Prognosis decision-making system for non-resectable hepatocellular carcinoma

Abstract

The invention discloses a prognosis decision-making system for unresectable hepatocellular carcinoma (uHCC). The prognosis decision-making system comprises the following steps: processing clinical data of a uHCC patient to obtain a clinical text input marker sequence and a CECT image block input marker sequence; a plurality of prognosis prediction models are obtained through training of the text input mark sequences and the image block input mark sequences of the multiple queues, and the decision-making unit constructs a decision-making support model based on the multiple prognosis prediction models and provides multiple treatment decisions. The decision-making system helps the uHCC patient to construct the optimal treatment decision and predict the corresponding prognosis survival score under various selectable treatment schemes, and provides powerful support for the personalized precise treatment of the uHCC so as to improve the non-surgical treatment effect and survival rate of the patient.

Description

Technical Field

[0001] The present application belongs to the field of biomedicine technology, and specifically, relates to a prognostic decision system for unresectable hepatocellular carcinoma. Background Art

[0002] Unresectable hepatocellular carcinoma (uHCC) is a global health challenge. It is a type of hepatocellular carcinoma that cannot be surgically removed due to tumor size, location, number, vascular invasion, or insufficient liver function reserve. Since the lesions cannot be effectively removed by surgery, the prognosis and survival rate are low. Currently, the main treatment strategies for uHCC are intra-arterial interventional therapy (IAIT), including transarterial chemoembolization (TACE) and hepatic arterialinfusion chemotherapy (HAIC), which rely on the unique heterogeneous imaging biomarkers of each patient. In addition, the combination of TACE or HAIC with molecular targeted drugs (MTAs) or immune checkpoint inhibitors (ICIs) also shows the potential to effectively prolong the survival of uHCC patients.

[0003] There is significant heterogeneity among uHCC patients, including tumor burden, liver function reserve, and general condition, which are usually quite different between individuals. The current non-surgical treatment options lack effective personalized decision-making, and treatment personnel often give clinical trial treatment options based on experience, which may be difficult to meet the needs of different patients. When formulating treatment plans for sick individuals, it is also necessary to comprehensively consider multiple factors such as their medical history, tumor characteristics, treatment goals, and previous plans. The current treatment decision-making process is often more complicated and requires collaboration and weighing of pros and cons by a multidisciplinary team, such as multi-department consultations, which will occupy additional medical resources; when medical personnel are insufficient, commonly used plans are usually used, which may not be consistent with the actual situation of the patient, and it is difficult to effectively improve the success rate of patient conversion treatment. Therefore, an accurate automated decision-making system is needed to assist treatment personnel in conducting prognosis assessments under multiple plans and thereby improve the efficiency of patient hospitalization management.

[0004] CN113470782A discloses a clinical prognosis prediction model before surgery for initially unresectable hepatocellular carcinoma, which is based on the variables of five patient samples: pre-SLR alpha-fetoprotein, number of cancer foci, portal vein cancer thrombus, maximum diameter of the main tumor, and change in tumor load after TACE treatment. The Nomogram model is trained and the patients suitable for salvage liver resection after TACE treatment and their prognosis survival are predicted. However, CN113470782A does not consider the features in the clinical images of uHCC patients and predict the prognosis of the disease, such as quantitative analysis of CT enhanced images (CECT), measurement of the volume, location, morphology, etc. of uHCC lesions, so as to provide more accurate data support for its prediction model and improve the accuracy of decision-making. In addition, it also does not publicly predict patients suitable for combined treatment with multiple treatment methods during the course of the disease and their prognosis, which is beneficial to the fact that receiving TACE monotherapy is not a feature of all uHCC patients.

[0005] Based on the above defects of uHCC prognosis method, this application is proposed. Summary of the invention

[0006] In order to solve the problems existing in the prior art, the purpose of this application is to provide a prognostic decision system for unresectable hepatocellular carcinoma. Multiple prognostic prediction models are trained through multimodal cohort data of clinical texts and CECT images of uHCC patients, and a modular decision support model is constructed using decision units and multiple prognostic prediction models. The optimal treatment decision and prognosis survival prediction under multiple treatment options can be given.

[0007] Specifically, this application involves the following aspects:

[0008] This application proposes a prognostic decision system for unresectable hepatocellular carcinoma, which includes the following functions:

[0009] A sample data preprocessing module is used to obtain clinical text variables of patients with unresectable hepatocellular carcinoma, and convert the clinical text variables into text input tag sequences through a text embedding unit; and to collect preoperative CT enhanced images of patients with unresectable hepatocellular carcinoma, segment the preoperative CT enhanced images into arterial phase CT enhanced images and portal phase CT enhanced images, and convert the arterial phase CT enhanced images and portal phase CT enhanced images into image block input tag sequences through an image embedding unit;

[0010] A prognosis prediction model building module inputs the text input tag sequence and the image block input tag sequence into a deep learning model to capture the global relationship between the text input tag sequence and the image block input tag sequence, and obtains a prognosis prediction model based on the deep learning model to predict the patient survival score;

[0011] The decision support model construction module includes a decision unit, which executes the optimal prognostic decision based on multiple prognostic prediction models; the decision support model construction module obtains multiple prognostic prediction models through training of multiple cohorts of patients with unresectable hepatocellular carcinoma, and constructs a decision support model based on the decision unit and multiple prognostic prediction models to provide the optimal prognostic decision results for patients with unresectable hepatocellular carcinoma.

[0012] According to one embodiment of the present application, converting clinical text variables into text input tag sequences by a text embedding unit includes:

[0013] The clinical text variables are input into the text embedding unit for linear mapping; the clinical text variables after linear mapping are tokenized to obtain a clinical text variable token sequence; and position embedding is added to each clinical text variable token sequence to obtain a text input token sequence.

[0014] According to one embodiment of the present application, converting the arterial phase CT enhanced image and the portal phase CT enhanced image into an image block input tag sequence by an image embedding unit includes:

[0015] The arterial phase CT enhanced image and the portal phase CT enhanced image are input into the image embedding unit, and feature maps of the arterial phase CT enhanced image and the portal phase CT enhanced image are obtained through a two-dimensional convolutional layer, each feature map contains multiple feature blocks; and position embedding is added to each feature block to obtain multiple image block input label sequences.

