Apparatus and computer readable storage medium for identifying prostate cancer

Abstract

The application discloses a device and a computer readable storage medium for identifying prostate cancer. The device comprises a processor and a memory having stored therein program instructions for identifying prostate cancer, which, when executed by the processor, cause the device to perform the following operations: acquiring multi-modal data; inputting the multi-modal data into a trained identification model for prostate cancer identification to obtain an identification result; wherein the identification model is trained by the following operations: defining expert knowledge text based on serological indicators as prompt words; splicing the prompt words in case text information to form target text information; and inputting imaging information and the target text information into the identification model for interactive learning to obtain the trained identification model. The scheme of the application can improve the accurate judgment ability of prostate cancer and reduce complications and waste of medical resources caused by unnecessary puncture.

Description

Technical Field

[0001] This application generally relates to the field of artificial intelligence in medical technology. More specifically, this application relates to an apparatus and a computer-readable storage medium for identifying prostate cancer. Background Technology

[0002] Prostate cancer is one of the most common malignant tumors in men, with high incidence and mortality rates. Clinically, prostate biopsy is the gold standard for diagnosing prostate cancer. However, unnecessary biopsies are also very common, leading to potential complications (such as infection, bleeding, sepsis, etc.) and a waste of medical resources. Currently, in prostate cancer screening and biopsy decisions, many studies and clinical practices still rely on simple numerical indicators such as prostate-specific antigen ("PSA") levels, PSA density, and prostate imaging and data scoring ("PI-RADS") as the basis for judgment. While these methods can help identify high-risk groups to some extent, they are not precise enough in distinguishing "clinically significant prostate cancer," especially when patients are in the intermediate risk range (such as a PI-RADS score of 3 or 4). Simple numerical values ​​or single scoring models often fail to accurately predict the true condition, easily leading to over-biopsy or missed diagnosis.

[0003] With the continuous development of natural language processing technology, general-purpose large language models have shown some potential in medical text interpretation. However, existing methods rely solely on the pre-training parameters of general models (such as patient medical records, examination reports, and imaging descriptions), which often limits the models to generalizing from surface-level data. This makes it difficult to capture deeper clinical logic and expert experience, leading to significant biases in model predictions. Furthermore, expert knowledge in clinical practice is often scattered across various guidelines, research papers, and the personal experiences of senior physicians, primarily in written or oral form, lacking a clear structured presentation. This approach, which only addresses expert experience at the level of "quantitative indicators" or "template rules," often fails to address individual differences in age, comorbid symptoms, and imaging findings among actual patients. It cannot fully incorporate the details of expert judgments in different scenarios, nor does it offer flexible handling of atypical cases, making it difficult to provide ideal support for clinical decision-making.

[0004] In view of this, there is an urgent need to provide a solution for identifying prostate cancer. This solution involves defining expert knowledge text and concatenating it with case text information to obtain target text information. This allows for unified analysis of structured and unstructured data. Furthermore, by deeply integrating the target text information and imaging information into the recognition model, the model can take into account individual differences when dealing with atypical cases. This reduces the rate of missed diagnoses and misdiagnoses, improves the accuracy of prostate cancer diagnosis, and reduces complications and waste of medical resources caused by unnecessary biopsies. Summary of the Invention

[0005] In order to at least address one or more of the technical problems mentioned above, this application proposes a scheme for identifying prostate cancer in several aspects.

[0006] In a first aspect, this application provides an apparatus for identifying prostate cancer, comprising: a processor; and a memory storing program instructions for identifying prostate cancer, wherein when the program instructions are executed by the processor, the apparatus performs the following operations: acquiring multimodal data, wherein the multimodal data includes serological marker information, imaging information, and case text information of a subject; inputting the multimodal data into a trained recognition model for prostate cancer identification, and obtaining a recognition result; wherein the recognition model is trained by: defining expert knowledge text based on the serological markers as prompt words; concatenating the prompt words into the case text information to form target text information; and inputting the imaging information and the target text information into the recognition model for interactive learning to obtain a trained recognition model.

[0007] In one embodiment, the method further includes performing corresponding preprocessing operations on the serological marker information, the imaging information, and the case text information, respectively.

