Radiology department image focus prediction system based on electric digital data processing

Abstract

The invention relates to the technical field of electrical digital data processing, and discloses a radiology department image focus prediction system based on electrical digital data processing, which comprises the following steps: generating a local computing environment fingerprint representing a current computing environment before processing image data, and carrying out association binding on the fingerprint and a generated prediction result; and the system also stores a golden fingerprint of a standard environment, and automatically executes an adjustment operation based on the difference between the local computing environment fingerprint and the golden fingerprint when the local computing environment fingerprint and the golden fingerprint are inconsistent, so that the computing environment identity is anchored for each data processing result, and a set of environment baseline self-adaptive regularization and self-healing mechanism is established; therefore, the problems of data comparability loss and systematic trust crisis caused by dynamic evolution of a computing environment are solved, and the context credibility and long-term value of a data processing result in the whole life cycle are ensured.

Description

TECHNICAL FIELD

[0001] The application relates to a radiology image lesion prediction system based on electric digital data processing and belongs to the technical field of electric digital data processing. BACKGROUND

[0002] Currently, a computing model is used to analyze high-precision data such as medical images, which is a general technical method for improving work quality. The technical basis of this method lies in the determinacy and traceability of the processing results. However, in the production environment, the computing environment on which the system runs, including the operating system, the driver program, the software function library and the processor microcode, is in a continuous dynamic update cycle. This update is a necessary operation for system maintenance and performance improvement. For the same input data, the system will produce result changes caused by the slight differences in the computing environment at different time points. This phenomenon makes the output data of the system on the vertical time axis lose strict comparability, and the long-term effectiveness of the data assets is also reduced.

[0003] To solve this problem, one idea in the field is to fix the computing environment through containerization and other technical means. However, this method increases the complexity of system operation and maintenance, and is contradictory to the objective need of continuous iteration and update of the system. Another idea focuses on improving the robustness of the algorithm model, but this does not solve the root of the problem. Because the result changes are not all caused by the deficiency of the algorithm, but by the lack of computing context information when the algorithm is generated, resulting in an uncertain source of any result difference. For example, the Chinese invention patent with the authorization publication number CN117153347B discloses a medical image processing system and method based on big data analysis and a storage medium. The core of this patent is to monitor and warn the physical risks such as equipment falling and damage caused by improper use of the user (such as excessive weight or too outward sitting posture) through the orderly cascade triggering of a series of detection devices such as body height, action trend, load position and image review. The technical solution of this patent focuses on protecting the physical safety of the equipment entity and the operating personnel. However, the limitation of this solution is also in this: its technical focus is completely fixed on the response to external physical behavior, and its monitoring and intervention object is the interaction between people and physical equipment, completely ignoring the inherent requirement of data processing task for the stability and consistency of the computing environment. Even if this system can perfectly prevent the equipment from falling due to improper load, it cannot guarantee that the image analysis results still have strict comparability after the update of the driver program or the core function library version. This design idea, which focuses on physics and ignores data, exposes the general shortcoming of the prior art in ensuring the credibility of the data throughout the life cycle, and does not solve the root of the problem.

[0004] Specifically, existing technologies have the following shortcomings: 1. Data processing results are separated from the computational environment context in which they were generated, lacking an inseparable technical connection; 2. When comparing and analyzing historical data results, there is a lack of an effective mechanism for pre-verifying the consistency of the computational environment between two processing tasks; 3. When unexpected changes occur in the processing results, it is impossible to definitively distinguish between differences in algorithmic logic and differences in the computational environment. Therefore, how to design a data processing system that can permanently correlate the computational environment characteristic parameters of each data processing with the digital results generated by that processing, under the condition of continuous evolution of the computational environment, and on this basis, establish a reliable data comparison and traceability analysis method, becomes the technical problem to be solved by this invention. Summary of the Invention

[0005] This invention provides a radiological imaging lesion prediction system based on electronic digital data processing. Its main purpose is to solve the problem in the existing data processing process that the separation between data results and their computational environment context information makes it impossible to make deterministic attributions and reliable comparisons of result differences throughout the continuously evolving life cycle of the system.

[0006] To achieve the above objectives, the present invention provides a radiological imaging lesion prediction system based on electronic digital data processing, the system comprising: An environment fingerprint generation module is configured to obtain a set of textualized environmental feature metadata of the computing node executing the radiology image data processing task before the task is triggered to generate a prediction result, and generate a local computing environment fingerprint representing the current computing environment based on the environmental feature metadata using a deterministic hash algorithm. A data binding module is configured to associate and store the local computing environment fingerprint as mandatory metadata with the prediction results generated by the task. A baseline management module is configured to store a gold fingerprint, predefined by the administrator, containing a list of standard environment characteristic parameters; An environment conditioning module is configured to execute the following judgment and control rules after the local computing environment fingerprint is generated: parse and compare the local computing environment fingerprint with the golden fingerprint to identify a set of differences between the lists of environmental feature parameters contained in the two; and when the set of differences is not empty, automatically trigger the execution of a preset adjustment operation script based on the content of the set of differences, whose internal instructions correspond to the parameter differences identified in the set of differences, so as to align the environmental feature parameters of the computing node in the direction defined by the golden fingerprint.

[0007] Preferably, the characteristic parameters of the computing environment include at least one of the following: operating system name and kernel version, software library version on which data processing depends, graphics processor model and driver version, and central processing unit model and microcode version.

[0008] Preferably, the system further includes a result analysis module, which is configured to, when comparing and analyzing at least two prediction results, first determine whether the local computing environment fingerprints associated with the at least two prediction results are consistent, and when it is determined that the local computing environment fingerprints are inconsistent, prevent direct numerical comparison of the at least two prediction results and issue a warning of environment inconsistency.

[0009] Preferably, the data binding module is configured to achieve association binding by writing the local computing environment fingerprint as metadata into the header area of ​​the file containing the prediction results.

