Medical diagnosis safety reasoning method and system based on auxiliary calculation party
By combining a three-party distributed computing architecture with lightweight security operators, the problem of high computational and communication overhead in medical diagnosis is solved, enabling low-cost and efficient secure inference for medical diagnosis, which is suitable for medical institutions with limited resources.
Patent Information
- Application Number
- CN202511000837.0
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-21
- Publication Date
- 2025-10-31
AI Technical Summary
In medical diagnostic scenarios, existing technologies incur excessive computational and communication overhead for secure multi-party computation, limiting the widespread adoption of smart healthcare, especially in resource-constrained medical institutions.
A three-party distributed computing architecture is adopted, including the patient side, the medical side, and the auxiliary computing side. Through three-party replication secret sharing technology and function secret sharing technology, the overhead of basic security operators is reduced. It is combined with lightweight security operator combinations, such as the use of distributed comparison functions in the high-level operator layer to reduce the communication and computing overhead of secure comparison.
It significantly reduces computing and communication costs while ensuring data security, improves the efficiency and scalability of medical diagnosis, and is suitable for medical institutions with limited resources.
Smart Images
Figure CN120878019A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the fields of data security and smart healthcare technology, and in particular to a medical diagnostic security reasoning method and system based on auxiliary computing. Background Technology
[0002] Secure multi-party computation allows multiple parties to collaboratively compute a function without disclosing input or intermediate data. In medical diagnostic scenarios, patients typically input their private medical examination data, while the medical staff input a machine learning model (considered a trade secret). The two parties perform secure inference through secure multi-party computation, and the patient receives the diagnostic results. Throughout this process, neither the patient's nor the medical staff's private data is leaked.
[0003] References: CrypTFlow2: Practical 2-Party Secure Inference (ACM CCS2020) proposes secure multi-party computation-based machine learning secure inference (abbreviated as: CrypTFlow2), a commonly used method for secure multi-party computation. The secure operators in CrypTFlow2 are mainly divided into two categories: linear operations (addition, multiplication, vector dot product, etc.) and nonlinear operations (mainly secure comparisons or operations that can be reduced to secure comparisons). In CrypTFlow2, the basic secure operators are implemented using a secret sharing method for two-party addition, which has high basic overhead. Furthermore, in CrypTFlow2, secure comparisons are implemented through parallel carry networks and unintentional transmissions, consuming the majority of communication and computational overhead. These two points severely affect the practicality of the CrypTFlow2 scheme; for example, a neural network with no more than 10 layers requires several seconds of runtime and hundreds of megabytes of communication cost for secure inference.
[0004] In medical diagnostic scenarios, patients receive their own encrypted computation results and the medical provider's encrypted computation results, which are then decrypted to obtain the plaintext computation results. In this approach, data security is achieved through cryptographic encryption, which uses randomness for encryption protection. However, the randomness of communication and computation between the two parties results in high costs. Especially in densely populated areas, where medical resources are limited, it is difficult to support the massive computational overhead, thus restricting the widespread adoption of smart healthcare solutions. Summary of the Invention
[0005] To overcome the problem of high computational overhead in the existing smart healthcare technologies, this invention proposes a medical diagnostic security reasoning method based on an auxiliary computing approach, which greatly reduces computational overhead while ensuring computational security and accuracy.
[0006] This invention proposes a medical diagnostic security reasoning method based on an auxiliary computing party. First, it constructs a three-party distributed computing architecture consisting of a patient, a medical party, and an auxiliary computing party; the patient provides physical examination data, and the medical party provides the computing model.
[0007] The three-party distributed computing architecture executes instructions to perform calculations on physical examination data and computing models through three-party replication and secret sharing computing technology, obtains diagnostic results, and outputs them through the patient.
[0008] Preferably, the execution instructions are obtained by compiling and setting the model. The execution instructions are then combined with the secret sharing of third-party replication to perform calculations on the physical examination data and the calculation model to obtain the diagnostic results.
[0009] The computation process for executing instructions includes: a basic operator layer, a high-level operator layer, and a machine learning layer.
[0010] Preferably, the basic operator layer uses addition or multiplication operators; the advanced operator layer uses vector dot product or safe comparison.
[0011] Preferably, secure comparison employs function secret sharing technology.
[0012] Preferably, the model is set to use logistic regression, XGBoost, or CNN.
[0013] Preferably, the parsing and compilation algorithm for the model is written in Python code.
[0014] Preferably, the three-party distributed computing architecture adopts a semi-honest model.
[0015] The present invention proposes a medical diagnostic security reasoning system based on an auxiliary computing party, comprising: a three-party distributed computing architecture, a parsing and compilation module, and a three-party security reasoning framework;
[0016] The three-party distributed computing architecture consists of a patient, a medical provider, and a computing support provider; the patient is used to upload physical examination data and output diagnostic results; the medical provider is used to upload computing models.
[0017] The parsing and compilation module connects to the patient, medical, and computational assistance sides respectively. The parsing and compilation module compiles the set model to obtain the execution instructions.