[0008] In another embodiment, performing corresponding preprocessing operations on the serological index information, the imaging information, and the case text information respectively includes: performing one or more preprocessing operations on the serological index information, such as outlier detection, standardization, or missing value imputation; performing one or more preprocessing operations on the imaging information, such as registration, sequence differentiation, or annotation; and performing one or more preprocessing operations on the case text information, such as text recognition, structured storage, or normalization.

[0009] In another embodiment, the serological information includes at least total prostate-specific antigen (PSA), PSA ratio, and PSA density; the imaging information includes at least one or more magnetic resonance imaging (MRI) sequences such as T1WI, T2WI, DWI, and ADC; and the case text information includes at least historical case information, symptom description information, and test record information.

[0010] In yet another embodiment, defining expert knowledge text as prompt words based on the serological indicators includes: comparing each indicator in the serological indicators with a corresponding threshold; and defining expert knowledge text as prompt words based on the comparison results.

[0011] In yet another embodiment, the method of concatenating the prompt word into the case text information to form target text information includes concatenating the prompt word before or after the case text information to form the target text information.

[0012] In another embodiment, before inputting the imaging information and the target text information into the recognition model for interactive learning, the method further includes: using the recognition model to extract the sequence features of each magnetic resonance sequence in the imaging information; and performing a fusion operation on the sequence features of each magnetic resonance sequence to obtain fused features.

[0013] In yet another embodiment, the fusion operation includes splicing, weighted averaging, or fully connected operations.

[0014] In yet another embodiment, the process of inputting the imaging information and the target text information into the recognition model for interactive learning includes: setting a self-attention layer in the recognition model; and using the self-attention layer to interactively learn the fused features and the target text.

[0015] In a second aspect, this application provides a computer-readable storage medium having stored thereon computer-readable instructions for identifying prostate cancer, which, when executed by one or more processors, perform operations performed by the apparatus as described in the various embodiments of the first aspect above.

[0016] Using the above-described scheme for identifying prostate cancer, this application defines expert knowledge text, which is then concatenated with case text information to obtain target text information. This allows for unified analysis of structured and unstructured data. By inputting the target text information and imaging information into the recognition model for deep fusion, the recognition model can take into account individual differences when facing atypical cases, thereby reducing the rate of missed diagnoses and misdiagnoses, improving the accuracy of prostate cancer diagnosis, and reducing complications and waste of medical resources caused by unnecessary biopsies. Attached Figure Description

[0017] The above and other objects, features, and advantages of exemplary embodiments of this application will become readily understood by reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of this application are illustrated by way of example and not limitation, and the same or corresponding reference numerals denote the same or corresponding parts, wherein:

[0018] Figure 1 This is an exemplary structural block diagram illustrating an apparatus for identifying prostate cancer according to an embodiment of this application;

[0019] Figure 2This is an exemplary flowchart illustrating a preprocessing operation for multimodal data according to an embodiment of this application;

[0020] Figure 3 This is an exemplary flowchart illustrating the training of a recognition model according to an embodiment of this application;

[0021] Figure 4 This is an exemplary flowchart illustrating a method for identifying prostate cancer according to an embodiment of this application;

[0022] Figure 5 This is an exemplary structural block diagram illustrating a device for identifying prostate cancer according to an embodiment of this application. Detailed Implementation

[0023] The technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the embodiments described in this specification are only some embodiments provided by this application for the purpose of facilitating a clear understanding of the solutions and complying with legal requirements, and are not all embodiments that can be implemented in this application. Based on the embodiments disclosed in this specification, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0024] Figure 1 This is an exemplary structural block diagram illustrating an apparatus 100 for identifying prostate cancer according to an embodiment of this application. Figure 1 As shown, the device 100 may include a processor 101 and a memory 102. The processor 101 may include, for example, a general-purpose processor (“CPU”) or a dedicated graphics processor (“GPU”), and the memory 102 stores program instructions executable on the processor. In some embodiments, the memory 102 may include, but is not limited to, resistive random access memory (RRAM), dynamic random access memory (DRAM), static random access memory (SRAM), and enhanced dynamic random access memory (EDRAM).

[0025] Furthermore, the aforementioned memory 102 may store program instructions for identifying prostate cancer. When the program instructions are executed by the processor, the device 100 performs the following operations: acquiring multimodal data, wherein the multimodal data includes the subject's serological marker information, imaging information, and case text information; inputting the multimodal data into a trained recognition model for prostate cancer identification; and obtaining the identification result. The recognition model is trained through the following operations: defining expert knowledge text based on serological markers as prompt words; concatenating prompt words into the case text information to form target text information; and inputting the imaging information and target text information into the recognition model for interactive learning to obtain a trained recognition model.