[0016] According to one embodiment of the present application, clinical text variables include baseline variables, laboratory variables and imaging variables; baseline variables include age, gender, medical history, ECOG grade, complications, HBV, ascites, and continuous local treatment; laboratory variables include ALB, TBIL, ALT, AST, PT, INR, PLT, and AFP; imaging variables include Barcelona Clinic Liver Cancer staging, portal vein cancer thrombus diagnosis results, tumor metastasis results, tumor size, tumor number, vascular invasion, tumor metastasis, and Child-Pugh grade.

[0017] According to one embodiment of the present application, a deep learning model is composed of m first neural network units and n second neural network units; wherein each text input tag sequence and image block input tag sequence is input into the first first neural network unit and respectively processed by the first normalization layer, the first neural network layer, the second normalization layer and the feedforward neural network, and enters the next first neural network unit for processing until it is processed by the mth first neural network unit; the text input tag sequence and the image block input tag sequence processed by the m first neural network units are subjected to unified tag processing to obtain a multimodal unified tag sequence that integrates the text input tag sequence information and the image block input tag sequence information.

[0018] According to one embodiment of the present application, the multimodal unified label sequence is further input into the first second neural network unit and processed through the third normalization layer, the second neural network layer, the fourth normalization layer and the feedforward neural network, and enters the next second neural network unit for processing until it is processed by the nth second neural network unit; the multimodal unified label sequence processed by the n second neural network units is passed through the classification unit and converted into a prognostic survival score for patients with unresectable hepatocellular carcinoma; the difference between the prognostic survival score of patients with unresectable hepatocellular carcinoma and the true prognostic survival score of patients with unresectable hepatocellular carcinoma is measured by a loss function, and the loss function is minimized to update the first neural network unit and the second neural network unit until the loss value reaches a specified level and a prognostic prediction model is obtained.

[0019] According to one embodiment of the present application, multiple prognosis prediction models are obtained by training multiple cohorts of patients with unresectable hepatocellular carcinoma, including:

[0020] The first prognostic prediction model is obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with HAIC; the second prognostic prediction model is obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with TACE; the third prognostic prediction model is obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with HAIC combined therapy; and the fourth prognostic prediction model is obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with TACE combined therapy. Further comprising:

[0021] The HAIC survival score is obtained through the first prognostic prediction model; the first TACE survival score and the second TACE survival score are obtained through the second prognostic prediction model; the first HC survival score and the second HC survival score are obtained through the third prognostic prediction model; the first TC survival score and the second TC survival score are obtained through the fourth prognostic prediction model.

[0022] According to one embodiment of the present application, the decision unit performs the optimal prognosis decision based on multiple prognosis prediction models, including the following steps:

[0023] HAIC survival score prediction for patients with unresectable hepatocellular carcinoma; in response to HAIC survival score lower than MSS1, a first TACE survival score prediction for patients with unresectable hepatocellular carcinoma; in response to HAIC survival score not lower than MSS1, a second TACE survival score prediction for patients with unresectable hepatocellular carcinoma; in response to the first TACE survival score lower than MSS2, a first HC survival score prediction for patients with unresectable hepatocellular carcinoma; in response to the second TACE survival score not lower than MSS2, a first TC survival score prediction for patients with unresectable hepatocellular carcinoma; in response to the second TACE survival score lower than MSS2, a second HC survival score prediction for patients with unresectable hepatocellular carcinoma; in response to the first HC In response to a survival score of no less than MSS3, patients with unresectable HCC were switched to HC; in response to a first HC survival score less than MSS3, patients with unresectable HCC were switched to HAIC; in response to a first TC survival score of no less than MSS4, patients with unresectable HCC were switched to TC; in response to a first TC survival score less than MSS4, patients with unresectable HCC were switched to TACE; in response to a second HC survival score of no less than MSS3, patients with unresectable HCC were switched to HC; in response to a second HC survival score less than MSS3, patients with unresectable HCC were predicted by a second TC survival score; and in response to a second TC survival score of no less than MSS4, patients with unresectable HCC were switched to TC.

[0024] According to one embodiment of the present application, MSS1 is the median HAIC survival score of the first prognostic prediction model training set; MSS2 is the median TACE survival score of the second prognostic prediction model training set; MSS3 is the median HC survival score of the third prognostic prediction model training set; MSS4 is the median TC survival score of the fourth prognostic prediction model training set.

[0025] According to one embodiment of the present application, HC is HAIC combination therapy, which includes HAIC treatment combined with TKI (tyrosine kinase inhibitor) treatment, ICI (immune checkpoint inhibitor) treatment and / or local treatment; TC is TACE combination therapy, which includes TACE treatment combined with TKI (tyrosine kinase inhibitor) treatment, ICI (immune checkpoint inhibitor) treatment and / or local treatment.

[0026] According to one embodiment of the present application, the HAIC survival score, the first TACE survival score, and the second TACE survival score are the objective response rate (ORR) of patients with unresectable hepatocellular carcinoma; and the first HC survival score, the second HC survival score, the first TC survival score, and the second TC survival score are the one-year overall survival (OS) of patients with unresectable hepatocellular carcinoma.

[0027] Beneficial Effects

[0028] The prognostic decision system for unresectable hepatocellular carcinoma described in this application has the following technical effects:

[0029] 1. The uHCC prognosis decision system based on deep learning proposed in this application aims at the limitations and improvement needs of the decisions and prognoses provided by the existing prediction models, and provides a prognosis prediction model that can globally learn the features of medical images and clinical texts, and learn the treatment prognosis feature representation of multimodal medical clinical data. The prognosis prediction model represents the complex relationship between the patient's baseline characteristics, clinical data, treatment regimen and prognosis results in the form of a neural network, thereby providing an artificial intelligence-type medical atlas in combination with the decision-making unit. In this way, the decision support model can comprehensively analyze and utilize this multidimensional information, provide patients with a variety of prognosis prediction results, and provide the best treatment decision based on this.

[0030] 2. The prognosis prediction model proposed in this application learns the global relationship in multimodal clinical data to extract the comprehensive information of clinical structured text, unstructured text and CECT image blocks, and can analyze the rich prognostic feature representations in the patient cohort. Therefore, it has better population generalization, lower method and computational complexity, and ensures high accuracy. In addition to accurately providing personalized combined treatment or single treatment decisions for uHCC patients among the four treatment options, the decision support model can also recommend further conservative treatment for potential patients whose prognosis is not ideal under multiple treatment options, avoid treatment deviations, alleviate patient pain, and reduce ineffective economic expenditures. The application of the decision-making system of this application will help provide patients with personalized treatment decisions and provide strong support for helping patients achieve transformative treatment or further achieve rehabilitation.