[0018] The three-party secure reasoning framework connects the parsing and compilation module, the patient, the medical staff, and the computational assistance provider. The three-party secure reasoning framework executes instructions and performs calculations on the encrypted physical examination data and the encrypted computational model through three-party copying and secret sharing to obtain the diagnostic results.
[0019] Preferably, the three-party distributed computing architecture includes a basic operator layer, an advanced operator layer, and a machine learning layer. The three-party secure inference framework performs calculations on the encrypted physical examination data and the encrypted computation model through three-party replication and secret sharing to obtain diagnostic results; the advanced operator layer uses function secret sharing to implement secure comparison operations.
[0020] The present invention proposes a storage medium storing a computer program, which, when executed, is used to implement the aforementioned medical diagnostic security reasoning method based on auxiliary computing.
[0021] The advantages of this invention are:
[0022] (1) The medical diagnosis security reasoning method proposed in this invention first adjusts the system framework by adding an auxiliary computing party in addition to the patient and medical parties to assist in the calculation. Secondly, by introducing the auxiliary computing party to form a three-party secure computing system model, the basic security operator can be implemented by replacing the two-party additive secret sharing with a lightweight three-party copy secret sharing, thereby reducing the basic overhead.
[0023] (2) In this invention, secure operator operations are performed through the basic operator layer, the advanced operator layer and the machine learning layer, which allows for a more flexible combination of secure operators. For example, the advanced operator layer uses the distributed comparison function in the preprocessed function secret sharing technology, which reduces the computational overhead of secure comparison communication.
[0024] (3) In this invention, the patient's physical examination data, the medical institution's diagnostic model, and all intermediate results are private data and must not be disclosed during the calculation process. The diagnostic results are only available to the patient, thus ensuring data security. Attached Figure Description
[0025] Figure 1 This is a flowchart of a medical diagnostic security reasoning method based on auxiliary calculation proposed in this invention;
[0026] Figure 2 This is a schematic diagram of a medical diagnostic security reasoning method based on auxiliary calculation proposed in this invention;
[0027] Figure 3 This is a structural diagram of the three-party distributed computing architecture proposed in this invention. Detailed Implementation
[0028] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.
[0029] like Figure 1 , Figure 2 , Figure 3 As shown, a medical diagnostic security reasoning method based on auxiliary computing includes the following steps.
[0030] S1. Construct a three-party distributed computing architecture consisting of patients, medical staff, and auxiliary computing providers; patients provide physical examination data, and medical staff provide computing models.
[0031] In a three-party distributed computing architecture, the patient provides physical examination data, receives and outputs diagnostic results; the medical provider provides the computational model (i.e., the diagnostic model) without output; and the auxiliary computing provider has neither input nor output. These three parties constitute three independently cooperating computing nodes, and each party's computing and communication resources should be capable of completing the computational tasks allocated by the secure multi-party computing protocol. The security model of the three-party distributed computing architecture is a semi-honest model in cryptography, meaning that the participating parties (patient, medical, and auxiliary computing providers) execute tasks according to a predetermined protocol, but are allowed to infer the original data from the received ciphertext.
[0032] S2. The set model is compiled to obtain the execution instructions; the patient, medical staff and auxiliary computing staff run the execution instructions, and calculate the ciphertext of the physical examination data and the calculation model through the secret sharing of the three parties to obtain the diagnosis results, which are output by the patient.
[0033] Specifically, the computation process for executing instructions includes a basic operator layer, a high-level operator layer, and a machine learning layer.
[0034] S21. The basic operator layer performs basic calculations on the physical examination data and the calculation model to obtain data C; the basic calculations can be performed by addition or multiplication, etc.
[0035] S22. The advanced operator layer performs calculations on data C to obtain data D. The advanced operator layer uses algorithms including vector dot product and secure comparison. The secure comparison specifically uses function secret sharing technology.
[0036] S23. The machine learning layer processes the data D to obtain the diagnostic result C; the machine learning layer can be a linear layer or a non-linear layer.
[0037] The compilation process in step S2 is executed through a compilation layer; this layer is used to execute the parsing compilation algorithm. The model can be a machine learning algorithm written in Python code, such as logistic regression, XGBoost, or CNN. The compilation layer executes the parsing compilation algorithm on the model, thereby outputting instructions that the third-party secure inference framework can execute, i.e., execution instructions; the third-party secure inference framework then calls secure multi-party computation operators to compute the diagnostic results. The third-party secure inference framework consists of a basic operator layer, a high-level operator layer, and a machine learning layer. The framework is primarily implemented using two secure multi-party computation techniques: three-party copy secret sharing and function secret sharing. In this embodiment, the third-party secure inference framework calls the three-party copy secret sharing algorithm to process the output of the compilation layer, thereby obtaining the diagnostic results.