[0026] In some embodiments, the aforementioned serological indicators may include, but are not limited to, total prostate-specific antigen (PSA), PSA ratio, and PSA density. The aforementioned imaging information may include, but is not limited to, one or more magnetic resonance imaging (MRI) sequences such as T1WI, T2WI, DWI, and ADC. The aforementioned case text information may include, but is not limited to, historical case information, symptom descriptions, and test records (e.g., digital rectal examination results), as well as, for example, the attending physician's initial assessment.

[0027] In some embodiments, before training the recognition model, the present application further includes performing corresponding preprocessing operations on serological marker information, imaging information, and case text information, respectively. In some embodiments, one or more preprocessing operations, such as outlier detection, standardization, or missing value imputation, may be performed on the serological marker information. One or more preprocessing operations, such as registration, sequence discrimination, or annotation, may be performed on the imaging information. One or more preprocessing operations, such as text recognition, structured storage, or normalization, may be performed on the case text information.

[0028] Specifically, outlier detection is performed on serological marker information, and outliers are removed based on clinical experience or statistical distribution. Standardization of serological marker information can ensure consistent feature distribution, while missing value imputation can be achieved through methods such as interpolation or averaging. In the preprocessing of imaging information, registration or resampling can unify resolution and slice thickness, sequence differentiation is achieved by identifying T1WI, T2WI, DWI, or ADC, and lesion areas are labeled. In the preprocessing of case text information, key symptoms or diagnostic points are extracted through text recognition, structured storage is achieved by establishing mappings for field names or values, and normalization operations are performed by removing noise information or stop words.

[0029] Based on the multimodal information obtained above, including serological markers, imaging data, and medical record text, the multimodal data is input into a trained recognition model for prostate cancer identification to obtain the identification result. In some embodiments, the aforementioned recognition model can be, for example, a Transformer-based visual-text fusion model. During the training of the recognition model, the device 100 first defines expert knowledge text based on serological markers as prompt words. In some embodiments, each serological marker can be compared with a corresponding threshold, and then expert knowledge text can be defined based on the comparison result as prompt words. In some implementation scenarios, expert knowledge text P_{\text{guidelines}} can be defined as prompt words by comparing with, for example, prostate cancer diagnosis and treatment guidelines, literature, or clinical expert experience, and providing corresponding annotations.

[0030] For example, in some exemplary scenarios, "when PI-RADS = 3 and PSA density ("PSAD") exceeds a certain threshold, it can be considered a potentially high risk"; "if PSA > 10 ng / mL and PI-RADS ≥ 4, puncture is strongly recommended"; "if PSAD > 0.15 ng / mL..." 2 "This indicates a higher risk." In this scenario, prompts can be generated using, for example, "IF-THEN" style or natural language descriptions, forming one or more "readable" texts (i.e., prompt words), denoted as P. guidelines .

[0031] Next, a prompt word is appended to the case text information to form the target text information. In some embodiments, the prompt word is appended before or after the case text information to form the target text information. As an example, assuming the case text information is denoted as X, the prompt word P is appended before or after X. guidelines This can form target text information X′, where X′ = Concat(P guidelines For example, [cue word: "Based on prostate cancer guidelines: special vigilance should be exercised when PSA>10..."] + [patient medical record: PSA=12, high signal on DWI...].

[0032] Furthermore, the device 100 further performs interactive learning by inputting the imaging information and target text information into the recognition model to obtain a trained recognition model. In some embodiments, before inputting the imaging information and target text information into the recognition model for interactive learning, this application embodiment may further include: using the recognition model to extract the sequence features of each magnetic resonance sequence in the imaging information, and performing a fusion operation on the sequence features of each magnetic resonance sequence to obtain fused features. This allows for consideration of differences in image sequences. In some embodiments, the aforementioned fusion operation may include splicing, weighted averaging, or fully connected operations, etc.