[0031] 3. The decision-making system proposed in this application can also help researchers or medical personnel determine the treatment and hospitalization plans for early uHCC patients, thereby rationally regulating the use of medical resources. For example, it can help medical institutions determine the purchase plan of medicines and medical devices based on recent patient cohorts, coordinate the allocation of hospitalization beds in departments based on patient survival scores, etc., to provide feasible solutions for making full use of limited medical resources. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] Figure 1 The figure shows a schematic diagram of the training process of the prognosis prediction model according to an embodiment of the present application accepting uHCC patient cohort data.

[0033] Figure 2 The figure shows a schematic diagram of the structure of the first neural network unit of the prognosis prediction model according to an embodiment of the present application.

[0034] Figure 3The figure shows a flow chart of constructing a decision support model using four prognosis prediction models according to an embodiment of the present application.

[0035] Figure 4 The figure illustrates a flow chart for providing treatment decisions and prognosis predictions according to a decision support model according to an embodiment of the present application.

[0036] Figure 5A The figure shows a schematic diagram of the performance comparison of HAIC and HC prognosis survival score prediction provided by the decision support model according to an embodiment of the present application in the training set and the validation set of the patient cohort.

[0037] Figure 5B The figure shows a performance comparison diagram of TACE and TC prognosis survival score prediction provided by the decision support model according to an embodiment of the present application in the training set and the validation set of the patient cohort. DETAILED DESCRIPTION

[0038] The present application is further described below in conjunction with examples. It should be understood that the examples are only used to further describe and illustrate the present application and are not used to limit the present application.

[0039] Unless otherwise defined, the technical and scientific terms in this specification have the same meaning as those generally understood by those skilled in the art. Although similar or identical methods and materials as described herein may be used in experiments or practical applications, materials and methods are described herein below. In the event of a conflict, the present specification including the definitions therein shall prevail. In addition, materials, methods and examples are provided for illustration only and are not restrictive. The present application is further described below in conjunction with specific embodiments, but is not intended to limit the scope of the present application.

[0040] definition

[0041] Hepatic arterial infusion chemotherapy

[0042] Hepatic arterial infusion chemotherapy (HAIC) is a type of intra-arterial interventional therapy (IAIT), which is a method of directly infusing chemotherapy drugs into the liver through the hepatic artery to achieve the purpose of treatment. Specifically, HAIC can be used to treat diseases such as uHCC. Its advantage is that there is a higher blood drug concentration in the hepatic artery, which enables liver cancer tissue to absorb more chemotherapy drugs than systemic chemotherapy, thereby improving the efficacy of liver cancer. For a period of time after treatment, uHCC patients may need to follow specific nursing and rehabilitation instructions to ensure the effectiveness of treatment and reduce the occurrence of complications.

[0043] Transcatheter hepatic arterial chemoembolization

[0044] Transarterial chemoembolization (TACE) is another interventional treatment for liver cancer in intra-arterial interventional therapy (IAIT). Its principle is to deliver chemotherapy drugs and embolic agents directly to the blood supply artery of the liver tumor through catheter technology, so as to achieve the purpose of local high-concentration drug administration and block the blood supply of the tumor. However, TACE is not suitable for all liver cancer patients, especially in clinical reports in Asia, which show that it has large individual differences. Therefore, when deciding whether to perform TACE treatment, it is necessary to comprehensively consider factors such as the patient's basic characteristics, disease progression, liver function status and possible complications.

[0045] Objective response rate

[0046] Objective Response Rate (ORR) is a core indicator used to evaluate the efficacy of tumor treatment. Specifically, ORR refers to the proportion of patients whose tumor volume shrinks to a predetermined value and can be maintained for a certain period of time during tumor treatment. This indicator directly measures the anti-tumor activity of the drug on the tumor and is an important tool for evaluating efficacy, especially in early clinical trials.

[0047] Overall survival

[0048] Overall Survival (OS) is a key indicator for evaluating the efficacy of anti-tumor drugs. Generally, if a drug can significantly prolong the patient's OS, then the drug is considered to have good efficacy. In clinical practice, OS data one year after surgery / treatment is considered to be able to effectively reflect the efficacy and is of great significance in guiding the formulation of treatment decisions. Doctors often determine the appropriate treatment plan based on the patient's one-year OS expectation.

[0049] This application proposes a prognostic decision system for unresectable hepatocellular carcinoma (uHCC), which may include the following functions:

[0050] A sample data preprocessing module that obtains clinical text variables of patients with unresectable hepatocellular carcinoma and converts them into text input token sequences; and

[0051] Collect preoperative CT enhanced images of patients with unresectable hepatocellular carcinoma, segment the preoperative CT enhanced images into arterial phase CT enhanced images and portal phase CT enhanced images and convert them into image block input label sequences;

[0052] A prognosis prediction model construction module inputs a text input tag sequence and an image block input tag sequence into a deep learning model to capture the global relationship between the text input tag sequence and the image block input tag sequence, thereby obtaining a prognosis prediction model based on a deep learning model;

[0053] A decision support model construction module includes a decision unit, which executes the optimal prognostic decision based on multiple prognostic prediction models; the decision support model construction module obtains multiple prognostic prediction models through training of multiple cohorts of patients with unresectable hepatocellular carcinoma, and constructs a decision support model based on the decision unit and multiple prognostic prediction models to provide the optimal prognostic decision results for patients with unresectable hepatocellular carcinoma.

[0054] In this application, the preoperative CT enhanced (Contrast-Enhanced Computed Tomography, hereinafter referred to as CECT) images of uHCC patients increase the contrast between tissues by intravenously injecting contrast agents into patients, thereby more clearly showing the pathological conditions of blood vessels and tissue structures. CECT of liver blood vessels has different manifestations in different phases. HCC will show significant enhancement in the arterial phase, which is manifested as the development of a rich blood supply tumor characterized by arterial blood supply. In the portal phase, the contrast agent is widely distributed in the portal system of the liver, and the enhancement characteristics of HCC lesions can be evaluated in the portal phase, such as tumor size, morphology, growth characteristics, and whether there is distant metastasis. Therefore, enhanced images in the arterial and portal phases can provide detailed information on tumor blood supply and surrounding tissues, which can be used to more comprehensively evaluate the characteristics and progression of liver cancer. By combining multiple images and providing them to the prognosis prediction model to learn the rich pathological information therein, the model can better combine clinical text mining with high-precision treatment plans.