[0010] Preferably, the data binding module is also configured to generate a composite evidence hash to cryptographically link the current association binding record with the historical association binding record; the system also includes an integrity verification module, which is configured to verify the integrity of the historical record chain by recursively verifying the consistency between the composite evidence hash and the historical hash value in reverse, starting from the latest association binding record.

[0011] Preferably, the system further includes: a behavior analysis module configured to process a set of standard image data under a specific local computing environment fingerprint to obtain a set of corresponding prediction results, and to calculate a variance index characterizing the macroscopic prediction behavior of the model under that computing environment based on the confidence level of the prediction results; and a drift diagnosis module configured to calculate the drift degree of the variance index obtained under at least two different local computing environment fingerprints. And in drift degree When the drift exceeds a preset threshold determined based on historical statistical data, an alarm signal is generated indicating that the model's macroscopic predictive behavior has drifted. The drift degree is... The calculation method is as follows ,in For drift degree, To provide a variance metric for fingerprints in a local computing environment. This is a variance metric for fingerprints in another local computing environment.

[0012] Preferably, the system further includes: a benchmark verification module configured to execute a predefined standardized benchmark computation task with fixed input and whose correct output and benchmark execution time are known in advance, after the environment fingerprint generation module generates the local computing environment fingerprint and before the prediction result is actually generated; and an execution control module configured to determine whether to authorize continued execution to generate the prediction result based on whether an execution result of the benchmark computation task meets a preset verification criterion, wherein the verification criterion includes: the consistency of bit-level comparison between the computation output value of the benchmark computation task and the known correct output result, and the stability of comparing its execution time with the benchmark execution time.

[0013] Preferably, the system further includes: a historical fingerprint repository for storing multiple local computing environment fingerprints associated with historical prediction results; and a security audit module configured to: receive a query request defining at least one software component with a security vulnerability and its version information, and, based on the query request, search the historical fingerprint repository to identify historical prediction results generated by a computing environment containing a software component with a security vulnerability.

[0014] Preferably, when applied to a federated learning scenario, the system further includes: an environment quality assessment module configured to score the local computing environment fingerprint submitted by a federated learning participant and associated with its local model update based on a preset scoring rule, to obtain an environment quality score, wherein the scoring rule includes: if the local computing environment fingerprint is consistent with the golden fingerprint, a standard score is added; if the software component version contained in the local computing environment fingerprint is the latest stable version, an advancement score is added; and if the local computing environment fingerprint contains a software component version known to have security vulnerabilities, a security score is deducted; and a model aggregation module configured to use the environment quality score to weight the model updates of each participant when aggregating the local model updates of multiple participants.

[0015] Preferably, the environment conditioning module is further configured to: after triggering the execution of the adjustment operation script, instruct the environment fingerprint generation module to regenerate the updated local computing environment fingerprint, and compare the updated local computing environment fingerprint with the gold fingerprint again to verify the correctness of the execution of the adjustment operation script.

[0016] Compared with the prior art, the beneficial effects of the present invention are: 1. By acquiring the characteristic parameters of the computing environment and generating a computing environment fingerprint before triggering the image data processing task, and then associating and binding this fingerprint with the generated prediction results, and using the consistency of the fingerprint as a pre-judgment criterion for executing differential analysis logic during subsequent result analysis, a fixed association and analysis method between data processing results and their computing environment context is established. Compared with the existing data processing flow, where the processing results are independent of the hardware and software environment in which they are generated, and when faced with inconsistent processing results, it is impossible to distinguish whether the difference is caused by changes in algorithm logic or changes in the computing environment, this invention, through the above mechanism, ensures that each data processing result carries the environmental state information at the time of its generation. When performing result tracing or comparative analysis, the environmental consistency is first identified through the fingerprint, separating the uncertain factor of environmental change from the complex analysis process, making the attribution path of result differences clear and definite.

[0017] 2. After the aforementioned fingerprint generation and before the actual generation of the prediction result, a step of executing a standardized benchmark calculation task is added. The decision to continue with the subsequent prediction result generation process is based on whether the execution result of this benchmark calculation task meets a preset verification standard. Existing technologies typically focus on the static configuration information of the computing environment, but lack effective means to detect instantaneous fluctuations in the runtime environment caused by factors such as virtualization scheduling resource contention. These instantaneous fluctuations can also affect the certainty of the final calculation result. This invention provides a way to dynamically check the health of the actual runtime environment before data processing tasks occur by immediately executing a standardized benchmark calculation task sensitive to computing resources after the static environment fingerprint verification is passed, and verifying its output value and execution time. This avoids high-cost data processing in unstable instantaneous environments and prevents potential distortion of calculation results caused by runtime environment jitter.

[0018] 3. A golden fingerprint defined as the standard computing environment is stored. After generating the fingerprint of the current computing environment, it is compared with the golden fingerprint. If they are inconsistent, an adjustment operation is automatically performed based on the difference to align the current computing environment with the standard computing environment. In a distributed computing cluster, due to configuration updates, patch upgrades, and other operations on each node, the environment configuration will gradually deviate, forming multiple comparability islands, which weakens the value of the aforementioned fingerprint comparison mechanism in global applications. This invention introduces the golden fingerprint as a unified baseline, transforming the locally generated fingerprint from a passive record of information into an active control signal. When a deviation is detected between the local fingerprint and the golden fingerprint, the deviation information is directly used to trigger a specific environment normalization script, forming a closed-loop adjustment path of fingerprint generation-difference comparison-automatic normalization-re-verification. This continuously converges the computing environments of each node in the distributed environment towards a unified standard, ensuring the comparability of data processing results across the entire system.

