Chemotherapy-induced peripheral neuropathy prediction method and system based on quantum enhanced traditional Chinese and western medicine fusion, and electronic equipment

Abstract

The invention discloses a chemotherapy-induced peripheral neuropathy prediction method and system based on quantum enhanced traditional Chinese and western medicine fusion and electronic equipment, relates to the technical field of disease risk prediction, and aims to solve the problems that an existing prediction method is low in accuracy and insufficient in early warning capability and neglects individual differences. The method comprises the following steps: collecting original multi-modal data of an existing patient, extracting a feature vector set, carrying out multi-modal fusion, constructing and training a chemotherapy-induced peripheral neuropathy prediction model based on quantum enhanced traditional Chinese and western medicine fusion, and predicting the chemotherapy-induced peripheral neuropathy (CIPN) risk of the patient to be predicted through the model. According to the method, western medicine pharmacology data and traditional Chinese medicine diagnosis insight are deeply fused through quantum calculation and a deep learning technology through a quantum enhanced multi-mode fusion framework, precise prediction of CIPN is realized, a CIPN prevention and treatment strategy is optimized, individualized treatment is realized, the life quality of a patient is improved, and chemotherapy related complications are reduced.

Description

Technical Field

[0001] This invention relates to the field of disease risk prediction technology, and in particular to a method, system and electronic device for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine. Background Technology

[0002] Chemotherapy-induced peripheral neuropathy (CIPN) is the most common dose-limiting toxicity during chemotherapy in cancer patients. The incidence of CIPN can be as high as 90% in patients receiving platinum-based, taxane-based, or vinca alkaloid-based chemotherapy. CIPN often manifests as sensory abnormalities, motor dysfunction, and autonomic nervous system disorders, severely impacting patients' quality of life and frequently leading to chemotherapy dose reduction or discontinuation, directly threatening the effectiveness of cancer treatment and patient survival.

[0003] Currently, the prediction and assessment of CIPN mainly rely on three types of methods: clinical assessment scales based on the NCI-CTCAE classification criteria (National Cancer Institute Common Adverse Event Terminology Standard), invasive neurophysiological examinations, and simplified scoring systems based on risk factors such as age and comorbidities. These methods have significant limitations: the predictive accuracy is generally low (approximately 60%–70%); assessments are mostly conducted after symptom onset, lacking early warning capabilities; the assessment dimensions are limited, failing to fully consider individual genetic background and physical differences, especially neglecting individualized information such as traditional Chinese medicine constitution and symptoms; and they cannot dynamically predict the development trajectory and severity evolution of CIPN.

[0004] Traditional Chinese medicine categorizes CIPN under the terms "Bi syndrome" and "numbness," and its diagnostic system has unique advantages in identifying patients' constitutions and symptoms, providing important references for individualized prevention and treatment. However, traditional Chinese medicine diagnosis relies on doctors' subjective experience and lacks objective quantitative standards, making it difficult to effectively integrate with the modern medical system.

[0005] In recent years, technologies such as artificial intelligence and quantum computing have provided new tools for medical prediction. However, current technologies lack a CIPN risk assessment scheme that can systematically integrate Western medicine pharmacology data and traditional Chinese medicine diagnostic information, and utilize quantum enhancement and deep learning for multimodal fusion. Therefore, developing a system capable of early, dynamic, and personalized accurate prediction has become an urgent need to optimize clinical decision-making in CIPN. Summary of the Invention

[0006] To address the aforementioned problems, this invention aims to provide a method, system, and electronic device for predicting chemotherapy-induced peripheral neuropathy (CIPN) based on quantum-enhanced integration of traditional Chinese and Western medicine. Through a quantum-enhanced multimodal fusion architecture, it deeply integrates Western medicine pharmacological data with traditional Chinese medicine diagnostic insights using quantum computing and deep learning technologies to achieve accurate prediction of CIPN, optimize CIPN prevention and treatment strategies, realize personalized treatment, improve patients' quality of life, and reduce chemotherapy-related complications.

[0007] To achieve the above objectives, the technical solution adopted by the present invention is as follows: On one hand, this invention provides a method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine, the method comprising: Collect and preprocess the original multimodal data of existing patients to extract feature vector sets; Multimodal fusion of feature vectors in the feature vector set; A predictive model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine was constructed and trained. The trained model was used to predict chemotherapy-induced peripheral neuropathy using multimodal clinical data of patients to be predicted.

[0008] Optionally, the multimodal fusion of feature vectors in the feature vector set includes: Convert the symptom vectors in the eigenvector set into quantum states; Deep feature extraction of chemotherapy drugs; Align all features with a time dimension with time series data. Adaptive normalization and fusion are performed on all processed features.

[0009] Optionally, the process of constructing and training a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine includes: The data input module is used to input the fused feature vector; The quantum TCM syndrome coding module is used to model the relationships between TCM diagnostic features; A pharmacokinetic-syndrome interaction network module is used to capture the interaction patterns between chemotherapy drug characteristics and TCM diagnostic characteristics; The Spatiotemporal Feature Fusion Transformer module is used for deep fusion of spatiotemporal features; The adaptive risk trajectory modeling module continuously predicts the risk trajectory of chemotherapy-induced peripheral neuropathy based on the outputs of the quantum TCM syndrome coding module, the pharmacokinetics-syndrome interaction network module, and the spatiotemporal feature fusion Transformer module. The data output module is used to output the prediction results.

[0010] Optionally, the specific operation steps of the quantum TCM syndrome encoding module are as follows: Entangled-state representation of the relationships between TCM diagnostic features using quantum gate operations; The syndrome status is determined by quantum measurement through clinical observation.