[0033] In some exemplary scenarios, assuming that the sequence features extracted by the recognition model from T1WI, T2WI, DWI, and ADC correspond to FT1, FT2, FDWI, and FADC, they can be fused using the following formula to obtain the fused feature F. img :

[0034] Fimg=φ(FT1,FT2,FDWI,FADC)

[0035] Here, φ(·) represents the fusion operation, which can include simple splicing, weighted averaging, or learnable fully connected layer / attention mechanism.

[0036] After obtaining the fusion features of the sequence features of each magnetic resonance imaging sequence, a self-attention layer can be set in the recognition model. The self-attention layer is used to interact and learn with the fusion features and the target text to deeply integrate the target text information and imaging information. This allows the recognition model to take into account individual differences when facing atypical cases, thereby reducing the rate of missed diagnosis and misjudgment and improving the accuracy of prostate cancer diagnosis.

[0037] Specifically, interactive learning can be represented by the following formula:

[0038] P(Cancer|X)=σ(W f F img +W t T text +b)

[0039] Among them, F img T represents the fusion characteristics of sequence features from various magnetic resonance sequences. text Let W represent the integrated vector of case text information and prompt words (i.e., the target text information X′ mentioned above), σ(·) represent the activation function of the recognition model (such as Sigmoid), and W f With W t This represents the learnable weight matrix (obtainable through a self-attention mechanism), where b is the bias term. P(Cancer|X) represents the probability value of prostate cancer predicted by the recognition model after interactive learning based on fused features and target text. Using this probability value, if the recognition result is higher than a preset threshold, the patient is deemed to have a high probability of clinically significant prostate cancer, and further biopsy is recommended; if the probability value is low, temporary follow-up observation is recommended.

[0040] As described above, this application's embodiments integrate multimodal data (serological indicators, imaging information, and case text information) into a recognition model for prostate cancer identification. Specifically, by defining expert knowledge text and concatenating it with case text information to obtain target text information, the target text information incorporating expert knowledge and imaging information are deeply fused to train the recognition model. Based on this, by uniformly parsing structured and unstructured data and injecting clinicians' experience and guidelines into the training of the recognition model (i.e., through deep fusion of a large multimodal model and expert prior knowledge), the model can make more discriminative judgments for intermediate risk groups such as PI-RADS 3 and 4. This not only strengthens the mining of multimodal data itself but also allows expert knowledge to be reflected in the handling of atypical cases. This enables the recognition model to consider individual differences when facing atypical cases, reducing the rate of missed diagnoses and misdiagnoses, thereby balancing the reduction of missed diagnoses with the prevention of overtreatment. This achieves accurate prediction of prostate biopsy-related cancer, reducing complications and waste of medical resources caused by unnecessary biopsies.

[0041] Figure 2 This is an exemplary flowchart illustrating preprocessing operations on multimodal data according to embodiments of this application. Figure 2 As shown, in step S201, the subject's serological marker information, imaging information, and case text information are acquired. As previously mentioned, before training the recognition model, corresponding preprocessing operations can be performed on the serological marker information, imaging information, and case text information respectively. For example, in step S202, outlier detection, standardization, or missing value imputation is performed on the serological marker information. In step S203, registration, sequence discrimination, or annotation is performed on the imaging information. In step S204, text recognition, structured storage, or normalization is performed on the case text information to obtain the preprocessed data in step S205. This not only facilitates input into the recognition model but also improves the accuracy of subsequent recognition.

[0042] Figure 3 This is an exemplary flowchart illustrating the training of a recognition model according to embodiments of this application. Figure 3As shown, in steps S301, S302, and S303, the subject's serological marker information, imaging information, and case text information are acquired, respectively. Next, in step S304, expert knowledge text is defined based on the serological markers to serve as prompts. In some embodiments, each indicator in the serological markers can be compared with a corresponding threshold based on, for example, prostate cancer treatment guidelines, literature, or clinical expert experience, and experts can provide corresponding annotations as prompts. In step S305, the prompts are concatenated into the case text information to form target text information. Specifically, the prompts can be concatenated before or after the case text information to form the target text information.

[0043] Further, in step S306, the imaging information is input into the recognition model so that in step S307, the recognition model can first extract the sequence features of each magnetic resonance sequence in the imaging information. Then, in step S308, the sequence features of each magnetic resonance sequence are fused to obtain fused features. The fusion operation may include concatenation, weighted averaging, or fully connected operations. Based on the obtained fused features, in step S309, the fused features and the target text are interactively learned to further fine-tune the recognition model, resulting in a trained recognition model. For more details on training the recognition model, please refer to the aforementioned... Figure 1 The description of the subject matter will not be repeated here.