[0019] 4. Based on preset scoring rules, the computational environment fingerprints submitted by federated learning participants and associated with their local model updates are scored to obtain an environment quality score. This environment quality score is then used to weight the model updates of each participant when aggregating the local model updates of multiple participants. In collaborative data processing tasks like federated learning, the contribution of each participant is usually measured only by the amount of data they provide. However, the standardization of their computational environment and the sophistication of their hardware and software also directly affect the quality of the model update, a factor that is generally overlooked in existing technologies. This invention transforms the computational environment fingerprints submitted along with the model update into a quantifiable environment quality score through scoring rules. This score is directly used as a weighting coefficient to adjust the proportion of the participant's current model update in the global model aggregation. This provides a way to measure computational contribution for the federated learning system, incentivizing each participant to maintain a high-quality computational environment, thereby improving the overall training quality of the global model. Attached Figure Description

[0020] Fig. 1 This is a schematic diagram of the system collaborative architecture and data closed-loop process of the present invention; Fig. 2 This is a performance indicator verification curve for the core functions of the system of this invention; Fig. 3 This is a logic block diagram showing the modular structure of the system functions of the present invention. Detailed Implementation

[0021] To make the objectives, technical solutions, and advantages of the present invention clearer, the technical solutions of the present invention will be described in detail below. 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 skilled in the art without creative effort are within the scope of protection of the present invention.

[0022] The radiological imaging lesion prediction system based on electro-digital data processing claimed in this invention is mainly composed of an environmental fingerprint generation module, a data binding module, a baseline management module, and an environmental normalization module. The data processing flow of each module is as follows: the environmental fingerprint generation module captures and hashes the instantaneous state of the computing node's execution environment; then, the data binding module associates this state information with the output result of the data processing task; finally, the environmental normalization module performs continuous closed-loop calibration of the computing node's local environment with reference to the global standards stored in the baseline management module, thereby achieving a dynamic and variable environment normalization system. In a distributed computing cluster, a stable, traceable, and globally comparable execution benchmark is provided for all electronic digital data processing activities. In electronic digital data processing applications requiring high determinism, such as medical image analysis, the output results of data processing are not only related to the algorithm logic and input data, but also affected by the computing environment in which the processing is performed. To address this technical issue, the environmental fingerprint generation module in this invention is configured to obtain a set of textual environmental feature metadata characterizing the current computing environment before a radiology image data processing task is triggered to generate a prediction result. This is done by calling standardized query instructions built into the computing node's operating system. This acquisition process utilizes the existing processors in the system. This metadata may include, but is not limited to, the operating system name and kernel version obtained by executing the uname -a instruction, the software library version obtained by executing the ldd instruction to query the core dynamic link library, the graphics processor model and driver version obtained by executing the nvidia -smi instruction, and the central processing unit model and microcode version obtained by executing the lscpu instruction. Then, the environmental fingerprint generation module concatenates the above-mentioned set of textual metadata in a preset deterministic order to form a long string, and uses... Using a standard cryptographic hash algorithm, such as Secure Hash Algorithm 256 (SHA-256), the long string is processed to generate a unique, fixed-length string. This string is defined as the local computing environment fingerprint representing the current computing environment. A concrete numerical example is: if the obtained kernel version is 5.4.0-80-generic and the driver version is 470.57.02, the module concatenates them into 5.4.0-80-generic;470.57.02, and after performing the SHA-256 operation on this string, a string like a8c7e is obtained...The fixed-length hash value of 1fb9d serves as the local computing environment fingerprint. Thus, this module generates a compact and verifiable digital identity for the computational context of each data processing task at the moment of execution. To ensure an inseparable link between the generated local computing environment fingerprint and its corresponding processing result, the data binding module is configured to associate and store this fingerprint as metadata with the prediction result generated by the task. This association binding process provides at least two alternative implementation methods to ensure its reliability under different system architectures. The preferred implementation path is to directly write the local computing environment fingerprint as metadata into the header area of ​​the file containing the prediction result. For example, for a prediction result file stored in JSON format, the fingerprint can be written into a reserved field named `computation_fingerprint`. The alternative implementation path is to package the original result file and a separate metadata file containing fingerprint information into a single archive file with a checksum, such as a tar archive, at the file system level. In this way, the data binding module ensures that each data output object carries its own generated computing environment information, which can be directly read by any downstream analysis process, from the moment it is generated.

[0023] In distributed computing clusters with a large number of nodes, the environments of each node are prone to deviation due to asynchronous maintenance operations, forming multiple comparable silos and weakening the value of data comparison. To address this issue, this system introduces a proactive environment adjustment mechanism, specifically implemented by a baseline management module and an environment normalization module. The baseline management module is configured to store a predefined "golden fingerprint" containing a list of standard environment characteristic parameters, defined by the system administrator. The creation of this golden fingerprint follows a standardized calibration procedure: selecting a optimally configured and fully verified physical server as the baseline node, executing the aforementioned environment fingerprint generation process on that node, and storing its output hash value as the system's golden fingerprint in a shared storage area accessible to all computing nodes. Correspondingly, the environment normalization module is configured to immediately execute a set of judgment and control rules after the local computing environment fingerprint is generated on each computing node. It first parses and compares the local computing environment fingerprint with the global golden fingerprint to identify a set of differences between the lists of environment characteristic parameters contained in the two. For example, if the comparison reveals differences in the local node's graphics processor driver version... If the version is 470.57.02, and the standard version corresponding to the gold fingerprint is 510.47.03, then the difference set will contain this difference record. If and only if the difference set is not empty, the environment conditioning module will automatically trigger the execution of a preset adjustment operation script based on the content of the difference set. The internal instructions of this script correspond to the parameter differences identified by the difference set. Taking the aforementioned driver version difference as an example, the system will trigger an automated deployment script pointing to the driver installation program for version 510.47.03. It should be noted that, to form a management closed loop, the environment conditioning module is... Further configuration involves the instruction environment fingerprint generation module regenerating the updated local computing environment fingerprint after triggering the execution of the adjustment operation script. The updated local computing environment fingerprint is then compared with the golden fingerprint again. This verification-repair-re-verification process will continue to loop until the difference set is empty. In this way, the mechanism transforms the locally generated fingerprint from a passive record of information into an active control signal, continuously converging the computing environments of various nodes in the distributed environment towards a unified standard, thus providing a basis for the comparability of data processing results across the entire system.

