language Model (LLM), such as... Figure 8 , Figure 8This is a schematic diagram of the pre-trained large language model structure of this application. Learnable linear projection units are set in the pre-trained large language model to collect facial video streams and skin conductance response signals of patients. The facial video streams and skin conductance response signals are encoded and mapped, and video tokens and physiological tokens are obtained using linear projection units.

[0058] In one possible implementation, such as Figure 2 , Figure 2 yes Figure 1 A flowchart of a specific implementation of step 100; step 100 includes:

[0059] Step 110: Encode the frame sequence of the facial video stream using the video coding unit of the pre-trained large language model to obtain the high-dimensional feature vector of the video. Step 120: Map the high-dimensional feature vector of the video to the hidden unit dimension using the linear projection unit to obtain the video token. Step 130: Segment the skin conductance response signal using the physiological coding unit of the pre-trained large language model and extract its features to obtain the physiological feature vector. Step 140: Map the physiological feature vector to the hidden unit dimension using the linear projection unit to obtain the physiological token.

[0060] In this embodiment, the patient's facial video stream and skin conductance response signals are simultaneously acquired using an input acquisition unit. Then, a visual encoding unit and a physiological encoding unit encode the patient's facial video stream and skin conductance response signals, respectively. The visual encoding unit can employ a pre-trained Vision Transformer (VTR). Encode the facial video frame sequence and extract the high-dimensional feature vector of the video. The physiological coding unit segments the skin conductance response signal (EDA signal) and extracts physiological feature vectors. The mapping unit uses a learned linear projection layer (Projector) to respectively... and The mapped hidden dimension generates visual tokens and physiological tokens.

[0061] Step 200: Obtain patient-specific data, construct patient-specific instruction pairs based on the patient-specific data, and use QLoRA technology to fine-tune the personalized parameters and select the structure of the pre-trained large language model to obtain personalized embedding vectors. Based on the personalized embedding vectors, obtain the personalized high-dimensional cancer pain representation space.

[0062] In this embodiment, a patient-specific instruction pair is constructed according to the "one person, one dataset" mechanism to achieve data isolation and instruction construction. Data from different patients are not mixed for training. An independent instruction pair is constructed for each patient, including the patient ID, disease stage, and the corresponding token sequence.

[0063] In one possible implementation, such as Figure 3 , Figure 3 yes Figure 1 A flowchart of a specific implementation of step 200; step 200 includes:

[0064] Step 210: Obtain patient-specific data and use the patient-specific data to train the pre-trained large language model to obtain the first personalized LoRA adapter; Step 220: Connect the first personalized LoRA adapter in parallel next to the attention unit of the pre-trained large language model and freeze the parameters of the pre-trained large language model to perform personalized parameter fine-tuning.

[0065] In this embodiment, the QLoRA (Quantized Low-Rank Adaptation) technique specifically involves fine-tuning the parameters through a parallel adapter. The QLoRA technique freezes more than 99% of the parameters of the base model. A first personalized LoRA adapter (low-rank matrix) is connected in parallel next to the attention unit. By training the first personalized LoRA adapter, the patient's unique "physiological signal-subjective feeling" mapping function is fitted, rather than learning the commonalities of the group.

[0066] In one possible implementation, after step 220, as Figure 4 , Figure 4 yes Figure 3 A flowchart of a specific implementation method following step 220; step 200 also includes:

[0067] Step 230: Use the multimodal fusion hidden output unit as the output unit of the pre-trained large language model; Step 240: Perform attention-weighted pooling on the output of the multimodal fusion hidden output unit to obtain personalized embedding vectors, and use the set of personalized embedding vectors to obtain a personalized high-dimensional cancer pain representation space.

[0068] In this embodiment, the output layer (LMHead) of the model is stripped away, and the hidden output unit of the multimodal fusion is selected as the hidden layer output for multimodal fusion, serving as the feature extraction layer. The output of the multimodal fusion hidden output unit is then weighted and pooled to obtain a feature extraction layer with dimension [missing information]. Personalized embedding vectors .all The set of values ​​constitutes the patient's personalized high-dimensional cancer pain representation space.

[0069] Step 300: In the early stage of hospitalization, collect the patient's individual pain data and calculate the patient's relative pain index based on the individual pain data and the individual high-dimensional cancer pain representation space.