[0044] Figure 4 This is an exemplary flowchart illustrating a method for identifying prostate cancer according to an embodiment of this application. Figure 4 As shown, in step S401, multimodal data such as serological marker information, imaging information, and case text information are acquired. In step S402, the multimodal data is input into the trained recognition model for prostate cancer recognition to obtain the recognition result. Further details regarding the training of the recognition model can be found in the aforementioned... Figure 1 The description of the subject matter will not be repeated here.

[0045] In some implementation scenarios, this application embodiment can also include clinical deployment and iterative optimization. On the one hand, the locally deployed large model can run on the hospital server, ensuring that sensitive data is not leaked; on the other hand, doctors can view the system's process and key decision prompts for analyzing patient characteristics and MRI images through a visual interface. After clinicians actually perform puncture biopsies, they can feed the actual results back to the system for continuous model learning and parameter updates, forming a closed-loop optimization. Especially for the intermediate-risk group with PI-RADS scores of 3 and 4, their actual pathological results and corresponding MRI features can be recorded in detail, allowing the model to further improve its discrimination against this population in subsequent versions. Finally, to better address individual differences and atypical cases, this application embodiment supports doctors adding personalized instructions or external knowledge to the interface. For example, for patients with other urinary tract diseases or who have undergone prostate surgery, their MRI manifestations and PSA levels may deviate from the conventional standards. In this case, experts can write relevant experience into prompts or additional modules, allowing the model to take these additional variables into account during inference, thereby reducing missed diagnoses and preventing overtreatment, and effectively improving the accuracy of clinical decision-making. Through the fusion of the aforementioned multimodal large model with expert knowledge, phased fine-tuning, and continuous iteration of local deployment, the embodiments of this application can achieve higher sensitivity and specificity in determining the necessity of prostate puncture, reduce complications, and save medical resources.

[0046] Based on the foregoing description, this application's embodiments significantly improve the accuracy of prostate cancer diagnosis by combining a multimodal large model with expert prior knowledge, reducing complications and waste of medical resources caused by unnecessary biopsies. From a clinical practice perspective, this approach introduces multi-source data fusion: high-dimensional information such as serological indicators, MRI sequences, and medical record texts, combined with in-depth analysis of PI-RADS intermediate scores (3 and 4 points), effectively identifying patients with truly significant clinical risk of prostate cancer. Simultaneously, through specific prompt word fine-tuning and expert knowledge injection, the model can consider individual differences when facing atypical cases, thereby reducing the rate of missed diagnoses and misdiagnoses. In actual deployment, running on an in-hospital server not only ensures data security and patient privacy but also allows the model to continuously iterate and optimize based on clinical feedback, forming a closed-loop learning mechanism. For medical personnel, the visual interface of this application's embodiments helps doctors review core indicators and the model's recommendation rationale, increasing their trust in and interpretability of artificial intelligence decisions. For patients, reducing unnecessary biopsies effectively lowers the risk of complications such as infection and bleeding, improving the overall medical experience. Ultimately, through multimodal data integration, embedding of domain expert knowledge, and iterative large model technology, the embodiments of this application have higher accuracy and feasibility in prostate cancer screening and biopsy management, providing a forward-looking artificial intelligence-assisted tool for clinical diagnosis and treatment decision-making.

[0047] Optionally or additionally, embodiments of this application may also directly use traditional statistical or machine learning models, such as those based on logistic regression or multilayer perceptrons, supplemented by a pre-built rule engine (IF-THEN form). By encoding expert knowledge offline, a high clinical risk is indicated when certain triggering conditions are identified. Optionally or additionally, embodiments of this application may also utilize a retrieval-augmented model. A small knowledge base or knowledge graph is first built, containing information such as indications for prostate biopsy and PI-RADS scoring strategies. When the model processes medical records or image descriptions, it first retrieves the most relevant expert entries and then inputs them along with the original text into the recognition model to obtain outputs that better fit the guidelines. Optionally or additionally, embodiments of this application may also introduce a visual annotation system. For example, local feature information is extracted multiple times from the same MRI or medical record and aligned with domain expert opinions. This allows the model to learn how to associate the mapping relationship between "specific lesion location" and "specific rules" during the fine-tuning stage, thereby automatically matching similar performance during inference. Alternatively or additionally, embodiments of this application can also add a private interface to the large cloud model, transmitting the latest clinical rules for prostate cancer to the model for secondary parsing via API. This eliminates the need to store the large model ontology within the hospital and also addresses some of the data shortage issues. These optional solutions each have their own characteristics in terms of flexibility, deployment cost, and update / iteration modes, and can be selected or combined based on the hospital's size, data security strategy, and existing technological reserves.