[0024] To leverage the established context association mechanism, this system also includes a result analysis module. When comparing and analyzing at least two prediction results, such as predictions of the same image from models of versions V2.1 and V2.2, this module is configured to first determine whether the local computing environment fingerprints associated with these two prediction results are consistent. If the fingerprints are found to be inconsistent, the module will prevent direct numerical comparison of the two prediction results and output a clear warning of environment inconsistency to the user interface or system log. Simultaneously, it will highlight the differences between the two fingerprints, thus definitively distinguishing between changes in algorithm logic and changes in the computing environment as the cause of the result differences. To ensure the long-term integrity and immutability of historical records, the data binding module can also be configured to introduce a chain-like evidence storage mechanism when performing association binding, i.e., for the nth prediction result... During the binding process, the module will additionally obtain the final hash value of the (n-1)th binding record. And a summary based on the current prediction results. Current environment fingerprint and historical hash values ,pass The calculation method generates a new composite evidence hash. This allows for the cryptographic linking of current association binding records with historical association binding records. Correspondingly, the system also includes an integrity verification module, which can verify the integrity and tamper-proof status of the entire historical record chain by recursively verifying the consistency between the composite storage hash and the historical hash value, starting from the latest association binding record. This makes any single tampering with the historical record invalid in mathematical verification of the entire subsequent chain.

[0025] To manage the composite evidence storage hash chain, the system sets up an append-only evidence storage log. This log can be a local file or a dedicated database table structure, where each record contains three fields: a unique identifier for the current record, the composite evidence storage hash for the current record, and so on. And the composite evidence hash of the previous record. The integrity verification module's execution logic is a reverse recursive verification algorithm. When verifying the integrity of the historical record chain, the verification task starts with the latest record in the evidence log. The module first reads the content of the latest record (denoted as the nth record) and independently recalculates its composite evidence hash. The calculation method is to use the prediction result digest associated with the nth record, the local computing environment fingerprint, and the hash value of the (n-1)th record in the evidence log. The recalculated hash value is compared with the hash value of the nth record stored in the evidence log. The chain is compared. If they do not match, the chain is determined to have been tampered with. If they match, the chain is checked in reverse order for the (n-1)th and (n-2)th records, until the first record is checked. Only when the hash values ​​of all records match is the entire historical record chain confirmed to be complete and tamper-free.

[0026] Furthermore, to address situations where the static configuration of the computing environment is consistent but the runtime state is unstable, such as instantaneous fluctuations in computing resources caused by virtualization scheduling or multi-task resource contention, this system also includes a benchmark verification module and an execution control module. The benchmark verification module is configured to insert a pre-verification step after the environment fingerprint generation module generates the local computing environment fingerprint and before the core image prediction process is actually executed. This step executes a predefined standardized benchmark calculation task with fixed input and whose correct output and benchmark execution time are known in advance, such as a microprogram that performs 100 multiplications on a 1024x1024 floating-point matrix. Correspondingly, the execution control module determines whether to authorize the continued execution of subsequent prediction result generation processing based on whether the execution result of the benchmark calculation task meets a preset verification standard. The verification standard here includes two aspects: first, comparing the calculated output value of the benchmark calculation task with the known correct output result at the bit level to determine their consistency; second, comparing the current execution result with the expected correct output result. The runtime is compared with the baseline runtime calibrated at the baseline node, such as 50ms, to determine its stability, such as whether the fluctuation range is within ±5%. Only when both static fingerprint verification and dynamic runtime baseline verification pass are the core data processing tasks authorized to execute; otherwise, the task is rejected and specific runtime environment anomaly information is recorded. This provides the system with a way to dynamically check the health of the actual runtime environment before data processing tasks occur, preventing potential distortion of calculation results caused by runtime environment jitter. Furthermore, the system can use long-term accumulated fingerprint data for system analysis and management. For example, the system may include a behavior analysis module and a drift diagnosis module. The behavior analysis module automatically calls the current model when a new, previously unrecorded local computing environment fingerprint first appears, processes a small set of fixed, representative standard image data, and calculates a variance index characterizing the model's behavior in that computing environment based on the confidence level of these prediction results. The drift diagnostic module is configured to calculate the drift of variance metrics acquired under fingerprints from at least two different local computing environments. and define a drift degree The calculation method is as follows In the formula, For drift degree, To provide a variance metric for fingerprints in a local computing environment. This is a variance metric for a fingerprint in another local computing environment. When the drift exceeds a preset drift threshold determined based on historical statistical data, such as three standard deviations of historical values, the system generates an alarm signal indicating that the model behavior has drifted. This allows operations and maintenance personnel to understand the impact of environmental changes on the overall stability of the model.

[0027] The historical fingerprint data accumulated by this system can also be used for network security auditing. To this end, the system can include a historical fingerprint repository and a security audit module. The historical fingerprint repository is used to archive all local computing environment fingerprints and their parsed metadata associated with historical prediction results. The security audit module is configured to receive a query request from a security administrator, defining at least one software component with a security vulnerability and its version information, such as a JSON object containing `{component::OpenSSL, vulnerable_versions:<1.1.1d}`. Based on the query request, it searches the historical fingerprint repository to identify a list of all historical prediction results generated by computing environments containing the software component with the security vulnerability. This extends the traditional security approach, which only scans the current system state, to the entire lifecycle of historical data assets. When applied to distributed data processing scenarios such as federated learning and multi-center collaboration, this system can also include an environmental quality assessment module to evaluate the computational contributions of each participant. The system comprises a module and a model aggregation module. When a federated learning participant submits its local model update, it must simultaneously submit the fingerprint of its local computing environment corresponding to this training. The environment quality assessment module scores this fingerprint based on a preset, quantifiable scoring rule to obtain an environment quality score. This scoring rule is a deterministic procedure with clearly defined addition and subtraction items. For example, if the local computing environment fingerprint is completely consistent with the gold fingerprint recommended by the federated network, a standard score of 50 points is added; if the software library version contained in the fingerprint is the latest stable version, a state-of-the-art score of 10 points is added; if the fingerprint contains a software component version known to have high-risk security vulnerabilities, a security score of 30 points is deducted. Finally, when aggregating the local model updates of all participants, the model aggregation module uses their respective environment quality scores to weight the model updates of each participant. A participant with a higher environment quality score will have a larger proportion of contribution in the global model. This provides the federated learning system with a way to measure computational contribution beyond simply the amount of data, incentivizing all participants to maintain a high-quality, standardized computing environment.