[0070] In one possible implementation, such as Figure 5 , Figure 5 yes Figure 1 A flowchart of a specific implementation of step 300; step 300 includes:

[0071] Step 310: Obtain the patient's individual pain data during the initial stage of hospitalization, and determine the pain-free anchor point and risk anchor point in the individual high-dimensional cancer pain representation space using K-means based on the individual pain data; Step 320: Perform real-time detection on the patient's individual pain data to obtain real-time individual pain data, and map the real-time individual pain data into a real-time individual pain vector; Step 330: Calculate the relative pain index based on the pain-free anchor point, risk anchor point, and individual pain vector.

[0072] In this embodiment, a vector geometric matching mechanism is used to achieve "comparison with itself".

[0073] First, individualized anchor points are set. In the early stages of patient admission, using collected individual pain data, such as "no pain" and "breakthrough pain" data, a pain-free anchor point is determined in the individualized high-dimensional cancer pain representation space using K-means. and risk anchors .

[0074] Then, the Relative Pain Index (RPI) is calculated, mapping real-time personalized pain data to real-time personalized pain vectors. .calculate The normalized distance from the anchor point, i.e., the relative pain index (RPI):

[0075] = (1),

[0076] Among them, RPI This index only reflects the degree of deviation of a patient from their baseline, thus resolving the issue of inconsistent perceptions of the same score among different patients.

[0077] In one implementation, it is also possible to... Perform confidence level determination, if the real-time vector If the data falls into a low-density area of ​​historical data distribution (i.e., an unknown state), the system automatically lowers the RPI confidence level and prompts medical staff to conduct manual verification to avoid misjudgment caused by model illusion.

[0078] Step 400: Obtain the latest steady-state physiological data of the patient's condition at the new stage in real time, and update the adapter and relative pain index of the pre-trained large language model based on the latest steady-state physiological data.

[0079] In one possible implementation, such as Figure 6 , Figure 6 yes Figure 1 A flowchart of a specific implementation of step 400; step 400 includes:

[0080] Step 410: Collect the latest homeostatic physiological data according to the latest clinical protocol; Step 420: Update the pain-free anchor point and risk anchor point using an exponential moving average method based on the latest homeostatic physiological data to update the relative pain index.

[0081] In this embodiment, when the analgesia regimen is adjusted clinically and the patient is confirmed to have entered a new steady state, anchor recalibration is required. The system automatically collects the latest steady-state physiological data for that period and updates it using an exponential moving average. and The coordinate position allows the assessment benchmark to follow changes in patient tolerance and update the relative pain index.

[0082] In one possible implementation, such as Figure 7 , Figure 7 yes Figure 6 A flowchart of a specific implementation following step 420; step 400 further includes:

[0083] Step 430: If the relative pain index changes, start parameter fine-tuning and train a second personalized LoRA adapter based on the latest steady-state physiological data; Step 440: Perform weight interpolation on the first personalized LoRA adapter based on the second personalized LoRA adapter to fine-tune the parameters.

[0084] In this embodiment, if the relative pain index changes, such as a significant shift in the patient's pain pattern (e.g., the development of drug tolerance), the system initiates incremental fine-tuning. A new LoRA adapter (Adapter-v2), i.e., the second personalized LoRA adapter, is trained using only the latest limited data, and weight interpolation is performed between it and the original adapter (Adapter-v1), i.e., the first personalized LoRA adapter.

[0085] (2),

[0086] in Forgetting factor, For the first personalized LoRA adapter weight, For the second personalized LoRA adapter weight.

[0087] In this embodiment, by constructing exclusive instruction pairs and employing QLoRA technology to fine-tune the parameters of the pre-trained large language model, personalized embedding vectors are obtained, achieving isolated parameter fine-tuning for each individual. The model is no longer dominated by the distribution of group data and can accurately capture the unique pain expression patterns of special cancer patients, such as the elderly, those with complex underlying diseases, and those with facial muscle stiffness. By utilizing personalized pain data and a personalized high-dimensional cancer pain representation space to calculate the relative pain index of patients, the clinical problem of different pain levels despite similar NRS scores is avoided, enabling individualized longitudinal comparisons. Model optimization is achieved by fine-tuning and updating the adapter parameters of the pre-trained large language model based on the latest steady-state physiological data.

[0088] To address the aforementioned technical problems, this application also provides a cancer pain level assessment system 500 based on a personalized representation space, employing the cancer pain level assessment method based on a personalized representation space from the first aspect, such as... Figure 9 , Figure 9 This is a schematic diagram of an embodiment of a cancer pain level assessment system based on a personalized representation space, as described in this application; including:

[0089] The first acquisition module 501 is used to pre-train a large language model and set learnable linear projection units in the pre-trained large language model. It uses the encoding units and mapping units in the pre-trained large language model to encode and map the facial video stream and skin electroreactivity signals, and uses the linear projection units to obtain video tokens and physiological tokens.