[0048] Figure 5 This is an exemplary structural block diagram illustrating a device 500 for identifying prostate cancer according to an embodiment of this application. It will be understood that the device implementing the solution of this application can be a single device (e.g., a computing device) or a multifunctional device including various peripheral devices.

[0049] like Figure 5 As shown, the device of this application may include a central processing unit (“CPU”) 511, which may be a general-purpose CPU, a dedicated CPU, or other information processing and program execution unit. Furthermore, the device 500 may also include a mass storage device 512 and a read-only memory (“ROM”) 513. The mass storage device 512 may be configured to store various types of data, including various serological indicators related to the subject, imaging information and case text information, prompt words, target text information, image sequence fusion features, algorithm data, intermediate results, and various programs required to run the device 500. The ROM 513 may be configured to store power-on self-test of the device 500, initialization of various functional modules in the system, drivers for the system's basic input / output, and data and instructions required to boot the operating system.

[0050] Optionally, device 500 may also include other hardware platforms or components, such as the tensor processing unit (“TPU”) 514, graphics processing unit (“GPU”) 515, field-programmable gate array (“FPGA”) 516, and machine learning unit (“MLU”) 517 shown. It is understood that although various hardware platforms or components are shown in device 500, they are merely exemplary and not limiting, and those skilled in the art can add or remove appropriate hardware as needed. For example, device 500 may implement the method for identifying prostate cancer of this application by including only a CPU, associated storage devices, and interface devices.

[0051] In some embodiments, to facilitate data transmission and interaction with external networks, the device 500 of this application further includes a communication interface 518, through which it can connect to a local area network / wireless local area network (“LAN / WLAN”) 505, and further through the LAN / WLAN to connect to a local server 506 or to the Internet (“Internet”) 507. Alternatively or additionally, the device 500 of this application can also directly connect to the Internet or cellular network via the communication interface 518 based on wireless communication technology, such as wireless communication technology based on 3G (“3G”), 4G (“4G”), or 5G (“5G”). In some application scenarios, the device 500 of this application can also access the server 508 and database 509 of an external network as needed to obtain various known algorithms, data, and modules, and can remotely store various data, such as various data or instructions used to present, for example, serological index information, imaging information and case text information of the test subject, prompt words, target text information, image sequence fusion features, etc.

[0052] Peripherals of device 500 may include a display device 502, an input device 503, and a data transmission interface 504. In one embodiment, the display device 502 may include, for example, one or more speakers and / or one or more visual displays, configured to provide voice prompts and / or display images and videos for the identification of prostate cancer according to this application. The input device 503 may include, for example, a keyboard, mouse, microphone, posture capture camera, and other input buttons or controls, configured to receive audio data input and / or user commands. The data transmission interface 504 may include, for example, a serial interface, parallel interface, or Universal Serial Bus interface (“USB”), Small Computer System Interface (“SCSI”), Serial ATA, FireWire (“FireWire”), PCI Express, and High Definition Multimedia Interface (“HDMI”), configured for data transmission and interaction with other devices or systems. According to the scheme of this application, the data transmission interface 504 can receive prostate images acquired by an MRI acquisition device and transmit prostate images or various other types of data or results to device 500.

[0053] The CPU 511, mass storage 512, ROM 513, TPU 514, GPU 515, FPGA 516, MLU 517, and communication interface 518 of the device 500 of this application can be interconnected via bus 519, and can interact with peripheral devices through this bus. In one embodiment, the CPU 511 can control other hardware components in the device 500 and its peripheral devices through this bus 519.

[0054] The above combination Figure 5 This application describes a device for identifying prostate cancer that can be used to execute this application. It should be understood that the device structure or architecture described herein is merely exemplary, and the implementation methods and entities of this application are not limited thereto, but can be modified without departing from the spirit of this application.