[0038] The following compares the computational costs of the same operators in the medical diagnostic security inference method based on auxiliary computation proposed in this invention and the CrypTFlow2 algorithm (machine learning security inference based on secure multi-party computation), and the results are shown in Table 1:
[0039] Table 1: Comparison of computational costs of different operators in the two inference methods
[0040]
[0041] In Table 1, n is the bit length of the input data, and k is the number of input data.
[0042] As can be seen, the multiplication operator applied to the CryptoFlow2 algorithm and the method of this invention has the same communication round complexity, but the method of this invention achieves lower communication cost;
[0043] The safety comparison operator is applied to the CrypTFlow2 algorithm and the method of this invention, which achieves improvements in both communication cost and number of communication rounds.
[0044] The vector dot product operator is applied to both the CrypTFlow2 algorithm and the method of this invention. The communication round complexity is the same, but the method of this invention has made great progress in terms of communication cost.
[0045] It is evident that, for any operator applied to this method, the complexity of the communication rounds is constant, and the communication cost is always less than that in the CryptoFlow2 algorithm.
[0046] As can be seen, this invention achieves a medical diagnostic secure reasoning system with low communication costs by combining an auxiliary computing party with a third-party secure reasoning framework.
[0047] Furthermore, during the specific implementation of the method of the present invention, the diagnostic results obtained were the same as the plaintext calculation results, proving the correctness of the method of the present invention.
[0048] Of course, those skilled in the art will recognize that the present invention is not limited to the details of the exemplary embodiments described above, but also includes the same or similar structures that can be implemented in other specific forms without departing from the spirit or essential characteristics of the invention. Therefore, the embodiments should be considered illustrative and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of equivalents of the claims are intended to be included within the present invention. No reference numerals in the claims should be construed as limiting the scope of the claims.
[0049] Furthermore, it should be understood that although this specification describes embodiments, not every embodiment contains only one independent technical solution. This narrative style is merely for clarity. Those skilled in the art should consider the specification as a whole, and the technical solutions in each embodiment can also be appropriately combined to form other embodiments that can be understood by those skilled in the art.
[0050] The technologies, shapes, and structures not described in detail in this invention are all known technologies.
Claims
1. A medical diagnostic security reasoning method based on auxiliary computation, characterized in that, First, a three-party distributed computing architecture is constructed, consisting of patients, medical personnel, and auxiliary computing providers; patients provide physical examination data, and medical personnel provide computing models. The three-party distributed computing architecture executes instructions to perform calculations on physical examination data and computing models through three-party replication and secret sharing computing technology, obtains diagnostic results, and outputs them through the patient.
2. The medical diagnostic security reasoning method based on auxiliary computation as described in claim 1, characterized in that, The execution instructions are obtained by compiling and setting the model. The execution instructions are then combined with the secret sharing of third-party replication to perform calculations on the physical examination data and the calculation model to obtain the diagnostic results. The computation process for executing instructions includes: a basic operator layer, a high-level operator layer, and a machine learning layer.
3. The medical diagnostic security reasoning method based on auxiliary computation as described in claim 2, characterized in that, The basic operator layer uses addition or multiplication operators; the advanced operator layer uses vector dot product or safe comparison.
4. The medical diagnostic security reasoning method based on auxiliary computation as described in claim 3, characterized in that, Secure comparison employs a function secret sharing technique.
5. The medical diagnostic security reasoning method based on auxiliary computation as described in claim 1, characterized in that, The model can be set to use logistic regression, XGBoost, or CNN.
6. The medical diagnostic security reasoning method based on auxiliary computation as described in claim 5, characterized in that, The parsing and compilation algorithm for the model is written in Python code.
7. The medical diagnostic security reasoning method based on auxiliary computation as described in claim 1, characterized in that, The three-party distributed computing architecture adopts a semi-honest model.
8. A medical diagnostic security reasoning system based on auxiliary computation, characterized in that, include: A third-party distributed computing architecture, a parsing and compilation module, and a third-party secure inference framework; The three-party distributed computing architecture consists of a patient, a medical provider, and a computing support provider; the patient is used to upload physical examination data and output diagnostic results; the medical provider is used to upload computing models. The parsing and compilation module connects to the patient, medical, and computational assistance sides respectively. The parsing and compilation module compiles the set model to obtain the execution instructions. The three-party secure reasoning framework connects the parsing and compilation module, the patient, the medical staff, and the computational assistance provider. The three-party secure reasoning framework executes instructions and performs calculations on the encrypted physical examination data and the encrypted computational model through three-party copying and secret sharing to obtain the diagnostic results.
9. The medical diagnostic security reasoning system based on auxiliary computing as described in claim 8, characterized in that, The three-party distributed computing architecture includes a basic operator layer, an advanced operator layer, and a machine learning layer. The three-party secure inference framework performs calculations on the encrypted physical examination data and the encrypted computation model through three-party replication and secret sharing to obtain diagnostic results. The advanced operator layer uses function secret sharing to implement secure comparison operations.
10. A storage medium, characterized in that, The system contains a computer program that, when executed, is used to implement the medical diagnostic security reasoning method based on an auxiliary computing party as described in any one of claims 1-7.