[0011] Optionally, the specific operation steps of the pharmacokinetic-syndrome interaction network module are as follows: Encode the drug- and syndrome-related parts of the fused feature vector to establish drug node and syndrome node representations; By modeling edge relationships, causal interaction relationships are established between nodes; The importance weights of different drug-syndrome pairs in the occurrence of chemotherapy-induced peripheral neuropathy are learned through a message passing mechanism. The drug-symptom pairs are weighted using an attention mechanism to generate graph representation vectors.

[0012] Optionally, the risk dynamic equation in the adaptive risk trajectory modeling module is expressed as: ; In the formula, The situation is at risk. For drug exposure function, For the syndrome evolution function, For baseline risk, for Hierarchical functions, These are the learning parameters.

[0013] Optionally, in the construction of a quantum-enhanced integrated traditional Chinese and Western medicine prediction model for chemotherapy-induced peripheral neuropathy, and during the training of the model, the composite loss function in the end-to-end training phase is expressed as: ; In the formula, For binary classification, cross-entropy loss, For probability regression loss, For trajectory prediction loss, For the standardization of consistency between traditional Chinese and Western medicine, for Graded loss prediction.

[0014] Optionally, the step of using a trained model to predict chemotherapy-induced peripheral neuropathy using multimodal clinical data of the patient to be predicted includes: Collect multimodal clinical data of patients to be predicted and perform preprocessing and multimodal fusion; The fused feature vector is input into the trained prediction model of chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine to predict the risk of chemotherapy-induced peripheral neuropathy. Generate and analyze the prediction results.

[0015] On the other hand, the present invention also provides a prediction system for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine, comprising: The multimodal data acquisition and preprocessing module is used to acquire raw multimodal data from existing patients and extract feature vector sets; The feature vector fusion module is used to perform multimodal fusion of feature vectors from the feature vector set; The model building and training module is used to build a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, and to train and optimize the model performance. The clinical prediction module uses a trained model to predict the risk of chemotherapy-induced peripheral neuropathy in patients to be predicted. Among them, the multimodal data acquisition and preprocessing module, feature vector fusion module, model construction and training module, and clinical prediction module are based on the previously described quantum-enhanced integrated traditional Chinese and Western medicine prediction method for chemotherapy-induced peripheral neuropathy.

[0016] In another aspect, the present invention also provides an electronic device, including at least one processor; and a memory communicatively connected to the processor; wherein the memory stores instructions executed by the processor, the instructions being executed by the processor to enable the processor to perform the previously described method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine.

[0017] The beneficial effects of this invention are: (1) This invention uses a quantum-enhanced multimodal fusion architecture to achieve deep feature extraction and interactive modeling of Western medicine pharmacology data and traditional Chinese medicine diagnostic information. The model achieves a prediction accuracy of 82.3% and an AUC-ROC value of 0.82 in the test, which is significantly better than traditional methods. In particular, the model can provide an early warning of the risk of CIPN 22 days in advance, providing a sufficient and critical time window for clinical implementation of early preventive intervention.

[0018] (2) This invention introduces quantum computing principles to encode TCM syndromes for the first time. By representing the complex coexistence relationship of syndromes through quantum superposition, it solves the problem of standardization and quantification of TCM information. The fusion model constructed in this way not only conforms to the internationally accepted CTCAE assessment standard, but also incorporates the individualized characteristics of TCM constitution and syndromes, retains the individualized advantages of TCM, and creates a new paradigm of quantitative integration of TCM and Western medicine. It realizes a truly individualized risk assessment, thereby generating a tailor-made risk prediction and prevention plan for each patient.

[0019] (3) This invention uses Neural ODE to model the risk of CIPN over continuous time, which can output the risk evolution trajectory over the next few weeks, providing a prospective basis for the dynamic adjustment of chemotherapy regimens. At the same time, the model integrates an interpretability module, which can output key influencing factor analysis, risk level determination and specific prevention and treatment suggestions, effectively assisting clinical decision-making and enhancing clinical guidance value.

[0020] (4) This invention, through accurate prediction and early intervention, is expected to significantly reduce the incidence of clinical CIPN, decrease related hospitalization needs, and significantly improve patients' quality of life. Simultaneously, by optimizing the allocation of medical resources, it can save patients' treatment time and reduce unnecessary medical expenses, demonstrating significant clinical benefits and health economic value. Furthermore, this method provides feasible technical paths and research paradigms for cutting-edge interdisciplinary fields such as integrated traditional Chinese and Western medicine, artificial intelligence medicine, and quantum medicine, and has positive significance for promoting interdisciplinary development. Attached Figure Description

[0021] Figure 1 This is a flowchart of the chemotherapy-induced peripheral neuropathy prediction method based on quantum-enhanced integration of traditional Chinese and Western medicine in this invention.

[0022] Figure 2 This is a schematic diagram of the pharmacokinetic-syndrome interaction network module of the present invention.

[0023] Figure 3 This is an architecture diagram of the spatiotemporal feature fusion Transformer module in this invention.

[0024] Figure 4 This is a flowchart of the training process for the chemotherapy-induced peripheral neuropathy prediction model based on quantum-enhanced integration of traditional Chinese and Western medicine in this invention.

[0025] Figure 5 This is a training error curve of the prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine in this invention.

[0026] Figure 6 This is a test error curve of the chemotherapy-induced peripheral neuropathy prediction model based on quantum-enhanced integration of traditional Chinese and Western medicine in this invention.

[0027] Figure 7 This is a training accuracy curve of the prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine in this invention.

[0028] Figure 8 This is a curve showing the accuracy of the prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine in this invention.

[0029] Figure 9 This is a comparison chart of ROC curves for predicting the risk of chemotherapy-induced peripheral neuropathy using different prediction methods in this invention.