[0055] An automatic segmentation method is used to divide CECT images and obtain arterial phase and portal phase CECT images to help the prognosis prediction model further identify and quantify the diseased tissue in the two phases and improve the quality of the image data input into the prognosis prediction model. Common segmentation methods can be used to segment the uHCC patient images of this application, for example, using an nnUNet model pre-trained with liver tumor images for automatic segmentation. Vascular enhancement has different manifestations in different phases, and the images of the two phases can provide detailed information on the tumor blood supply and surrounding tissues, which helps to more comprehensively evaluate the characteristics and progression of liver cancer. This combination of multi-phase liver vascular images helps to improve the diagnostic and prediction accuracy of the model.

[0056] In particular, before segmenting the CECT images, the CECT images can also be preprocessed. The preprocessing methods include interpolation, denoising, voxel intensity quantization, etc., to improve image quality and useful information and improve the reliability of the segmentation algorithm.

[0057] In the present application, the sample data preprocessing module converts clinical text variables into text input tag sequences, which may include:

[0058] The clinical text variables are converted using a text embedding method, which may include the following steps: linear mapping of the clinical text variables; tokenization of the clinical text variables after linear mapping to obtain a clinical text variable token sequence; and adding position embedding to each clinical text variable token sequence to obtain a text input token sequence. The clinical text variables are converted into word vectors by linear mapping of the input clinical text variables, and the word vectors are segmented to obtain multiple token sequences. It should be noted that the clinical text variables used in this application may include structured text variables and unstructured text variables, so the unstructured text variables can be segmented into multiple word units and processed together with the structured text variables.

[0059] In particular, the position embedding process is performed on the clinical text variable tag sequence, the purpose of which is to encode the position information of words and word units in the tag sequence into a vector representation, so that the tag sequence has complete position information. In this way, the prognosis prediction model can capture the semantic relationship between words in the text input tag sequence, such as the relationship between patient baseline characteristics, laboratory indicators and imaging diagnosis.

[0060] In the present application, clinical text variables may include baseline variables, laboratory variables, and imaging variables. Specifically, baseline variables may include: age, gender, medical history, ECOG grade, complications, HBV, ascites, and continuous local treatment. Laboratory variables may include: ALB, TBIL, ALT, AST, PT, INR, PLT, and AFP. Imaging variables may include: Barcelona Clinical Liver Cancer Staging, portal vein tumor thrombus diagnosis results, tumor metastasis results, tumor size, tumor number, vascular invasion, tumor metastasis, and Child-Pugh grade.

[0061] In the present application, the sample data preprocessing module converts the arterial phase CT enhanced image and the portal phase CT enhanced image into an image block input tag sequence, which may include:

[0062] The method of converting the arterial phase CT enhanced image and the portal phase CT enhanced image by using the image embedding method may include the following steps: inputting the arterial phase CT enhanced image and the portal phase CT enhanced image into a two-dimensional convolution layer to obtain feature maps of the arterial phase CT enhanced image and the portal phase CT enhanced image, each feature map including a plurality of feature blocks; adding position embedding to each feature block to obtain an image block input tag sequence. Specifically, the sliding convolution kernel of the two-dimensional convolution layer is used to extract the local features of the arterial phase CECT image and the portal phase CECT image of the patient and form a feature map, each feature map is further divided to include a plurality of feature blocks (patches).

[0063] In particular, since the prognosis prediction model cannot directly understand the two-dimensional spatial structural information between feature blocks, it is also possible to first use label embedding, such as a linear layer to convert the feature block into an embedding vector representation; and use position embedding to add its position information to the embedding vector of the feature block to provide the prognosis prediction model with the two-dimensional spatial position information of the feature block. Therefore, multiple feature block embedding vectors with position embedding added can be further merged into a sequence, that is, the image block input label sequence as the input of the prognosis prediction model.

[0064] In this way, the image block input tag sequence contains the local features of its corresponding arterial phase CECT image and portal phase CECT image. The prognosis prediction model can use the text input tag sequence and the image block input tag sequence as a unified tag sequence, thereby mining the complex relationship between text and image modality data, improving the accuracy of the prognosis prediction model, and avoiding model overfitting.

[0065] In particular, after adding position embeddings for each clinical variable label sequence, the method may further include: adding a random deactivation layer to randomly discard part of the text input label sequence weights in each training of the prognosis prediction model; and after adding position embeddings for each feature block, the method may further include adding a random deactivation layer to randomly discard part of the image block input label sequence weights in each training of the prognosis prediction model. In this way, when the prognosis prediction model uses the clinical variable label sequence and the image block input label sequence as input for training, by randomly discarding part of the label sequence, its dependence on the specific features of the part of the label sequence can be reduced, and the number of effective parameters updated during training is reduced, thereby improving the generalization performance of the prognosis prediction model in the external validation set and further avoiding overfitting.

[0066] In the present application, the prognosis prediction model is obtained based on deep learning model training. The deep learning model constructed in the present application can be composed of m first neural network units and n second neural network units. In one embodiment, the first neural network unit is a transformer-based multimodal input neural network. Among them, each text input tag sequence and image block input tag sequence are respectively input into the first first neural network unit, and pass through the first normalization layer, the first neural network layer, the second normalization layer and the feedforward neural network respectively, and enter the next first neural network unit for processing until it is processed by the last first neural network unit. Figure 1 An example of a deep learning model consisting of 2 first neural network units and 10 second neural network units is given.

[0067] In a first neural network unit, the text input tag sequence and the image block input tag sequence are respectively input into the first neural network layer after passing through a first normalization layer. In the first neural network layer, the deep learning model learns the unimodal representations in the two input tag sequences respectively, and also learns the multimodal representations between the two input tag sequences. Figure 2 The structure and learning path of the first neural network layer are illustrated, where Q, K, and V represent the first, second, and third vectors of the text input tag sequence / image block input tag sequence, respectively, Matrix.Mul represents matrix multiplication, and Scaled Softmax represents the scaled Softmax function.

[0068] In one embodiment, the first neural network layer includes: a vector generation layer, a similarity calculation layer, a weight layer and a weighted layer, wherein the vector generation layer converts each text input tag sequence and each image block input tag sequence into a first vector, a second vector and a third vector. The similarity calculation layer obtains and calculates the similarity value between the first vector of the text input tag sequence / image block input tag sequence and its second vector and the second vector of the image block input tag sequence / text input tag sequence; wherein the similarity value calculation method may include calculating the cosine similarity, Euclidean distance, Pearson correlation coefficient, etc. between the first vector and the second vector.