[0028] Example 1: This example demonstrates the specific operation of the technical solution in a particular scenario. In a hospital information system continuously running and deploying a radiology lesion prediction system, the operations and maintenance department upgraded a lung nodule prediction model from version V2.1 to V2.2 as planned. For routine quality control review, the system was instructed to use the V2.2 model to repeat the processing of a previously processed CT image (PID-0753) that had been processed by version V2.1. After processing, the system operator discovered that the risk value given by version V2.2 for the same suspicious nodule at the same location in the image deviated from the historical risk value given by version V2.1, and the coordinates of its predicted bounding box also shifted. This phenomenon required verification of data consistency. At this time, the result analysis module in the system of this invention was triggered to compare and analyze the results of these two processing steps. According to the established procedures, this module did not... Instead of directly comparing numerical values, the system first extracts local computing environment fingerprints bound to the processing results of versions V2.1 and V2.2. Upon comparison, the system determines that the two hash strings are inconsistent, indicating that the two data processing tasks were executed in different computing environments. Subsequently, the result analysis module automatically parses the specific environmental feature metadata contained in the two fingerprints and identifies the content of the difference set as the graphics processor driver version item. Specifically, the parameter value in the fingerprint bound to the V2.1 result is 470.57.02, while the corresponding parameter value in the V2.2 result is 510.47.03. This difference arises because the server operating system performed a routine driver update during the two model version iterations. Through the collaborative operation of the environment fingerprint generation module and the result analysis module, a data processing result drift problem is automatically and deterministically transformed into a computing environment context mismatch determination problem.

[0029] Based on this determination, the system prevented direct numerical comparison of the risk values ​​and bounding box coordinates of the two prediction results and output an alarm message to the monitoring interface of the operations and maintenance department. The message read: "The two predictions were generated in different computing environments, and the results are not directly comparable. The environment difference is that the graphics processor driver version changed from 470.57.02 to 510.47.03." The operations and maintenance personnel thus understood that the source of the result deviation was not a defect in the model algorithm itself, but rather a change in the underlying computing environment. This avoided incorrect troubleshooting of the algorithm team and prevented the use of a data point generated in an inconsistent environment and lacking comparability for subsequent longitudinal evaluation of system performance. The system's mechanism of associating each data processing process with its computing environment resolves the contradiction between the loss of data comparability and the uncertainty of the attribution of result differences caused by the continuous evolution of the software and hardware environment throughout the system's lifecycle. The system log automatically recorded the entire comparative analysis process, including the fingerprints of the two different local computing environments, the specific differences identified, and the system's control behavior that prevented direct numerical comparison, providing a complete and auditable contextual traceability record for the entire lifecycle of this data asset.

[0030] Example 2: To verify the ability of this invention to identify and manage the comparability of data processing results in different computing environments, this example designed and executed a comparative verification experiment. The experimental platform consisted of a server configured with an Intel Xeon W-2223 CPU, an NVIDIA RTX A4000 GPU, and 64GB of memory, running Ubuntu 20.04.3LTS. The experimental data used a set of DICOM files containing 10 anonymized CT images as a fixed gold verification image set, which remained unchanged across all experimental groups. The experimental steps included constructing four discrete, controllable computing environments to simulate environment drift conditions, specifically set as follows: Environment A served as the baseline environment, with GPU driver version 470.57.02, and its computing environment fingerprint was set as the system's gold fingerprint; Environment B simulated a driver upgrade scenario, with its driver version updated to 510.47.03, and all other configurations identical to Environment A; Environment C served as a reference environment, with all hardware and software configurations consistent with Environment A; Environment D was configured identically to Environment B and was used to verify system behavior when a specific module was missing.

[0031] The experiment was set up with two main groups: a control group, which used a conventional data processing program without any modules of the present invention; and an experimental group, which used the system of the present invention fully described in the aforementioned specific embodiments. The purpose of this experiment was to verify two functions, namely, passive identification and active normalization of environmental differences. First, the passive identification function was verified by temporarily disabling the environmental normalization module in the experimental group system, so that the system only performed fingerprint generation, binding and analysis functions. Then, the gold verification image set was processed in environments A, B and C, respectively, and the local computing environment fingerprint generated by each processing, an exemplary output risk value, and the behavior of the system when attempting to compare the results in environments A, B and C pairwise were recorded. The experimental process and data records are shown in Table 1.

[0032] Table 1: Verification data for the environmental difference recognition function.