[0030] Figure 10 This is a flowchart illustrating the clinical application of the chemotherapy-induced peripheral neuropathy prediction model based on quantum-enhanced integrated traditional Chinese and Western medicine in this invention. Detailed Implementation

[0031] To enable those skilled in the art to better understand the technical solutions of the present invention, the technical solutions of the present invention will be further described below with reference to the accompanying drawings and embodiments. Obviously, the described embodiments are only a part of the embodiments, and not all of the embodiments.

[0032] Example 1:

[0033] Example 1 provides a method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, as shown in the attached figure. Figure 1 As shown, the method specifically includes the following steps: Step 1: Collect the original multimodal data of existing patients and preprocess it to extract the feature vector set; Specifically, the existing patient's original multimodal data includes four types: chemotherapy regimen information, traditional Chinese medicine (TCM) diagnostic information, CTCAE assessment data, and clinical baseline data. The chemotherapy regimen information includes drug type, cumulative dose, and administration regimen; the TCM diagnostic information includes information obtained from tongue diagnosis, pulse diagnosis, inquiry, and observation. Tongue diagnosis requires the use of standardized tongue image acquisition equipment to obtain tongue color, coating, and texture characteristics; pulse diagnosis requires recording 28 pulse types, including cun, guan, chi, superficial, middle, and deep; inquiry requires collecting syndrome-related symptoms and constitution type; and observation requires recording the patient's expression, morphology, and local signs; the CTCAE assessment data includes current grading, historical grading sequence, and sensory / motor neuropathy grading; and the clinical baseline data includes patient age, gender, past medical history (e.g., diabetes), and comorbidities.

[0034] To ensure data integrity and reliability, the original multimodal data was first cleaned to check its integrity, ensuring that over 90% of the required fields were complete, and identifying, correcting, or deleting obviously erroneous records. For missing data, multiple imputation methods were used. Then, the original data was standardized and normalized to eliminate the influence of different feature units and numerical ranges. Next, feature encoding was performed, converting non-numerical categorical text data into numerical format using a standardized scale. Finally, a feature vector set containing 290 dimensions was constructed. This included 128 dimensions for chemotherapy drug features (drug category code, cumulative dose, dosing regimen, and time series); 106 dimensions for TCM diagnostic features (tongue diagnosis features 40, pulse diagnosis features 38, syndrome vector 16, and constitution type 12); 24 dimensions for CTCAE assessment features (current grade 6, historical sequence 12, and sensory / motor dissociation 6); and 32 dimensions for baseline clinical features (demographic 8, past medical history 12, and laboratory indicators 12).

[0035] Step 2: Perform multimodal fusion on the feature vectors in the feature vector set; Optionally, step 2 includes the following sub-steps: Sub-step 201: Convert the symptom vectors in the eigenvector set into quantum states; Specifically, the traditional symptom vector is initially converted into a form that can be processed by quantum computing. The 16-dimensional symptom vector is encoded into a 16-dimensional complex vector using a 4-qubit circuit. This involves the following two steps: (1) Quantum state initialization: The 16-dimensional syndrome vector in TCM diagnostic features is mapped to the quantum state space, and the computation basis of 8 basic syndromes is defined: Qi deficiency|000, blood stasis|001, phlegm dampness|010, Yin deficiency|011, Yang deficiency|100, heat toxicity|101, rheumatism|110, and kidney essence deficiency|111.

[0036] (2) Representing the coexistence of multiple symptoms through the principle of quantum superposition. (1) In the formula, This represents the complete quantum state of the syndrome vector in the TCM diagnostic characteristics of a patient; This represents the syndrome state index (from 0 to 7, corresponding to 2³=8 possible TCM syndrome states). Indicates the first Individual syndrome ground state (pure syndrome); Indicates the state of syndrome The complex amplitude of , where the real part represents the intensity of the syndrome and the imaginary part represents the phase relationship between the syndromes.

[0037] Sub-step 202: Perform deep feature extraction on chemotherapy drug characteristics; Specifically, deep learning models such as one-dimensional convolutional neural networks (1D-CNN) or recurrent neural networks (RNN) are used to automatically extract high-level, neurotoxicity-related feature patterns from time-series data of 128-dimensional chemotherapy drug features, and output a 256-dimensional deep feature vector to capture the dynamic exposure patterns of the drug in vivo.

[0038] Sub-step 203: Align all features with a time dimension to a time series; Specifically, the Dynamic Time Warping (DTW) algorithm is used to align sequence data from different sources that may have misaligned timestamps, ultimately outputting modal features with a unified time scale, ensuring that the model can correctly learn cross-modal temporal causal relationships.

[0039] Sub-step 204: Perform adaptive normalization and fusion on all processed features; Specifically, based on the distribution of different types of features, the most suitable normalization strategy is automatically selected to unify all processed features to the same scale and perform initial concatenation, outputting a 512-dimensional, standardized, and deeply fused feature vector to be predicted. More specifically, the fusion process is completed through a three-stage workflow, which merges different types of feature data into a unified representation: Phase 1: Collect all features processed in the previous steps, including: a 16-dimensional complex vector of syndromes representing the fundamental quantum states; a 256-dimensional deep chemotherapy drug feature vector describing the properties and interactions of chemotherapy drugs; time-series data of chemotherapy drug features and TCM diagnostic features synchronized over time; and 32-dimensional baseline clinical features containing information such as age, gender, and medical history. Arrange and concatenate these features side-by-side to create a large feature vector of approximately 550-560 dimensions.

[0040] Phase 2: Each set of features is analyzed using adaptive normalization to determine their distribution. For normally distributed data (e.g., blood test results), z-score normalization is used to center the data to zero with a standard deviation of 1; for count data (e.g., drug dosage), logarithmic transformation is used to handle the large numerical range; for binary data (yes / no features), they are left unchanged; for data containing outliers, robust scaling based on the median and interquartile range is used; for unknown distributions, rank-based normalization is used.