[0069] Furthermore, the weight layer obtains the weight value of the text input tag sequence / image block input tag sequence corresponding to each first vector based on the similarity value; the weighted layer weightedly sums the third vector of the text input tag sequence and image block input tag sequence of each first vector corresponding to the weight value based on the obtained weight value, and outputs the text input tag sequence and image block input tag sequence with different weight values, so that each text input tag sequence and image block input tag sequence has a different contribution to the first neural network unit.

[0070] Therefore, in the first neural network layer, each text input tag in the text input tag sequence not only pays attention to the similarity of its relationship with other text input tags, but also pays attention to the similarity of its relationship with other image block input tags; similarly, each image block input tag also pays attention to its relationship with other image block input tags and text input tags. Therefore, the first neural network layer can help the deep learning model obtain a comprehensive multimodal representation, which enables the prognosis prediction model obtained after training to capture the intricate relationship between the patient's clinical variables and medical images, thereby improving its prediction accuracy in complex diagnosis and treatment environments.

[0071] In the present application, the use of the first neural network unit is to help the model better learn the connections between different modalities, such as the complex intrinsic correlations between image lesions and laboratory parameters, baseline characteristics and diagnostic results, etc. The present application first uses the first neural network unit to perform interactive learning between input data of different modalities, which cannot be accomplished by the second neural network unit. The labeled output with complex multimodal features obtained by the first neural network unit is then input into the second neural network unit, which can provide better model generalization and medical data comprehension capabilities for the prognosis prediction model of the present application.

[0072] Furthermore, the text input tag sequence and the image block input tag sequence processed by multiple first neural network units are uniformly labeled to obtain a multimodal unified tag that integrates the text input tag sequence information and the image block input tag sequence information. After the text input tag sequence and the image block input tag sequence pass through the first neural network unit, each tag therein contains information of all tags with different weights. During the training process of the deep learning model, these weights will be updated with the first neural network unit, reflecting the role and dependency between multimodal medical data. Therefore, the two tag sequences can be unified, and the attention mechanism can be used to further learn the feature representations therein. For example, multiple second neural network layers can be used to further obtain the global representation of the multimodal unified tag sequence.

[0073] Specifically, in the present application, each multimodal unified marker sequence can be further input into the first second neural network unit and processed through the third normalization layer, the second neural network layer, the fourth normalization layer and the feedforward neural network therein, and enter the next second neural network unit for processing until it is processed through the last second neural network unit. Finally, the multimodal unified marker sequence processed by multiple second neural network units can be input into a classification unit to convert the marker sequence into a prognostic survival score as an output. Specifically, the classification unit may include multiple feedforward neural networks, such as a multilayer perceptron, and include using a classifier, such as the Sigmoid activation function of logistic regression to convert the marker sequence into a prognostic prediction survival score for patients with unresectable hepatocellular carcinoma.

[0074] In the present application, the distance between the prognosis prediction survival score of patients with unresectable hepatocellular carcinoma and the actual prognosis survival score of patients with unresectable hepatocellular carcinoma can be measured by a loss function, and the loss function can be minimized by minimizing the distance to update the first neural network unit and the second neural network unit until the loss value reaches a specified level and a prognosis prediction model is obtained. For example, the label weights in the first neural network unit and the second neural network unit can be updated by a cross entropy loss function to obtain a prognosis prediction model.

[0075] In the present application, the decision support model construction module can construct a decision support model based on the decision unit and multiple prognosis prediction models. Among them, the multiple prognosis prediction models can include a first prognosis prediction model obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with HAIC; a second prognosis prediction model obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with TACE; a third prognosis prediction model obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with HAIC combined therapy (hereinafter referred to as HC); and a fourth prognosis prediction model obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with TACE combined therapy (hereinafter referred to as TC).

[0076] It should be noted that the four prognostic prediction models can be set to output HAIC survival score, TACE survival score, HC survival score and TC survival score respectively. Among them, the decision support model can predict the TACE survival score of uHCC patients multiple times, such as predicting the first TACE survival score and the second TACE survival score; similarly, the decision support model can also predict the HC survival score and TC survival score multiple times, such as predicting the first HC survival score, the second HC survival score and the first TC survival score, the second TC survival score. The uHCC patient survival scores predicted by the decision support model may include HAIC ORR, TACE ORR, HC one-year OS and TC one-year OS, such as Figure 3 As shown, they are used as the basis for decision-making prediction of treatment options for uHCC patients.

[0077] In this application, the decision support model's decision unit can be based on the four prognosis prediction models to perform the following steps: Figure 4 The following example shows the steps of optimal prognostic decision making to obtain the optimal prognostic decision results for uHCC patients:

[0078] HAIC score prediction of survival in patients with unresectable hepatocellular carcinoma;

[0079] In response to the HAIC survival score being lower than MSS1, a first TACE survival score prediction was performed for patients with unresectable hepatocellular carcinoma; in response to the HAIC survival score being not lower than MSS1, a second TACE survival score prediction was performed for patients with unresectable hepatocellular carcinoma;

[0080] In response to a first TACE survival score lower than MSS2, HAIC was performed in patients with unresectable HCC and the first HC survival score was simultaneously predicted;

[0081] In response to the second TACE survival score being no less than MSS2, TACE is performed on patients with unresectable HCC and the first TC survival score is predicted at the same time; in response to the second TACE survival score being less than MSS2, the second HC survival score is predicted on patients with unresectable HCC;

[0082] In response to the first HC survival score being no less than MSS3, patients with unresectable HCC were switched to HC; in response to the first HC survival score being less than MSS3, patients with unresectable HCC were switched to HAIC;

[0083] In response to the first TC survival score being no less than MSS4, patients with unresectable HCC were switched to TC; in response to the first TC survival score being less than MSS4, patients with unresectable HCC were switched to TACE;

[0084] In response to the second HC survival score being no less than MSS3, HC is performed on patients with unresectable HCC; in response to the second HC survival score being less than MSS3, a second TC survival score is predicted for patients with unresectable HCC; and

[0085] TC was performed for patients with unresectable HCC in response to a second TC survival score no less than MSS4.

[0086] In the present application, MSS1 is the median HAIC survival score of the first prognostic prediction model training set; MSS2 is the median TACE survival score of the second prognostic prediction model training set; MSS3 is the median HC survival score of the third prognostic prediction model training set; MSS4 is the median TC survival score of the fourth prognostic prediction model training set.

