PET / CT personalized parameter calculation system with adaptive model optimization function
The PET/CT personalized parameter calculation system with adaptive model optimization function solves the problems of unstable image quality and low scanning efficiency in the existing technology, realizes uniform image quality and efficient scanning, and has continuous adaptation and model transferability capabilities.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHONGSHAN HOSPITAL FUDAN UNIV
- Filing Date
- 2026-04-14
- Publication Date
- 2026-07-31
AI Technical Summary
Current PET/CT clinical scanning protocols are rigid and cannot adapt to different equipment, populations, and drug characteristics, resulting in unstable image quality and low scanning efficiency.
The PET/CT personalized parameter calculation system with adaptive model optimization function realizes dynamic, closed-loop parameter adjustment and optimization through patient feature data extraction, personalized parameter calculation, system integration interface, image quality feedback and adaptive optimization engine.
It achieves uniformity and stability of image quality, improves scanning efficiency, reduces human error, and has continuous adaptive capability and model transferability.
Smart Images
Figure CN122492568A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to an intelligent calculation and optimization system for personalized acquisition parameters in positron emission tomography (PET) / computed tomography (CT) systems, belonging to the interdisciplinary field of medical imaging technology and medical artificial intelligence. Background Technology
[0002] Currently, the mainstream protocol for PET / CT clinical scanning follows international guidelines (such as EANM and SNMMI), employing a linear model based on patient weight to calculate the amount of radiopharmaceutical (e.g., 18 The injection dosage of F-FDG is specified. Simultaneously, image acquisition typically employs a fixed, empirical time scheme. This combination of a "weight-dose" model and fixed acquisition time constitutes the widely implemented standard operating procedure. However, this mainstream approach has the following inherent drawbacks:
[0003] First, simple weight models cannot distinguish the different effects of lean body tissue and adipose tissue on the distribution and attenuation of radiopharmaceuticals, resulting in unstable image quality in obese or abnormally muscular patients. Secondly, the fixed acquisition time ignores individual differences in patient body size, equipment sensitivity, and radiopharmaceutical kinetics, resulting in low scanning efficiency (too long for thin patients) or insufficient image quality (too short for tall patients). More importantly, the existing mainstream solution is a static, open-loop system. Once its parameters (such as drug activity per kilogram of body weight and fixed acquisition minutes) are set, they cannot be adjusted and optimized according to the actual imaging effect, and it lacks the ability to adapt to different devices, different populations and different drug characteristics.
[0004] Therefore, although existing guidelines provide a basic framework, the contradiction between their rigid models and the complex individualized needs in clinical practice is becoming increasingly prominent. There is an urgent need for a personalized parameter calculation system that can achieve dynamic, closed-loop, and adaptive operation to achieve homogeneous, stable image quality that meets diagnostic needs and efficient imaging examinations. Summary of the Invention
[0005] The technical problem this invention aims to solve is that the rigid models in existing PET / CT clinical scanning guidelines cannot meet clinical needs.
[0006] To address the aforementioned technical problems, the present invention discloses a personalized parameter calculation system for PET / CT with adaptive model optimization function, characterized in that it includes: The patient feature data extraction module is used to obtain multiple patient feature data and then send the integrated patient feature data to the personalized parameter calculation module. A personalized parameter calculation module maintains a parameterized model library, which stores different adjustable parameter calculation models for different radiopharmaceuticals. Each adjustable parameter calculation model is a calculation framework defined by a set of learnable model parameters. After obtaining the patient's feature data, the personalized parameter calculation module selects a corresponding adjustable parameter calculation model from the parameterized model library based on the radiopharmaceutical currently used by the patient. Using the patient's feature data as input, the adjustable parameter calculation model calculates and outputs personalized PET / CT clinical scan recommended parameters suitable for the current patient. The system integration interface is used to directly and securely push the personalized PET / CT clinical scan recommendation parameters generated by the personalized parameter calculation module to the PET / CT equipment control system. It is also used to read the equipment status of the PET / CT equipment and automatically load the scanning protocol of the PET / CT equipment. The image quality feedback acquisition module is used to analyze the image quality after the PET / CT device completes the scan of the current patient based on the personalized PET / CT clinical scan recommendation parameters, and obtain quantitative quality feedback data composed of objective evaluation parameter values generated based on the analysis results. An adaptive optimization engine is used to compare the quality feedback data generated by the image quality feedback acquisition module with a preset target value. If there is a significant deviation between the quality feedback data and the target value, the corresponding adjustable parameter calculation model is trained using a corresponding optimization algorithm, with each completed case after the PET / CT personalized parameter calculation system is put into use as a training sample. This completes the fine-tuning of the model parameters, and the optimized adjustable parameter calculation model is updated to the parameterized model library.