[0041] Phase 3: Final fusion and standardization of all normalized features. Using learned linear transformations (e.g., finding the most important combinations through weighted averaging), all features are projected onto a 512-dimensional scale; further refinement is performed to ensure all features are on the same scale; a 512-dimensional feature vector is output, representing all patient information in a unified format.

[0042] Step 3: Construct a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine (MICI-CIPN model) and train the model. The prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integration of traditional Chinese and Western medicine includes a data input module, a quantum TCM syndrome coding module (QTSE module), a pharmacokinetic-syndrome interaction network module (PSIN module), a spatiotemporal feature fusion Transformer module (TSFFT module), an adaptive risk trajectory modeling module (ARTM module), and a data output module. The data input module is used to input the fused feature vector from step 2; The quantum TCM syndrome coding module is used to model the relationship between TCM diagnostic features; The pharmacokinetic-syndrome interaction network module is used to capture the interaction patterns between chemotherapy drug characteristics and TCM diagnostic characteristics; The spatiotemporal feature fusion Transformer module is used for deep fusion of spatiotemporal features; The adaptive risk trajectory modeling module continuously predicts the risk trajectory of chemotherapy-induced peripheral neuropathy based on the outputs of the quantum TCM syndrome coding module, the pharmacokinetics-syndrome interaction network module, and the spatiotemporal feature fusion Transformer module. The data output module is used to output the prediction results.

[0043] In some embodiments, the quantum TCM syndrome encoding module models the complex relationships between TCM diagnostic features through quantum gate operations and entanglement. By performing deep, parameterized quantum operations on the 16-dimensional fundamental quantum state transformed in sub-step 201 using an 8-qubit system, the module uncovers the complex, non-classical interactions between syndromes and between syndromes and other TCM diagnostic features, ultimately outputting 128-dimensional quantum features. More specifically, the specific operation process of the quantum TCM syndrome encoding module includes: (1) Quantum entanglement between 16-dimensional fundamental quantum states and between tongue diagnosis and pulse diagnosis features can be created by applying controlled quantum gates. For example, quantum entanglement can be used to reflect the intrinsic correlation between tongue diagnosis and pulse diagnosis. Represented as (2) (2) Quantum measurements are performed through clinical observation to cause the quantum state to collapse, thereby determining the final enhanced and correlated syndrome state.

[0044] In some embodiments, the pharmacokinetic-syndrome interaction network module encodes the drug- and syndrome-related parts of the 512-dimensional fusion features obtained in step 2, establishing a heterogeneous graph structure of drug nodes, syndrome nodes, and interaction edges, and finally outputting a 256-dimensional graph structure representation vector to capture the complex interaction patterns between chemotherapy drug features and traditional Chinese medicine diagnostic features. More specifically, as shown in the appendix... Figure 2 As shown, the specific implementation steps of the pharmacokinetic-syndrome interaction network module include: (1) Node definition: Drug nodes are defined as: platinum (cisplatin, carboplatin, oxaliplatin), taxanes (paclitaxel, docetaxel), vinblastine alkaloids (vincristine, vinorelbine), and others, totaling 128 dimensions. Syndrome nodes are defined as 8 TCM syndrome types, totaling 64 dimensions.

[0045] (2) Edge Relationship Modeling: The strength of drug-syndrome interaction is calculated based on historical data. The causal effect is calculated using the do-calculus method, taking into account the influence of drug exposure time on the interaction. The do-calculus formula is as follows: (3) In the formula, This indicates that when we intervene in the settings Time result The probability, express The resulting variable, This indicates the therapeutic variables of drug administration. Indicates causal interference. Confounding variables representing patient characteristics and comorbidities. Indicates the confounding factor Sum of all possible values, Indicates a given and The conditional probability, This represents the prior probability of the confounding factor value.

[0046] The backdoor adjustment formula is: (4) In the formula, Indicates a specific intervention value (e.g., drug dosage). This represents a specific confounding factor value. This formula is used to block confounding pathways to isolate causal effects, thereby estimating the true effect of the drug on CIPN, independent of patient factors.

[0047] (3) Message passing mechanism: A 4-layer GraphSAGE is used for information aggregation. Through an attention mechanism, the model learns the importance weights of different drug-syndrome pairs in the occurrence of CIPN. The formula for information aggregation using GraphSAGE is as follows: (5) In the formula, This indicates the central node that is being updated. For nodes The expression, Represents a node neighborhood Each neighbor node in the network, Representing neighboring nodes In the Layer feature representation, Representation layer index, Indicates for All in the neighborhood , For aggregate functions, For activation function, For weights.

[0048] (4) Attention Weighting: An 8-head attention mechanism is used to weight important drug-syndrome pairs, ultimately outputting a 256-dimensional graph representation vector. The formula for calculating the attention weights is as follows: (6) In the formula, and Represents the graph node index in the drug-syndrome interaction network. This indicates the target node that is being followed. Indicates the source node that provides the information. Represents the normalized attention weights (i.e., node weights) For nodes (importance) Normalize the weights of all neighbors to a sum of 1. For nodes The expression, For nodes The expression .

[0049] In some embodiments, the spatiotemporal feature fusion Transformer module performs cross-modal attention calculation on the time series features and spatial distribution features in the 512-dimensional fused features obtained in step 2, thereby achieving deep fusion of spatiotemporal features. More specifically, as shown in the appendix... Figure 3 As shown, the spatiotemporal feature fusion Transformer module adopts a two-stream architecture, specifically including the following steps: (1) Time encoder: Input chemotherapy cycle time series data, use sine and cosine position encoding, capture time dependence through self-attention mechanism. The structure of the time encoder is a 6-layer Transformer encoder, each layer is 512-dimensional.

[0050] (2) Spatial encoder: Input the spatial distribution characteristics of TCM diagnosis and learn the relationship between syndromes through self-attention mechanism. The spatial encoder has a 6-layer Transformer encoder with 512 dimensions per layer.