[0087] In particular, the prognosis prediction model can also output the risk group of the patient under different treatment methods. Specifically, the decision-making system of the present application can use the prognosis prediction model to classify patients with HAIC survival scores not less than MSS1, TACE survival scores not less than MSS2, first / second HC survival scores not less than MSS3, and first / second TC survival scores not less than MSS4 as low-risk patient groups, and patients with opposite survival scores as high-risk patient groups. The decision unit can further use the classification results to determine the patient's optimal treatment prognosis. For patients whose predicted results for all treatment methods are in the high-risk patient group, the decision support model can provide the degree of risk caused by the treatment to avoid serious problems.

[0088] In this application, HC is HAIC combination therapy, which may include one or more of HAIC combined with TKI therapy, ICI therapy and local therapy; TC is TACE combination therapy, which may include one or more of TACE combined with TKI therapy, ICI therapy and local therapy. TKI and ICI are usually used as part of combination therapy in combination with other major treatment methods, such as major radiotherapy, chemotherapy, etc., to improve the treatment effect and prolong the patient's survival. The two main treatment methods of HAIC and TACE combined with the above-mentioned anticancer drugs and local therapy can effectively improve the prognosis. However, the drug tolerance in the patient population is highly heterogeneous, and the efficacy of combination therapy varies from person to person.

[0089] In the overall uHCC population, HAIC combined therapy and TACE combined therapy have been proven to have statistical value in improving patient survival. Therefore, when the clinical prognosis of a single treatment is poor, different forms of combined therapy will be considered. However, some patient groups still do not have survival benefits under combined therapy, and even have side effects. The decision support model of this application can capture the optimal path for patients under multiple treatment methods, which means that it can directly determine the treatment plan without trying them one by one, avoiding ineffective treatments and reducing unnecessary drug side effects and the economic burden on patients. For example, when a patient is not suitable for any of the single treatments, the decision support model finds that the combined treatment page cannot produce a significant effect on it. The decision unit may not recommend any treatment plan, reminding the patient not to expect subsequent targeted immunotherapy or molecular targeted drugs to avoid potential economic losses and complications; on this basis, the decision support model can further propose to the patient that a more active treatment plan or prevention plan should be implemented, emphasizing that the conservative treatment plan provided by the model may lead to an increased risk of death, and the monitoring and auxiliary maintenance system needs to be further strengthened.

[0090] Example

[0091] Example 1 Patient cohort design and data preprocessing

[0092] Between January 2014 and November 2021, a total of 4773 consecutive uHCC patients who underwent initial TACE or HAIC treatment in twelve tertiary hospitals were reviewed. 1448 registered review patients from three of the hospitals were selected and assigned to the proposed training set and internal validation set for training and optimizing the prognosis prediction model; 277 patients from the other nine hospitals were selected and assigned to the external validation set. All patients were diagnosed according to the European Association for the Study of the Liver (EASL) and American Association for the Study of Liver Diseases (AASLD) guidelines1,2, that is, they were considered to have uHCC if they had multiple intrahepatic metastases, distant metastases, or a tumor size greater than 10 cm. In addition, some suspected cases were confirmed by image-guided needle biopsy.

[0093] Specifically, all 1725 uHCC patients (mean age (52.8±11.5) years; 1529 males) underwent preoperative CECT scans, were deemed eligible for IAIT and were included in the cohort. First, the cohort was divided into two groups according to the interventional treatment method: TACE group (816 people) and HAIC group (909 people). In these two groups, further secondary division was performed according to the combined systemic therapy received by the patients, so that TACE and HAIC who received systemic therapy were divided into two groups, resulting in TC group (314 people) and HC group (443 people). Among these patients, they were randomly divided into training group and internal validation group at a ratio of 70% / 30%. The remaining patients were assigned to the external validation set. The prognostic prediction models corresponding to different cohorts were intensively trained using the training set to fine-tune the model parameters, which helps to enhance the ability of each prognostic prediction model to discern subtle changes in complex patient heterogeneity, in order to improve the accuracy of the model.

[0094] To ensure the uniformity and reliability of the image segmentation algorithm and the prognostic prediction model, several preprocessing steps were implemented on the patient CECT images. Specifically, one-dimensional spline interpolation was performed on all CECT images, and the slice thickness was standardized to 1 mm. In addition, the voxel intensity was limited to the range of -50 to 350 HU to eliminate non-liver and non-tumor tissues, thereby improving the sensitivity and accuracy of subsequent segmentation for subsequent analysis.

[0095] In this embodiment, the CECT images are finally segmented using the nnU-Net model pre-trained on the Liver and Tumor Segmentation (LITS) Challenge dataset. nnU-Net processes the arterial and portal venous phase CECT images of each patient in the cohort to provide liver and tumor contours. According to the location of the largest tumor area in the image, an informative slice (the slice containing the largest tumor area) is selected from the arterial and portal venous phase CECT images of each patient; using the bounding box covering the entire liver area of ​​the selected slice, the image is cropped to a size of 224mm*224mm and contains multiple patches as image input for the prognostic prediction model.

[0096] Example 2 Construction and performance verification of prognosis prediction models for four treatment methods

[0097] The four initial prognostic prediction models all used the preoperative CECT images and corresponding baseline parameters and clinical parameters of each patient in the four cohorts of Example 1 as input. It should be noted that in this example, the image segmentation results were further manually modified and corrected by two senior reviewers with more than ten years of expertise in the field of liver imaging to improve accuracy. A total of 3450 CECT slices from the four cohorts were used to train and validate four independent prognostic prediction models.

[0098] Each prognosis prediction model was trained using the proposed training set cohort. Specifically, the demographic baseline and quantitative clinical parameters of each patient were processed to obtain the input labels of this article, and the two pre-processed preoperative CECT images of each patient were processed to obtain the image block input label sequence. The global features of the input data were extracted by 2 first neural network units and 10 second neural network units in turn, and the dimension was reduced together with the clinical variables to obtain a unified label sequence, and the predicted patient prognosis results were output using a logistic regression model. The model was continuously trained with the reduced-dimensional features and the first neural network unit and the second neural network unit were updated using the cross-entropy loss function.

[0099] Compared with traditional CNN methods, the prognostic prediction model provides prognostic outcome predictions by uniformly processing multimodal inputs without the need for excessive fine-tuning at each data processing stage. In addition, the multimodal attention mechanism of the prognostic prediction model provides additional consideration of the interconnections between multimodal inputs at the global level. In contrast, non-uniform methods, such as traditional CNN methods, only consider the connections between global representations (i.e., features reduced during dimensionality reduction).