[0090] The second acquisition module 502 is used to acquire patient-specific data, construct patient-specific instruction pairs based on the patient-specific data, and use QLoRA technology to fine-tune the individual parameters and select the structure of the pre-trained large language model to obtain a personalized embedding vector. Based on the personalized embedding vector, a personalized high-dimensional cancer pain representation space is obtained.

[0091] The calculation module 503 is used to collect the patient's individual pain data in the early stage of hospitalization and calculate the patient's relative pain index based on the individual pain data and the individual high-dimensional cancer pain representation space.

[0092] The update module 504 is used to acquire the latest steady-state physiological data of the patient's condition in real time, and update the adapter and relative pain index of the pre-trained large language model based on the latest steady-state physiological data.

[0093] To address the aforementioned technical problems, this application also provides a computer device, including a memory and a processor. The memory stores computer-readable instructions, and when the processor executes the computer-readable instructions, it implements the steps of the cancer pain level assessment method based on personalized representation space as described in the first aspect.

[0094] This computer device can be a computer, server, workstation, or other similar device; it can also be a mobile device such as a mobile phone, tablet, or in-vehicle mobile terminal; or other devices with program execution capabilities. Its internal structure diagram can be as follows: Figure 10 As shown, Figure 10 This is a schematic diagram of a computer device according to an embodiment of the present application. The computer device includes a processor, a memory, and a network module. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores an operating system, instructions, or code. The internal memory provides an environment for the operation of the operating system and instructions or code in the non-volatile storage media. When the instructions or code are executed by the processor, they implement the functions or steps of a cancer pain level assessment method based on a personalized representation space. The network module of the computer device may include a network interface and / or a wireless network module, through which the computer device can communicate with other devices or service platforms. Furthermore, the computer device may also include a display screen and input devices, etc.

[0095] The memory is used to store computer programs, which include program instructions. The processor is configured to call the program instructions, and when the processor executes the instructions or code, it implements the steps of the cancer pain level assessment method based on the personalized representation space as described above.

[0096] To address the aforementioned technical problems, embodiments of this application also provide a computer-readable storage medium storing computer-readable instructions. When executed by a processor, the computer-readable instructions implement the steps of the cancer pain level assessment method based on a personalized representation space as described in the first aspect.

[0097] The computer-readable storage medium stores a computer program, which includes program instructions that are implemented when executed by a processor. Figures 1 to 7 The cancer pain level assessment methods based on personalized representation space provided in each step are detailed in the implementation methods provided in the above steps, and will not be repeated here.

[0098] The aforementioned computer-readable storage medium may be the apparatus of the cancer pain level assessment method based on personalized representation space provided in any of the foregoing embodiments, or the internal storage unit of the aforementioned terminal device, such as the hard disk or memory of a computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., provided on the computer device.

[0099] Furthermore, the computer-readable storage medium may include both internal storage units and external storage devices of the computer device. The computer-readable storage medium is used to store the computer program and other programs and data required by the computer device. The computer-readable storage medium may also be used to temporarily store data that has been output or will be output.

[0100] However, it should be understood that it is not required to implement all the components shown; more or fewer components may be implemented instead. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions, and its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.

[0101] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.

[0102] Compared with existing technologies, the embodiments of this application have the following main technical effects: By constructing exclusive instruction pairs and using QLoRA technology to fine-tune the parameters of the pre-trained large language model, personalized embedding vectors are obtained, achieving isolated parameter fine-tuning for each patient. The model is no longer dominated by the distribution of group data and can accurately capture the unique pain expression patterns of special tumor patients such as the elderly, those with complex underlying diseases, and those with facial muscle stiffness. By utilizing personalized pain data and personalized high-dimensional cancer pain representation space to calculate the relative pain index of patients, the clinical problem of different pain levels despite the same NRS score in traditional methods is avoided, enabling individualized longitudinal comparison. By fine-tuning and updating the parameters of the adapter of the pre-trained large language model based on the latest steady-state physiological data, model optimization is achieved.

[0103] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for assessing cancer pain levels based on a personalized representation space, characterized in that, include: Step 100: Construct a pre-trained large language model, and set a learnable linear projection unit in the pre-trained large language model. Collect the patient's facial video stream and skin conductance response signal, encode and map the facial video stream and skin conductance response signal respectively, and use the linear projection unit to obtain video token and physiological token. Step 200: Obtain patient-specific data, construct patient-specific instruction pairs based on the patient-specific data, and use QLoRA technology to fine-tune the personalized parameters and select the structure of the pre-trained large language model to obtain personalized embedding vectors. Obtain personalized high-dimensional cancer pain representation space based on the personalized embedding vectors. Step 300: In the early stage of hospitalization, collect the patient's individual pain data, and calculate the patient's relative pain index based on the individual pain data and the individual high-dimensional cancer pain representation space; Step 400: Obtain the latest homeostatic physiological data of the patient's condition at the new stage in real time, and update the adapter of the pre-trained large language model and the relative pain index based on the latest homeostatic physiological data.