[0051] (3) Cross-modal fusion: Deep fusion of spatiotemporal features is achieved through an 8-head cross-modal attention mechanism. Temporal encoding is used as the query, and spatial encoding is used as the key and value. Each of the 8 attention heads learns different spatiotemporal association patterns. The proportion of spatiotemporal features is dynamically adjusted through a gating fusion mechanism, and finally a 512-dimensional spatiotemporal fusion feature representation is output.

[0052] In some embodiments, the adaptive risk trajectory modeling module uses Neural ODE to perform adaptive risk trajectory modeling based on the outputs of the quantum TCM syndrome encoding module, the pharmacokinetics-syndrome interaction network module, and the spatiotemporal feature fusion Transformer module, and continuously predicts the risk trajectory of chemotherapy-induced peripheral neuropathy.

[0053] The risk dynamic equation is (7) In the formula, The situation is at risk. For drug exposure function, For the syndrome evolution function, For baseline risk, for Hierarchical functions, These are the learning parameters.

[0054] use The adaptive step-size solver is used to solve the problem, and the initial state is set. It automatically adjusts the step size based on error estimation and outputs a trajectory sequence of CIPN risk values ​​for each day over the next three weeks, starting from the current time point.

[0055] By integrating predictions to quantify uncertainty, specifically by training 10 independent models, calculating the mean and variance of the predictions, generating 95% confidence intervals, uncertainty is quantified, enabling continuous prediction of risk trajectories.

[0056] In some embodiments, the data output module is configured with multiple output layers in parallel, including binary judgment prediction results (occurrence / non-occurrence), probability scores (0~100%), risk levels (low / medium / high), 3-week risk trajectory prediction, key contributing factor analysis, CTCAE graded prediction, and personalized prevention and control suggestions.

[0057] Furthermore, the data output module also provides an interpretable multimodal decision tree (EMDT), in which the Western medicine branch includes assessment of drug toxicity, cumulative dose, and pharmacokinetic risk factors; the traditional Chinese medicine branch includes identification of syndrome patterns, constitution assessment, and tongue and pulse correlation analysis; integrating weighted combination, conflict resolution, and confidence calculation, it analyzes the contribution of each input feature to the final prediction result through SHAP value analysis, generates a report of key influencing factors, provides patient-specific interpretations, and enhances clinical credibility.

[0058] In some embodiments, the training process of the MICI-CIPN model specifically includes: The training parameters are set, and a preprocessed multimodal feature vector set is used. The MICI-CIPN deep learning network is trained in a supervised manner using the AdamW optimizer to measure the difference between the model's predictions and the actual occurrence of CIPN events, enabling the model to extract effective feature representations from the original multimodal data. Key training parameters include: a maximum of 200 training epochs, a multi-stage training strategy (50 epochs of component pre-training, 50 epochs of progressive unfreezing, 80 epochs of end-to-end training, and 20 epochs of adversarial training), 32 training samples per batch, and ensuring that 20% of the data is used as a validation set and 10% as a test set to avoid overfitting. The MICI-CIPN model training process is attached. Figure 4 As shown.

[0059] The core module parameter configurations of the MICI-CIPN model are shown in Table 1 below, and the training hyperparameter configurations are shown in Table 2 below.

[0060] Table 1. Parameter Configuration Table of Core Modules of the MICI-CIPN Model ; Table 2 Training Hyperparameter Configuration Table ; The multi-stage training strategy specifically includes the following four stages: Phase 1: Component pre-training (rounds 1-50) The QTSE module, PSIN module, and TSFFT module were pre-trained independently.

[0061] (1) QTSE module pre-training: Using the TCM syndrome reconstruction task, the quantum state representation of the syndrome is learned through an autoencoder architecture, and the loss function is the reconstruction error MSE. The quantum entanglement complexity is gradually increased during training, starting from 2 qubits and gradually expanding to 8 qubit systems.

[0062] (2) PSIN module pre-training: Construct a drug-syndrome interaction prediction task and learn the causal relationship graph structure using historical case data. A joint loss of link prediction and node classification is adopted with a weight ratio of 0.6:0.4.

[0063] (3) TSFFT module pre-training: Using the masked language model (MLM) strategy, 15% of the temporal features are randomly masked, and the model is trained to reconstruct the complete sequence. Contrastive learning is also introduced to make the representations of similar patients more similar.

[0064] Phase 2: Gradual Thawing (Rounds 51-100) The gradual unfreezing strategy starts from the output layer and gradually unfreezes the deeper parameters.

[0065] (1) Rounds 51-60: Only train the output layer and the last two layers, freeze the parameters of other layers, and set the learning rate to 0. .

[0066] (2) Rounds 61-80: Unfreeze the middle layer (layers 3-4), increase the learning rate to .

[0067] (3) Rounds 81-100: Unfreeze all layers, adjust the learning rate to This enables complete model adaptation.

[0068] Phase 3: End-to-end fine-tuning (rounds 101-180) During the end-to-end training phase, a composite loss function is used for joint optimization. The composite loss function is expressed as follows: (8) In the formula, For binary classification cross-entropy loss, use Imbalanced processing categories; To ensure risk score calibration, the MSE loss is calculated using probabilistic regression. This is the trajectory prediction loss, used to ensure temporal consistency; For consistency regularization between traditional Chinese medicine and Western medicine, the weight is 0.1; For CTCAE hierarchical prediction loss, multi-class cross-entropy is used.

[0069] The training process employs a gradient accumulation strategy, updating parameters every four mini-batches, with an effective batch size of 128. Mixed precision training (FP16) is used to accelerate computation while maintaining numerical stability.

[0070] Phase 4: Combat Training (Rounds 181-200) Adversarial examples are introduced during the adversarial training phase to enhance the robustness of the model.