[0033]

[0034] Referring to Table 1, in environments A and C, since the characteristic parameters of the computing environments are completely identical, the local computing environment fingerprints generated by the system are also exactly the same, both being a8c7e...1fb9d, and their output risk values ​​are stable at 0.8824. When the result analysis module compares these two results, they are determined to be directly comparable due to the identical fingerprints. However, in environment B, a change in only one parameter, the driver version, resulted in the system generating a completely different local computing environment fingerprint, b3d9f...2ac8e, and its output risk value changed to 0.0027. The drift was detected when the results analysis module attempted to compare the results of environment B with those of environment A. The module first detected fingerprint inconsistencies and prevented subsequent numerical comparisons. Simultaneously, it reported the specific difference—the graphics processor driver version—to the system log. The data in Table 1 shows that the system of this invention can capture subtle changes in the computing environment and make a definitive judgment on the comparability of the data results based on this information. Next, to verify the system's active environment warping function, the environment warping module was enabled in the experimental group system, and the gold verification image set was processed in environment D. Upon task triggering, the system first generates a local computing environment fingerprint for environment D: b3d9f...2ac8e. The environment normalization module immediately compares this fingerprint with the stored golden fingerprint: a8c7e...1fb9d, and identifies discrepancies in driver versions. Based on these discrepancies, the system automatically triggers a pre-defined driver version downgrade script, reverting the driver version of environment D from 510.47.03 to 470.57.02. After the normalization operation is complete, the system regenerates the local computing environment fingerprint, which is now a8c7e. The result is .1fb9d, consistent with the golden fingerprint. In this normalized environment, the system continues to process image data, and the final output risk value is 0.8824, which is completely consistent with the result in environment A. Experimental data shows that the system of the present invention can not only accurately identify and prevent direct comparison of incomparable data when facing environmental changes by calculating the environmental fingerprint mechanism, but also actively converge the computing environment that deviates from the benchmark to the unified golden fingerprint standard through the environmental normalization module, thereby maintaining the consistency and traceability of data processing results in the distributed system.

[0035] Example 3: This example combines Figs. 1 to 3 This describes a radiological imaging lesion prediction system based on electrical digital data processing, such as... Fig. 1As shown, this illustrates the interaction relationships and data flow paths between various functional units. The system administrator sends a security query request to the security audit module, which retrieves historical fingerprints from the historical fingerprint repository D2 and returns an audit report to the administrator. Simultaneously, the result analysis module provides an analysis report to the administrator based on historical fingerprints and historical prediction results from the prediction result database D3. In a data processing flow, the environment fingerprint generation module first generates a local computing environment fingerprint, which is then sent to the environment standardization module. The environment standardization module compares the fingerprint with the standard environment parameters set by the system administrator and the definition of the gold fingerprint stored in the gold fingerprint database D1. If there are discrepancies, an adjustment instruction is issued. The verified environment fingerprint and the aggregated global model in the federated learning scenario work together to generate the prediction result module. The output of this module is associated with the fingerprint and stored in the prediction result database D3. In the federated learning application, federated learning participants submit their local model updates and fingerprints. The model aggregation module processes the updates and forms the aggregated global model after referring to the standard fingerprint obtained from the gold fingerprint database D1. At the same time, the new environment fingerprints of the participants are stored in the historical fingerprint repository D2.

[0036] like Fig. 2 As shown, the graph uses the inspection cycle as the horizontal axis, the success rate (%) as the main vertical axis on the left, and the number of environmental change detections as the secondary vertical axis on the right. The graph contains three data curves: the solid line marked with circles represents the trend of the fingerprint consistency inspection success rate (%), the dashed line marked with squares represents the trend of the number of environmental change detections, and the dotted line marked with triangles represents the trend of the automatic normalization success rate (%).

[0037] like Fig. 3As shown, the system centers on the core digital data processing of the radiology lesion prediction system, showcasing twelve peripheral functional modules working in tandem with it, and a historical fingerprint data repository. These twelve functional modules include: an environmental fingerprint generation module responsible for acquiring environmental feature metadata and calculating environmental fingerprints; a drift diagnosis module responsible for drift degree calculation and behavioral drift alarms; an integrity verification module responsible for recursive hash verification and chain integrity verification; an environmental quality assessment module responsible for calculating environmental quality scores and contribution weights in federated learning scenarios; a data binding module responsible for associating fingerprints and generating composite evidence hashes; an environmental normalization module responsible for identifying discrepancy sets and performing environmental control over them; a security audit module responsible for retrieving historical fingerprints and generating risk exposure reports; an execution control module responsible for executing authorization decisions and task flow control; a benchmark verification module responsible for executing standardized benchmark tasks and time-consuming stability checks; a result analysis module responsible for fingerprint consistency judgment and environmental inconsistency warnings; a behavioral analysis module responsible for standard image processing and variance index calculation; and a baseline management module responsible for storing golden fingerprints and benchmark configuration management. In addition, a historical fingerprint repository is included for historical fingerprint archiving and metadata management.

[0038] Example 4: This example describes the calibration process of some core operating parameters of the system, as well as the deterministic execution logic of some internal modules. To determine the preset drift threshold in the drift diagnosis module of a specific model, a reproducible standardized engineering calibration procedure needs to be executed. This procedure first prepares a benchmark computing environment, the configuration of which is defined as the source of the gold fingerprint. At the same time, a set of computing environment perturbations known to cause slight changes in the calculation results are prepared, including three different versions of graphics processor drivers and two different versions of the core computing library. The first step of the calibration process is to process the gold verification image set in the benchmark environment using the behavior analysis module, calculate and record the benchmark variance index. The second step involves sequentially introducing each of the aforementioned environmental disturbances into the baseline environment. For each introduced change, the first step is repeated to obtain the variance index under that disturbance environment. and according to Calculate the corresponding drift value This process yields a set of drift sample data caused by known environmental changes. For example, after five different environmental disturbance tests, a drift sample set {0.0015, 0.0021, 0.0018, 0.0035, 0.0025} was obtained. The third step involves statistical analysis of this sample data, calculating its mean and standard deviation. In this sample data set, the mean is 0.00228, and the standard deviation is approximately 0.00074. The fourth step sets the preset drift threshold to the sum of the mean and three times the standard deviation. This threshold is then written into the system's configuration file as a quantitative basis for triggering drift alarm signals in subsequent production operations. Regarding the scoring rules for the environmental quality assessment module in the federated learning scenario, the values ​​of each bonus and deduction item are determined based on a quantitative impact assessment process. The specific value of the safety score, such as -30 points, is determined through the following simulation... The verification process was as follows: First, a federated learning simulation environment was constructed, comprising a central server and ten participants. Second, a standard federated learning aggregation process was run, and the performance metrics of the global model on the standard test set were recorded, achieving an accuracy of 95.0%. Third, one participant, accounting for 10% of the weight, was simulated with a known security vulnerability in its computing environment that could lead to gradient leakage. This participant was then required to submit a model update affected by this vulnerability. After aggregation, the performance of the global model was re-evaluated, revealing a drop in accuracy to 94.1%, a performance decrease of 0.9%. Finally, this performance decrease was multiplied by a preset risk coefficient, set at 33.3, to quantify the impact of this risk on the entire federated system, thus obtaining a deduction value. This method associates each coefficient value in the scoring rules with a measurable potential impact on system performance, providing an engineering basis for parameter setting.