[0055] Based on the foregoing description in conjunction with the accompanying drawings, those skilled in the art will understand that the embodiments of this application can also be implemented by software programs. Therefore, this application also provides a computer-readable storage medium storing computer-readable instructions for identifying prostate cancer. When executed by one or more processors, these computer-readable instructions can be used to implement the embodiments of this application in conjunction with the accompanying drawings. Figure 3 The method described is for identifying prostate cancer.

[0056] It should be noted that although the operations of the method of this application are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the steps depicted in the flowchart can be performed in a different order. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0057] It should be understood that when the terms "first," "second," "third," and "fourth," etc., are used in the claims, specification, and drawings of this application, they are used only to distinguish different objects and not to describe a specific order. The terms "comprising" and "including" as used in the specification and claims of this application indicate the presence of the described features, integrals, steps, operations, elements, and / or components, but do not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components, and / or collections thereof.

[0058] It should also be understood that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application. As used in this specification and claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term “and / or” as used in this specification and claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes such combinations.

[0059] Although the embodiments of this application are described above, the content is merely an example adopted for the purpose of facilitating understanding of this application and is not intended to limit the scope and application scenarios of this application. Any person skilled in the art described in this application may make any modifications and changes in the form and details of the implementation without departing from the spirit and scope disclosed in this application, but the scope of patent protection of this application shall still be determined by the scope defined in the appended claims.

Claims

1. A device for identifying prostate cancer, characterized by, Comprise: a processor; and a memory, wherein program instructions for identifying prostate cancer are stored, when the program instructions are executed by the processor, the device realizes the following operations: acquire multi-modal data, wherein the multi-modal data includes serological index information, imaging information and case text information of the examinee; input the multi-modal data into the trained identification model for prostate cancer identification to obtain an identification result; wherein the identification model is trained by: defining expert knowledge text based on the serological indicators as prompt words; splicing the prompt words in the case text information to form target text information; input the imaging information and the target text information into the identification model for interactive learning to obtain the trained identification model; wherein the serological index information at least includes total prostate specific antigen, prostate specific antigen ratio and prostate specific antigen density; the imaging information at least includes one or more magnetic resonance sequences in T1WI, T2WI, DWI and ADC; the case text information at least includes historical case information, symptom description information and detection record information; wherein before inputting the imaging information and the target text information into the identification model for interactive learning, it further comprises: extracting sequence characteristics of each magnetic resonance sequence in the imaging information using the identification model respectively; performing fusion operation on the sequence characteristics of each magnetic resonance sequence to obtain fusion characteristics; the inputting the imaging information and the target text information into the identification model for interactive learning to obtain the trained identification model comprises: further fine-tuning the identification model by interactive learning of fusion characteristics and target text to obtain the trained identification model.

2. The apparatus of claim 1, wherein, wherein it further comprises: performing corresponding preprocessing operations on the serological index information, the imaging information and the case text information respectively.

3. The apparatus of claim 2, wherein, wherein performing corresponding preprocessing operations on the serological index information, the imaging information and the case text information respectively comprises: performing one or more preprocessing operations of outlier detection, standardization or missing value filling on the serological index information; performing one or more preprocessing operations of registration, sequence differentiation or annotation on the imaging information; performing one or more preprocessing operations of text recognition, structured storage or regularization on the case text information.

4. The apparatus of claim 1, wherein, wherein defining expert knowledge text based on the serological indicators as prompt words comprises: comparing each indicator in the serological indicators with the corresponding threshold value; defining expert knowledge text based on the comparison result as the prompt words.

5. The apparatus of claim 4, wherein, wherein splicing the prompt words in the case text information to form target text information comprises: splicing the prompt words before or after the case text information to form the target text information.

6. The apparatus of claim 1, wherein, wherein the fusion operation comprises splicing, weighted average or full connection operation.

7. The apparatus of claim 1, wherein, wherein inputting the imaging information and the target text information into the identification model for interactive learning comprises: setting a self-attention layer in the identification model; The self-attention layer is used to interactively learn the fusion features and the target text.

8. A computer-readable storage medium, characterized in that, Computer readable instructions for identifying prostate cancer are stored thereon, which, when executed by one or more processors, implement operations performed by the apparatus of any one of claims 1-7.

Images