[0071] (1) Use FGSM (Fast Gradient Sign Method) to generate adversarial perturbations, perturbation strength =0.01.

[0072] (2) The adversarial samples and the original samples are mixed for training, with a ratio of 1:3.

[0073] (3) Reduce the learning rate to To avoid catastrophic amnesia.

[0074] (4) Introduce KL divergence regularization to maintain the original performance.

[0075] To improve the model's generalization ability, various data augmentation techniques were employed during training, expanding the original 8,000 training samples to 24,000. The data augmentation methods and parameters used during training are shown in Table 3 below.

[0076] Table 3 Data Augmentation Methods and Parameters ; During model training, the following key metrics are monitored and recorded in real time: (1) Main performance indicators: AUC-ROC, AUC-PR, accuracy, sensitivity, specificity.

[0077] (2) Calibration indicators: expected calibration error (ECE), Brier score.

[0078] (3) Training health: gradient norm, weight update magnitude, attention entropy, proportion of dead neurons.

[0079] Bayesian optimization is used to search for the optimal combination of hyperparameters. The search space includes: (1) Learning rate: Logarithmic scale, optimal value .

[0080] (2) Number of GNN layers: [2, 3, 4, 5, 6], with the optimal value being 4 layers.

[0081] (3) Hidden dimensions: [128, 256, 512], with the optimal value being 256.

[0082] (4) Dropout rate: [0.1, 0.5], with an optimal value of 0.31.

[0083] During model training, the model training error curve, test error curve, training accuracy curve, and test accuracy curve are monitored and plotted in real time, as shown in the attached figures. Figure 5 Appendix Figure 6 Appendix Figure 7 Appendix Figure 8 As shown, with the increase in training epochs, both training and testing errors steadily decreased and tended to converge, eventually converging to 0.21 for training and 0.22 for testing. Simultaneously, the accuracy steadily increased and eventually stabilized at a high level, reaching 86.5% for training and 82.3% for testing. This indicates that the model training process was effective, without serious overfitting or underfitting, and that the model possessed good learning and generalization abilities. Particularly noteworthy is that after introducing adversarial training in the 180th epoch, although the training error increased slightly (from 0.21 to 0.23), the testing error remained stable, demonstrating that adversarial training effectively improved the model's robustness without compromising its generalization performance.

[0084] The CIPN risk prediction performance of traditional machine learning methods, basic clinical scoring systems, traditional Chinese medicine syndrome differentiation methods, random guessing methods, and the MICI-CIPN model of this invention was compared, and the ROC curves of CIPN risk prediction for different prediction methods were plotted, as shown in the attached figure. Figure 9 As shown in the figure, the independent test set contains 1000 patient samples. The green curve represents the MICI-CIPN model of this invention, the blue curve represents the traditional machine learning method (random forest), the orange curve represents the traditional TCM syndrome differentiation method, the red curve represents the basic clinical scoring system (based on CTCAE standards), and the black dashed line represents random guessing. The results show that the AUC value (area under the ROC curve, the closer the value is to 1, the better the model performance) of the MICI-CIPN model of this invention is 0.820, the AUC value of the traditional machine learning method is 0.750, the AUC value of the traditional TCM syndrome differentiation method is 0.700, and the AUC value of the basic clinical scoring system is 0.680. The results indicate that the AUC value of the model of this invention reaches 0.820, which is significantly higher than that of the traditional method, proving that the quantum-enhanced TCM-Western medicine integration prediction method of this invention has excellent predictive ability and clinical applicability.

[0085] On the ROC curve, by maximizing the Youden index (sensitivity + specificity - 1), the optimal operating point of the MICI-CIPN model of this invention was determined to be a risk threshold of 0.425, with a sensitivity of 92.3%, specificity of 89.7%, positive predictive value (PPV) of 87.2%, negative predictive value (NPV) of 93.8%, F1 score of 0.897, and Youden index of 0.820. This means that at the optimal threshold, the model of this invention can correctly identify 92.3% of high-risk CIPN patients and correctly exclude 89.7% of low-risk patients. Among patients predicted to be high-risk, 87.2% actually develop CIPN, and among patients predicted to be low-risk, 93.8% do not actually develop CIPN.

[0086] The contribution of each module to the AUC value was analyzed through ablation experiments. The contribution of each module to the model performance is shown in Table 4 below.

[0087] Table 4. Contribution of each module to model performance ; The ablation experiment results showed that the QTSE module contributed the most to the model performance (6.7%), verifying the advantages of quantum computing in dealing with complex syndrome relationships in traditional Chinese medicine.

[0088] Furthermore, the integration of Chinese and Western medicine features produced a significant synergistic effect: the AUC value of Western medicine features alone was 0.732, the AUC value of traditional Chinese medicine features alone was 0.698, the AUC value of simple splicing fusion was 0.758, the AUC value of quantum-enhanced deep fusion in this invention was 0.820, and the synergistic gain was 0.820 - max(0.732, 0.698) = 0.088, that is, the integration of Chinese and Western medicine produced a synergistic gain of 8.8%.

[0089] The comparison of warning times for different prediction methods is shown in Table 5 below. As can be seen from the comparison data, the present invention can provide a warning of CIPN risk 22 days in advance, providing a sufficient time window for clinical intervention.

[0090] Table 5. Early Warning Schedules for Different Forecasting Methods ; The predictive performance of CIPN of different severities is shown in Table 6 below. It can be seen that the model has better predictive performance for severe CIPN, which is of great significance for clinical decision-making.

[0091] Table 6. Predictive performance of CIPN at different severities ; The stability of the model was evaluated using 20-fold cross-validation. The mean AUC was 0.820 ± 0.018, the minimum AUC was 0.785, the maximum AUC was 0.847, and the coefficient of variation was 2.2%. The low coefficient of variation indicates that the model performance is stable and there is no overfitting.