[0039] Furthermore, to ensure that the environment regulation module can deterministically invoke the correct adjustment script after identifying the difference set, its internal implementation employs a key-value lookup table structure. During system initialization, the module loads a configuration file that defines the mapping relationship between environment feature parameter names and specific adjustment script paths. The configuration file contains a record whose key is the string "graphics processor model and driver version," and its value is the string " / opt / scripts / regulate_gpu_driver.sh." When the difference set identified by the environment regulation module includes the difference item "graphics processor model and driver version," it uses this as the key to search the lookup table, obtains the corresponding script path, and invokes the script pointed to by that path to perform the environment adjustment operation. This lookup table-based implementation solidifies the logical relationship between difference identification and specific repair actions, eliminating uncertainties during execution.

[0040] Example 5: This example aims to supplement the deployment plan of the aforementioned technical solution in a specific boundary scenario, to illustrate its built-in mechanism for addressing potential data security risks; In a medical data center running the system of this invention on a large scale, when a third-party software supply chain security bulletin reveals a high-risk remote code execution vulnerability CVE-2022-XXXX in a widely used OpenSSL function library with a version number lower than 1.1.1d, a data security verification requirement arises. That is, it is necessary to immediately determine whether any data in the massive historical image prediction results accumulated by the system over the past few years was generated in a computing environment containing this vulnerable software component. Because the generation environment has known security flaws, the data integrity and security of associated information need to be reassessed; To address this requirement, The system security administrator invokes the security audit module included in the system of this invention. This module operates without interfering with the online real-time data processing flow. It provides a query interface through which the administrator inputs a structured query request defining the vulnerability information, specifically {component: OpenSSL, vulnerable_versions: <1.1.1d}. Upon receiving the request, the security audit module performs a reverse retrieval operation on the historical fingerprint repository. This repository archives the local computing environment fingerprints generated by each historical data processing task and bound to the prediction results. For each historical fingerprint in the repository, the audit module parses it, extracts the version number of the OpenSSL function library recorded therein, and compares it with the version range <1.1.1d defined in the query request.

[0041] After the retrieval and comparison operations are completed, the system automatically generates and outputs a risk exposure report. This report lists two-week time windows in the past three years during which some computing nodes in the system, due to incomplete patch updates, had OpenSSL versions of their operating environments that fell within the scope of the vulnerability. The report further provides a list of unique identifiers for all historical prediction results generated by these affected nodes during these two-week windows. Based on this list, the data management team can immediately isolate and review these potentially affected historical data records, or perform corresponding data tracing and risk assessment processes. A historical data security audit task that is difficult to complete in traditional data processing systems is transformed into an automated query that can be completed within one working day. This audit function utilizes the computing environment information recorded by the system to ensure operational reliability and applies it to the security status tracing of the entire data lifecycle.

[0042] Example 6: This example describes the calibration of the benchmark verification module's operating parameters and its response strategies under boundary conditions. In a scenario where the system of this invention is planned to be deployed to a new virtualized private cloud environment, an engineering challenge is to determine the runtime stability parameters in the verification standard for the benchmark verification module in the new environment, and to verify the module's boundary response capability when facing dynamic resource contention. The static configuration of the new environment has been verified to be consistent with the golden fingerprint, but its dynamic performance baseline is unknown. To address this, the system engineer executed a standardized pre-deployment calibration procedure. First, on the newly deployed and idle virtual machine, the defined standardized benchmark calculation task was executed 100 times continuously and independently. Second, the runtime of each execution was recorded, forming a dataset containing 100 runtime samples. Third, the dataset was statistically analyzed, and its mean was calculated. with standard deviation For example, the calculated mean is 75.2ms and the standard deviation is 1.5ms. Fourthly, based on these statistical results, the verification standard for the benchmark module in this environment is set as follows: the calculated output value must be consistent with the predicted result at the bit level, and the running time must not exceed an upper limit threshold. This upper limit threshold is determined to be... ,Right now This procedure transforms the setting of verification standards from a fixed empirical value into a calibration process based on actual hardware performance that can be reproduced through statistical methods.

[0043] After the above calibration is completed and the system is officially running, when a high-cost image data processing task is requested to be executed, the benchmark verification module in the system of this invention is first triggered according to the procedure to perform a standardized benchmark calculation task as a pre-verification. During a certain execution, due to other virtual machines on the physical host where the virtual machine is located suddenly starting to execute high-intensity calculation tasks, resource contention for CPU and memory bandwidth was caused, resulting in the benchmark calculation task taking 85.4ms to run. After obtaining this time, the execution control module compares it with the upper limit threshold of 79.7ms for this environment stored in the configuration file and determines that the current time has exceeded the threshold. Based on this judgment, the execution control module rejects the execution authorization of subsequent high-cost image data processing tasks and generates an alarm to the system monitoring platform, indicating that the instantaneous calculation benchmark has drifted and the runtime environment is unstable, thereby effectively preventing data processing in an instantaneous environment that may lead to distorted calculation results.

[0044] To further verify the necessity and beneficial effects of the environmental fingerprint association and regularization mechanism introduced in this invention, the following comparative example is set up, which uses a conventional data processing system that does not include the core module described in this invention, and is tested under the same hardware platform and task scenario as in Examples 1 and 2.