[0100] In this example, the trained prognostic prediction model was finally used to predict OOR and one-year OS under different uHCC treatment schemes, and its prediction performance was compared with the performance of the standard visual transformer (ViT) using only images as input, the pre-trained convolutional neural network ResNet-50 that extracts features from CECT images, and two models built based on clinical variables considered in existing clinical decision-making schemes (clinical model-1: Cox survival model; clinical model-2: logistic regression model). The prediction evaluation parameters of multiple models are shown in Table 1 below:

[0101] Table 1 Comparison of the prediction performance of four prognostic prediction models with existing prediction models

[0102]

[0103]

[0104] Among the deep learning methods, the conventional CNN method had the lowest regularization ability and the lowest AUC in the external validation set in the TC, HAIC, and HC groups (TACE: 0.695 [95% CI, 0.621, 0.762], p = 0.44; HAIC: 0.718 [95% CI 0.611, 0.809], p = 0.286; TC: 0.637 [95% CI, 0.498, 0.761], p = 0.218; HC: 0.685 [95% CI, 0.528, 0.817], p = 0.534). In the comparison of models under all four treatments, the prognostic prediction model clearly showed the highest performance value compared with other deep learning methods, especially in the external validation set. In the internal validation set and the external validation set, the AUC of ORR of the TACE prognostic prediction model was 0.788 (95% CI, 0.707, 0.856) and 0.761 (95% CI, 0.693, 0.820), respectively. In the internal validation set and the external validation set, the AUC of ORR of the HAIC prognostic prediction model was 0.854 (95% CI, 0.791, 0.904) and 0.805 (95% CI, 0.707, 0.881), respectively. In the internal validation set and the external validation set, the AUC of OS of the TC prognostic prediction model was 0.811 (95% CI, 0.677, 0.907) and 0.736 (95% CI, 0.608, 0.841), respectively. In the internal and external validation sets, the AUC of OS of the HC prognostic prediction model was 0.837 (95% CI, 0.737, 0.910) and 0.748 (95% CI, 0.599, 0.865), respectively. Finally, the DeLong test was used to verify that the performance of each prognostic prediction model was significantly better than the commonly used clinical model-1 in all validation set cohorts (P<0.05).

[0105] In addition, there was no significant difference in the ORR and OS calculated by the training set, internal validation set, and external validation set of the prognostic prediction model in the HAIC, TACE, and TC cohorts (p>0.05), indicating that the model has good generalization ability. There was a significant difference between the training set and the internal validation set of the HC cohort (p=0.019), but there was no significant difference between the training set and the external validation set, which further verified the usability of the prognostic prediction model in actual medical scenarios. In the training data sets of the four prognostic prediction models, the median survival scores (MSS) of the HAIC, TAIC, HC, and TC cohorts were 0.3368 (MSS1), 0.4035 (MSS2), 0.6458 (MSS3), and 0.6130 (MSS4), respectively. These parameters will serve as auxiliary values ​​for the decision unit to determine the optimal treatment plan.

[0106] Example 3 Decision support model construction and performance verification

[0107] The decision support model is constructed by using four independent prognostic prediction models and decision units obtained by training four cohorts in Example 2, and a comprehensive auxiliary treatment decision is provided for uHCC patients based on the survival score calculated by each prognostic prediction model.

[0108] Specifically, the survival score of the patient given by the prognostic prediction model is compared with the median survival score of the prognostic prediction model training set cohort, and the patient is divided into a high-risk group (the patient's survival score does not reach the median survival score) and a low-risk group (the patient's survival score reaches or exceeds the median survival score) according to whether the patient's survival score reaches the median survival score. In this embodiment, the decision support model further gives a decision result, that is, the patient's OOR / one-year OS. When the recommendation is consistent with the doctor's actual treatment choice, the patient's overall treatment response and one-year survival rate are significantly improved compared to when the decision support model recommendation is inconsistent with the doctor's actual treatment choice. At the same time, in the group where the decision support model does not recommend IAIT, that is, the group of patients in which any treatment plan is not recommended, the response rate and survival rate are also significantly lower than those of patients in the group with consistent decisions. Table 2, Figure 5A and Figure 5B The responses and survival benefits of the patient groups in the external validation set were compared when the predicted and true decisions were consistent.

[0109] Table 2 Actual objective response rate / one-year overall survival rate in different cohorts with consistent recommendations between the decision support model and clinicians, and actual overall response rate / one-year overall survival rate in cohorts with inconsistent or no recommendations from the decision support model

[0110]

[0111] For the TACE and HAIC groups, the ORR was significantly higher when the actual treatment strategy was consistent with the predicted recommendation of the decision support model (p<0.001). A significant difference was observed when the decision support model recommended not to use IAIT compared with the situation when the actual treatment strategy and the decision support model recommendation were consistent. The TC group also showed a similar trend, and the OS rate was significantly improved when the decision support model recommendation was consistent with the actual treatment decision. For the HC group, a larger sample size of patients is needed to prove that its prediction correctness is statistically significant.

[0112] In addition, compared with the group of patients who only received a single IAIT treatment regimen, the survival of the combined treatment group was longer, which means that in the overall population, the combined treatment provides a statistically significant improvement in the prognosis and survival of patients. This indicates that for high-risk uHCC patients, TACE or HAIC combined therapy should be the recommended first-line treatment method, which can prolong the OS of uHCC patients. Finally, for patients who are predicted to have OOR after receiving only the initial single IAIT, the decision support model can remind them that they do not need to expect the subsequent additional use of targeted immunotherapy or molecular targeted drugs, avoiding potential economic losses and complications for patients. For high-risk patients given by the prognostic prediction model and patients for whom the prognostic prediction model recommends not using any IAIT at all, the decision support model can suggest that they adjust other treatment options, postoperative prevention or monitoring strategies in a timely manner, emphasizing that due to their higher risk of death, they need intensive monitoring and other types of auxiliary systemic treatments.

[0113] In summary, the prognostic prediction model proposed in this application will become an immediately implementable prognostic diagnosis auxiliary clinical tool, which can provide risk stratification and accurate prognosis for doctors and patients. The application of the decision support model is to operate the survival score derived from the individual prognostic prediction model through the decision paradigm, which will become a promising potential method for optimizing personalized treatment plans in the prognostic management of personalized uHCC patients.