2. The cancer pain level assessment method based on personalized representation space according to claim 1, characterized in that, Step 100 includes: Step 110: Encode the frame sequence of the facial video stream using the video coding unit of a pre-trained large language model to obtain a high-dimensional feature vector of the video. Step 120: Using the linear projection unit, map the high-dimensional feature vector of the video to the hidden unit dimension to obtain the video token; Step 130: The skin conductance response signal is segmented and its features are extracted using the physiological coding unit of the pre-trained large language model to obtain a physiological feature vector. Step 140: Map the physiological feature vector to the hidden unit dimension using the linear projection unit to obtain the physiological token.

3. The cancer pain level assessment method based on personalized representation space according to claim 1, characterized in that, Step 200 includes: Step 210: Obtain the patient-specific data, and use the patient-specific data to train the pre-trained large language model to obtain the first personalized LoRA adapter; Step 220: Connect the first personalized LoRA adapter in parallel next to the attention unit of the pre-trained large language model, and freeze the parameters of the pre-trained large language model to fine-tune the personalized parameters.

4. The cancer pain level assessment method based on personalized representation space according to claim 3, characterized in that, Following step 220, step 200 further includes: Step 230: Use a multimodal fusion hidden output unit as the output unit of the pre-trained large language model; Step 240: The output of the multimodal fusion hidden output unit is subjected to attention-weighted pooling to obtain a personalized embedding vector. The set of the personalized embedding vectors is used to obtain the personalized high-dimensional cancer pain representation space.

5. The cancer pain level assessment method based on personalized representation space according to claim 1, characterized in that, Step 300 includes: Step 310: Obtain the patient's individual pain data during the initial stage of hospitalization, and determine the pain-free anchor point and risk anchor point in the individual high-dimensional cancer pain representation space using K-means based on the individual pain data; Step 320: Real-time detection of patient-specific pain data to obtain real-time personalized pain data, and mapping the real-time personalized pain data into a real-time personalized pain vector; Step 330: Calculate the relative pain index based on the pain-free anchor point, risk anchor point, and individual pain vector.

6. The cancer pain level assessment method based on personalized representation space according to claim 1, characterized in that, Step 400 includes: Step 410: Collect and obtain the latest homeostatic physiological data according to the latest clinical protocol; Step 420: Update the pain-free anchor point and risk anchor point using an exponential moving average based on the latest steady-state physiological data to update the relative pain index.

7. The cancer pain level assessment method based on personalized representation space according to claim 6, characterized in that, Step 400 further includes: Step 430: If the relative pain index changes, start parameter fine-tuning and train to obtain a second personalized LoRA adapter based on the latest homeostatic physiological data; Step 440: Perform weighted interpolation on the first personalized LoRA adapter based on the second personalized LoRA adapter to fine-tune the parameters.

8. A cancer pain level assessment system based on a personalized representation space, employing the cancer pain level assessment method based on a personalized representation space as described in any one of claims 1-7, characterized in that, include: The first acquisition module is used to pre-train a large language model and set a learnable linear projection unit in the pre-trained large language model. The encoding unit and mapping unit in the pre-trained large language model are used to encode and map the facial video stream and skin electroreactivity signal, and the linear projection unit is used to obtain the video token and physiological token. The second acquisition module is used to acquire patient-specific data, construct patient-specific instruction pairs based on the patient-specific data, and use QLoRA technology to fine-tune the individual parameters and select the structure of the pre-trained large language model to obtain a personalized embedding vector. Based on the personalized embedding vector, a personalized high-dimensional cancer pain representation space is obtained. The calculation module is used to collect and acquire the patient's individual pain data in the early stage of hospitalization, and calculate the patient's relative pain index based on the individual pain data and the individual high-dimensional cancer pain representation space. The update module is used to acquire the latest homeostatic physiological data of the patient's condition at a new stage in real time, and update the adapter of the pre-trained large language model and the relative pain index based on the latest homeostatic physiological data.

9. A computer device, characterized in that, The method includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the cancer pain level assessment method based on a personalized representation space as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions that, when executed by a processor, implement the steps of the cancer pain level assessment method based on a personalized representation space as described in any one of claims 1 to 7.

Images