[0092] Step 4: Use the trained model to predict chemotherapy-induced peripheral neuropathy using multimodal clinical data of the patients to be predicted. Optionally, step 4 includes the following sub-steps: Sub-step 401: Collect multimodal clinical data of the patient to be predicted and perform preprocessing and multimodal fusion; Specifically, following the same specifications and standards as the model training phase, four types of raw data from the patients to be predicted are collected: chemotherapy regimen information, traditional Chinese medicine diagnostic information, CTCAE assessment data, and clinical baseline data. The raw data undergoes data integrity verification to ensure that over 90% of the required fields are complete, meeting the basic input requirements of the model.

[0093] The newly acquired multimodal clinical data were cleaned, standardized, and feature-encoded in the same way as in step 1 to extract a feature vector set. Then, the feature vectors in the feature vector set were fused in a multimodal manner according to the method in step 2.

[0094] Sub-step 402: Input the fused feature vector into the trained prediction model of chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine to predict the risk of chemotherapy-induced peripheral neuropathy. Specifically, the fused feature vector is input into a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine. The model automatically calls the quantum TCM syndrome coding module, the pharmacokinetics-syndrome interaction network module, the spatiotemporal feature fusion Transformer module, and the adaptive risk trajectory modeling module to perform forward propagation calculations, ensuring that the response time of the entire inference process is less than 500 milliseconds to meet the real-time requirements of clinical practice.

[0095] Sub-step 403: Generate and analyze the prediction results; Specifically, the model outputs prediction results in multiple dimensions in parallel, which the system then integrates and analyzes, including the following: (1) Analyze the core prediction: Obtain and record the binary prediction results (occurrence / non-occurrence), probability of occurrence (0~100%), and risk level (low / medium / high).

[0096] (2) Analyze the dynamic trajectory: Extract the risk evolution trajectory sequence and the corresponding 95% confidence interval for the next 3 weeks from the ARTM module.

[0097] (3) Analyze CTCAE prediction: Obtain the sensory and motor neuron disease grading predicted by the model for the next assessment.

[0098] (4) Perform interpretability analysis: Call the integrated SHAP analyzer to calculate the contribution of each input feature to the prediction result and generate the top N key influencing factors and their contribution values.

[0099] Based on the above analysis results, a structured clinical report is automatically generated, mainly including four parts: risk summary, visualization trajectory, personalized prevention and treatment plan, and explanatory notes. The risk summary prominently displays the risk level and probability of occurrence; the visualization trajectory plots the risk change curve and confidence interval for the next three weeks; the personalized prevention and treatment plan includes risk stratification management strategies, chemotherapy dosage adjustment recommendations, traditional Chinese medicine (TCM) treatment plans, preventive interventions, follow-up monitoring plans, and CTCAE-level early warning. If the predicted risk is high, or the CTCAE level is predicted to be level 3 or higher, the report generates a significant warning and provides specific recommendations for chemotherapy dosage adjustment, TCM intervention plans, and enhanced monitoring plans. If the predicted risk is moderate, preventive recommendations are provided, such as TCM constitution conditioning and rehabilitation plans; the explanatory notes list key influencing factors to help clinicians understand the basis of the model's decisions. The clinical application process of the quantum-enhanced integrated TCM and Western medicine chemotherapy-induced peripheral neuropathy prediction model is attached. Figure 10 As shown.

[0100] Example 2:

[0101] Example 2 demonstrates the predictive performance and clinical value of the prediction method in Example 1 through a typical clinical case.

[0102] Clinical Case 1: Early Warning for High-Risk Patients This case demonstrates the ability of the prediction method in Example 1 to identify risk escalation before clinical symptoms become severe.

[0103] Patient profile: A 62-year-old female breast cancer patient who started adjuvant paclitaxel therapy (175 mg / m²). 2 According to traditional Chinese medicine assessment, the patient has a deficiency of both qi and blood, and has no history of neurological disorders.

[0104] Model input: Multimodal data of the patient were collected after the first cycle of chemotherapy, including cumulative dose of 350 mg / m², tongue appearance (pale tongue, thin white coating), pulse (weak and thin), and CTCAE grade 1.

[0105] Model prediction and intervention: (1) The patient's risk probability is 67% (high risk); (2) Risk trajectory predictions indicate that the risk will continue to rise to 94% within the next 3 weeks; (3) Key influencing factors include Qi and Blood Deficiency Syndrome (contribution 28%), Paclitaxel Cumulative Dosage (contribution 25%), and Tongue Characteristics (contribution 15%); the CTCAE classification prediction result is that it may progress to grade 3 after 3 weeks; (4) Clinical intervention plan: Based on the early warning, the clinicians reduced the chemotherapy dose by 20% and used traditional Chinese medicine intervention with the Qi-tonifying and blood-nourishing formula.

[0106] Actual outcome: The patient ultimately only developed grade 2 CIPN, successfully avoiding severe grade 3 neurotoxicity, and completed the chemotherapy cycle smoothly.

[0107] Clinical Use Case 2: Individualized Prevention Guided by Traditional Chinese Medicine Syndromes This case illustrates how TCM syndrome assessment and treatment can prevent CIPN in high-risk patients.

[0108] Patient profile: A 55-year-old male patient with colorectal cancer, undergoing oxaliplatin chemotherapy. Traditional Chinese medicine assessment indicated phlegm-dampness with blood stasis and high baseline risk factors.

[0109] The timeline of model predictions and interventions is shown in Table 7 below.

[0110] Table 7 Model Prediction and Intervention Timeline ; As shown in Table 1, the patient's baseline risk was 58% (intermediate risk). Personalized TCM (Traditional Chinese Medicine) recommendations included initiating TCM treatments to resolve phlegm and dampness, and promote blood circulation. After continuous monitoring of the syndrome evolution, the risk decreased to 52% in week 3 and stabilized at 48% in week 6. Ultimately, the patient did not develop CIPN throughout the entire chemotherapy cycle, and the TCM syndrome significantly improved, demonstrating the effectiveness of TCM intervention in risk control.