Claims

1. A prognostic decision system for unresectable hepatocellular carcinoma, the decision system comprising: The sample data preprocessing module obtains clinical text variables of patients with unresectable hepatocellular carcinoma and converts them into text input token sequences; Acquiring preoperative enhanced CT images of patients with unresectable hepatocellular carcinoma, segmenting the preoperative enhanced CT images into arterial phase enhanced CT images and portal phase enhanced CT images and converting them into image block input label sequences; A prognosis prediction model construction module inputs the text input tag sequence and the image block input tag sequence into a deep learning model to capture the global relationship between the text input tag sequence and the image block input tag sequence, and obtains a prognosis prediction model based on the deep learning model to predict the patient survival score; A decision support model construction module includes a decision unit, which executes the optimal prognostic decision based on multiple prognostic prediction models; the decision support model construction module obtains multiple prognostic prediction models through training of multiple cohorts of patients with unresectable hepatocellular carcinoma, and constructs a decision support model based on the decision unit and multiple prognostic prediction models to provide the optimal prognostic decision results for patients with unresectable hepatocellular carcinoma.

2. The decision-making system according to claim 1, wherein: The clinical text variables include baseline variables, laboratory variables and imaging variables; The baseline variables included age, sex, medical history, ECOG grade, comorbidities, HBV, ascites, and continuous local treatment; The laboratory variables include ALB, TBIL, ALT, AST, PT, INR, PLT, and AFP; The imaging variables included Barcelona Clinic Liver Cancer stage, portal vein tumor thrombus diagnosis results, tumor metastasis results, tumor size, tumor number, vascular invasion, tumor metastasis, and Child-Pugh grade.

3. The decision-making system according to claim 1, wherein: The deep learning model consists of m first neural network units and n second neural network units; Each of the text input mark sequence and the image block input mark sequence is input into the first first neural network unit and is processed by the first normalization layer, the first neural network layer, the second normalization layer and the feedforward neural network respectively, and enters the next first neural network unit for processing, until it is processed by the mth first neural network unit; The text input tag sequence and the image block input tag sequence processed by m of the first neural network units are subjected to unified tagging processing to obtain a multimodal unified tag sequence that fuses the text input tag sequence information and the image block input tag sequence information.

4. The decision system according to claim 3, wherein: The multimodal unified label sequence is further input into the first second neural network unit and processed by the third normalization layer, the second neural network layer, the fourth normalization layer and the feedforward neural network therein, and then enters the next second neural network unit for processing, until it is processed by the nth second neural network unit; The multimodal unified label sequence processed by the n second neural network units is passed through a classification unit and converted into a prognostic survival score for patients with unresectable hepatocellular carcinoma; The distance between the prognostic survival score of the patient with unresectable hepatocellular carcinoma and the true prognostic survival score of the patient with unresectable hepatocellular carcinoma is measured by a loss function, and the loss function is minimized to update the first neural network unit and the second neural network unit until the loss value reaches a specified level and the prognostic prediction model is obtained.

5. The decision system according to claim 1, wherein the plurality of prognosis prediction models obtained by training with a plurality of cohorts of patients with unresectable hepatocellular carcinoma comprises: The first prognostic prediction model was obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with HAIC; The second prognostic prediction model was obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with TACE; A third prognostic prediction model was obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with HAIC combination therapy; and The fourth prognostic prediction model was obtained by training a cohort of patients with unresectable hepatocellular carcinoma treated with TACE combined therapy.

6. The decision system according to claim 5, wherein: The HAIC survival score was obtained through the first prognostic prediction model; The first TACE survival score and the second TACE survival score were obtained through the second prognostic prediction model; The first HC survival score and the second HC survival score were obtained by the third prognostic prediction model; The first TC survival score and the second TC survival score were obtained through the fourth prognostic prediction model.

7. The decision system according to claim 6, wherein: The decision unit performs the optimal prognosis decision based on the multiple prognosis prediction models, including the following steps: The HAIC score is used to predict survival in patients with unresectable hepatocellular carcinoma; In response to the HAIC survival score being lower than MSS1, performing the first TACE survival score prediction on the patient with unresectable hepatocellular carcinoma; in response to the HAIC survival score being not lower than MSS1, performing the second TACE survival score prediction on the patient with unresectable hepatocellular carcinoma; In response to the first TACE survival score being lower than MSS2, predicting the first HC survival score for the patient with unresectable hepatocellular carcinoma; In response to the second TACE survival score being not lower than MSS2, predicting the first TC survival score for patients with unresectable hepatocellular carcinoma; in response to the second TACE survival score being lower than MSS2, predicting the second HC survival score for patients with unresectable hepatocellular carcinoma; In response to the first HC survival score being no less than MSS3, switching to HC for the patient with unresectable hepatocellular carcinoma; In response to the first HC survival score being less than MSS3, performing HAIC for the patient with unresectable hepatocellular carcinoma; In response to the first TC survival score being no less than MSS4, switching to TC for the patient with unresectable hepatocellular carcinoma; in response to the first TC survival score being less than MSS4, performing TACE for the patient with unresectable hepatocellular carcinoma; In response to the second HC survival score being not less than MSS3, performing HC on the patient with unresectable hepatocellular carcinoma; in response to the second HC survival score being less than MSS3, performing a second TC survival score prediction on the patient with unresectable hepatocellular carcinoma; and In response to the second TC survival score being no less than MSS4, TC is performed on patients with unresectable hepatocellular carcinoma.

8. The decision system according to claim 7, wherein: The MSS1 is the median of the HAIC survival score of the first prognostic prediction model training set; The MSS2 is the median of the TACE survival score of the second prognostic prediction model training set; The MSS3 is the median of the HC survival score of the third prognostic prediction model training set; The MSS4 is the median of the TC survival score of the fourth prognostic prediction model training set.

9. The decision-making system according to claim 6, wherein: The HC is a HAIC combination therapy, which includes HAIC therapy combined with TKI (tyrosine kinase inhibitor) therapy, ICI (immune checkpoint inhibitor) therapy and / or local therapy; The TC is TACE combined therapy, which includes TACE therapy combined with TKI (tyrosine kinase inhibitor) therapy, ICI (immune checkpoint inhibitor) therapy and / or local therapy.

10. The decision-making system according to claim 6, wherein: The HAIC survival score, the first TACE survival score, and the second TACE survival score are objective response rates (ORRs) in patients with unresectable hepatocellular carcinoma; and The first HC survival score, the second HC survival score, the first TC survival score, and the second TC survival score are one-year overall survival (OS) of patients with unresectable hepatocellular carcinoma.

Images