[0045] Comparative Example 1: This comparative example aims to simulate the operation of a radiology lesion prediction system without the core modules of environmental fingerprint generation, data binding, and environmental normalization described in this invention. The hardware platform, operating system, basic model (versions V2.1 and V2.2), and computed tomography (CT) image data to be processed (number PID-0753) used in the experiment were completely consistent with the conditions described in Examples 1 and 2 above. The experimental process and results are recorded in Table 2. The process started with the initial processing of the V2.1 version model in the initial computing environment. Subsequently, in order to simulate common operation and maintenance operations in the production environment, the technicians upgraded the server's graphics processor driver from version 470.57.02 to 510.47.03. Finally, in this changed computing environment, the new V2.2 version model was deployed, and the same image data PID-0753 was processed repeatedly.

[0046] Table 2: Functional verification data of conventional systems under environmental change scenarios.

[0047]

[0048] Experimental results show that, in the absence of the computational environment fingerprint generation and binding mechanism proposed in this invention, conventional data processing systems cannot perceive and record changes in their own operating environment. Therefore, when faced with computational result drift caused by environmental evolution, the system, in its design principle, cannot definitively distinguish the source of result differences between changes in algorithm logic and changes in the computational environment. This lack of attribution ability directly leads to incorrect fault diagnosis and causes unnecessary consumption of R&D resources and project delays.

[0049] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0050] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A radiological imaging lesion prediction system based on electro-digital data processing, characterized in that, The system includes: An environment fingerprint generation module is configured to obtain a set of textualized environmental feature metadata of the computing node executing the radiology image data processing task before the task is triggered to generate a prediction result, and generate a local computing environment fingerprint representing the current computing environment based on the environmental feature metadata using a deterministic hash algorithm. A data binding module is configured to associate and store the local computing environment fingerprint as mandatory metadata with the prediction results generated by the task. A baseline management module is configured to store a gold fingerprint, predefined by the administrator, containing a list of standard environment characteristic parameters; An environment conditioning module is configured to execute the following judgment and control rules after the local computing environment fingerprint is generated: parse and compare the local computing environment fingerprint with the golden fingerprint to identify a set of differences between the lists of environmental feature parameters contained in the two; and when the set of differences is not empty, automatically trigger the execution of a preset adjustment operation script based on the content of the set of differences, whose internal instructions correspond to the parameter differences identified in the set of differences, so as to align the environmental feature parameters of the computing node in the direction defined by the golden fingerprint.

2. The radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The characteristic parameters of the computing environment include at least one of the following: operating system name and kernel version, software library version on which data processing depends, graphics processor model and driver version, and central processing unit model and microcode version.

3. The radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The system also includes a result analysis module, which is configured to, when comparing and analyzing at least two prediction results, first determine whether the local computing environment fingerprints associated with the at least two prediction results are consistent, and when it is determined that the local computing environment fingerprints are inconsistent, prevent direct numerical comparison of the at least two prediction results and issue a warning of environment inconsistency.

4. The radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The data binding module is configured to write the local computing environment fingerprint as metadata into the header area of ​​the file containing the prediction results.

5. The radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The data binding module is also configured to generate a composite evidence hash to cryptographically link the current association binding record with the historical association binding records; the system also includes an integrity verification module, which is configured to verify the integrity of the historical record chain by recursively verifying the consistency between the composite evidence hash and the historical hash value in reverse, starting from the latest association binding record.

6. The radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The system also includes: a behavior analysis module configured to process a set of standard image data under a specific local computing environment fingerprint to obtain a set of corresponding prediction results, and to calculate a variance index characterizing the model's macroscopic prediction behavior under that computing environment based on the confidence level of the prediction results; and a drift diagnosis module configured to calculate the drift of the variance index obtained under at least two different local computing environment fingerprints. And in drift degree When the drift exceeds a preset threshold determined based on historical statistical data, an alarm signal is generated indicating that the model's macroscopic predictive behavior has drifted. The drift degree is... The calculation method is as follows ,in For drift degree, To provide a variance metric for fingerprints in a local computing environment. This is a variance metric for fingerprints in another local computing environment.

7. The radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The system also includes: a benchmark verification module configured to execute a predefined standardized benchmark computation task with fixed inputs and whose correct output and benchmark execution time are known in advance, after the environment fingerprint generation module generates the local computing environment fingerprint and before the prediction result is actually generated; and an execution control module configured to determine whether to authorize continued execution to generate the prediction result based on whether an execution result of the benchmark computation task meets a preset verification criterion, wherein the verification criterion includes: the consistency of bit-level comparison between the computation output value of the benchmark computation task and the known correct output result, and the stability of comparing its execution time with the benchmark execution time.

8. The radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The system also includes: a historical fingerprint repository for storing multiple local computing environment fingerprints associated with historical prediction results; and a security audit module configured to: receive a query request that defines at least one software component with a security vulnerability and its version information, and perform a retrieval in the historical fingerprint repository based on the query request.

9. A radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, When applied to federated learning scenarios, the system also includes: an environment quality assessment module, configured to score the local computing environment fingerprint submitted by a federated learning participant and associated with its local model update based on a preset scoring rule, to obtain an environment quality score. The scoring rule includes: adding a standard score if the local computing environment fingerprint matches the golden fingerprint; adding an advancement score if the software component version contained in the local computing environment fingerprint is the latest stable version; and deducting a security score if the local computing environment fingerprint contains a software component version known to have security vulnerabilities; and a model aggregation module, configured to use the environment quality score to weight the model updates of each participant when aggregating the local model updates of multiple participants.

10. A radiological imaging lesion prediction system based on electronic digital data processing according to claim 1, characterized in that, The environment conditioning module is further configured to: after triggering the execution of the adjustment operation script, instruct the environment fingerprint generation module to regenerate the updated local computing environment fingerprint, and compare the updated local computing environment fingerprint with the gold fingerprint again.

Images