[0111] Example 3:

[0112] Example 3 provides a prediction system for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, including a multimodal data acquisition and preprocessing module, a feature vector fusion module, a model construction and training module, and a clinical prediction module; The multimodal data acquisition and preprocessing module is used to acquire raw multimodal data from existing patients and extract feature vector sets; The feature vector fusion module is used to perform multimodal fusion of feature vectors from the feature vector set; The model building and training module is used to build a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, and to train and optimize the model performance. The clinical prediction module uses a trained model to predict the risk of chemotherapy-induced peripheral neuropathy in patients to be predicted. Among them, the multimodal data acquisition and preprocessing module, feature vector fusion module, model construction and training module, and clinical prediction module are implemented based on the chemotherapy-induced peripheral neuropathy prediction method based on quantum-enhanced integrated traditional Chinese and Western medicine described in Example 1.

[0113] Example 4:

[0114] Example 4 provides an electronic device, including at least one processor; and a memory communicatively connected to the processor; wherein the memory stores instructions that are executed by the processor, the instructions being executed by the processor to enable the processor to execute the chemotherapy-induced peripheral neuropathy prediction method based on quantum-enhanced integrated traditional Chinese and Western medicine described in Example 1.

[0115] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of this invention is defined by the appended claims and their equivalents.

Claims

1. A method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, characterized in that, The method includes: Collect and preprocess the original multimodal data of existing patients to extract feature vector sets; Multimodal fusion of feature vectors in the feature vector set; A predictive model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine was constructed and trained. The trained model was used to predict chemotherapy-induced peripheral neuropathy using multimodal clinical data of patients to be predicted.

2. The method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine according to claim 1, characterized in that, The multimodal fusion of feature vectors in the feature vector set includes: Convert the symptom vectors in the eigenvector set into quantum states; Deep feature extraction of chemotherapy drugs; Align all features with a time dimension with time series data. Adaptive normalization and fusion are performed on all processed features.

3. The method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine according to claim 2, characterized in that, The process of constructing and training a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine includes: The data input module is used to input the fused feature vector; The quantum TCM syndrome coding module is used to model the relationships between TCM diagnostic features; A pharmacokinetic-syndrome interaction network module is used to capture the interaction patterns between chemotherapy drug characteristics and TCM diagnostic characteristics; The Spatiotemporal Feature Fusion Transformer module is used for deep fusion of spatiotemporal features; The adaptive risk trajectory modeling module continuously predicts the risk trajectory of chemotherapy-induced peripheral neuropathy based on the outputs of the quantum TCM syndrome coding module, the pharmacokinetics-syndrome interaction network module, and the spatiotemporal feature fusion Transformer module. The data output module is used to output the prediction results.

4. The method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine according to claim 3, characterized in that, The specific operation steps of the quantum TCM syndrome coding module are as follows: Entangled-state representation of the relationships between TCM diagnostic features using quantum gate operations; The syndrome status is determined by quantum measurement through clinical observation.

5. The method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine according to claim 4, characterized in that, The specific operation steps of the pharmacokinetic-syndrome interaction network module are as follows: Encode the drug- and syndrome-related parts of the fused feature vector to establish drug node and syndrome node representations; By modeling edge relationships, causal interaction relationships are established between nodes; The importance weights of different drug-syndrome pairs in the occurrence of chemotherapy-induced peripheral neuropathy are learned through a message passing mechanism. The drug-symptom pairs are weighted using an attention mechanism to generate graph representation vectors.

6. The method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine according to claim 5, characterized in that, The risk dynamic equation in the adaptive risk trajectory modeling module is expressed as follows: ； In the formula, The situation is at risk. For drug exposure function, For the syndrome evolution function, For baseline risk, For CTCAE hierarchical functions, , , , , These are the learning parameters.

7. The method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine according to claim 6, characterized in that, The construction of a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, and the training of the model, are described. The composite loss function during the end-to-end training phase is expressed as follows: ； In the formula, For binary classification, cross-entropy loss, For probabilistic regression MSE loss, For trajectory prediction loss, For the standardization of consistency between traditional Chinese and Western medicine, Predict loss for CTCAE classification.

8. The method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine according to claim 7, characterized in that, The method of using a trained model to predict chemotherapy-induced peripheral neuropathy based on multimodal clinical data of the patient to be predicted includes: Collect multimodal clinical data of patients to be predicted and perform preprocessing and multimodal fusion; The fused feature vector is input into the trained prediction model of chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine to predict the risk of chemotherapy-induced peripheral neuropathy. Generate and analyze the prediction results.

9. A predictive system for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, characterized in that, include: The multimodal data acquisition and preprocessing module is used to acquire raw multimodal data from existing patients and extract feature vector sets; The feature vector fusion module is used to perform multimodal fusion of feature vectors from the feature vector set; The model building and training module is used to build a prediction model for chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine, and to train and optimize the model performance. The clinical prediction module uses a trained model to predict the risk of chemotherapy-induced peripheral neuropathy in patients to be predicted. The multimodal data acquisition and preprocessing module, feature vector fusion module, model construction and training module, and clinical prediction module are implemented based on the quantum-enhanced integrated traditional Chinese and Western medicine chemotherapy-induced peripheral neuropathy prediction method described in any one of claims 1-8.

10. An electronic device, characterized in that, The method includes at least one processor; and a memory communicatively connected to the processor; wherein the memory stores instructions that are executed by the processor to enable the processor to perform the method for predicting chemotherapy-induced peripheral neuropathy based on quantum-enhanced integrated traditional Chinese and Western medicine as described in any one of claims 1-8.

Images