[0007] Preferably, the patient feature data extraction module is integrated with an existing hospital information collection system to automatically retrieve the feature data from the hospital information collection system.
[0008] Preferably, the personalized PET / CT clinical scan recommended parameters calculated and output by the adjustable parameter calculation model include injection activity and acquisition time.
[0009] Preferably, during the initialization of the PET / CT personalized parameter calculation system, the model parameters of the adjustable parameter calculation model in the parameterized model library are general model parameters or initial model parameters set based on relevant literature.
[0010] Preferably, the adaptive optimization engine optimizes the adjustable parameter calculation model using the following steps: Step 1: The adaptive optimization engine obtains the quality feedback data, extracts the corresponding target indicators, and then calculates the deviation between the quality feedback data and the corresponding target indicators. ; Step 2: Determine the deviation value Is it within the preset allowable range? If so, the current optimization process ends, and the corresponding adjustable parameter calculation model in the parameterized model library remains unchanged; If not, proceed to step 3; Step 3: Based on the deviation value Obtain the corresponding deviation pattern, and obtain the corresponding optimization strategy based on the different deviation patterns; Step 4: Directly use the optimization strategy obtained in the previous step to perform optimization calculations and obtain the optimized model parameters; Or, for the deviation value The root causes of the deviation are evaluated, and the optimization strategy obtained in the previous step is adaptively adjusted based on the evaluation results. The adjusted optimization strategy is then used to perform optimization calculations to obtain the optimized model parameters. Step 5: After optimizing and updating the adjustable parameter calculation model using the optimized model parameters obtained in the previous step, perform security verification and model performance verification on the adjusted adjustable parameter calculation model: If the verification is successful, the optimized adjustable parameter calculation model will be updated to the parameterized model library. If the verification fails, adjust the optimization strategy and return to step 4 to recalculate the optimization until the verification is passed.
[0011] Preferably, in step 4, when the adaptive optimization engine optimizes the adjustable parameter calculation model, each completed case after the PET / CT personalized parameter calculation system is put into use is used as a training sample. The label of the training sample is the actual image quality. After all the training samples constitute the training dataset, the adjustable parameter calculation model is trained based on the training dataset using the optimization strategy to obtain the optimized model parameters.
[0012] Preferably, the adaptive optimization engine continuously utilizes newly acquired training samples to iteratively update model parameters with the goal of optimizing image quality, thereby continuously optimizing the adjustable parameter calculation model. After a period of operation, a calculation model with personalized model parameters that is most suitable for the current usage environment and the specific patient population in the current usage environment is obtained.
[0013] The system disclosed in this invention can not only calculate personalized parameters based on the patient's physical characteristics, but also automatically adjust and optimize its internal calculation model by continuously learning the actual effect of each scan, thereby adapting to specific equipment and specific populations, and ultimately achieving uniformity and optimization of image quality.
[0014] Compared with existing technical solutions, the present invention has the following beneficial effects: 1. Image quality uniformity: Through closed-loop feedback, the system can automatically compensate for factors such as equipment performance degradation and differences in technician operation, so that the image quality obtained at different times, for different patients or for the same patient remains stable. 2. Continuous adaptive capability: When new equipment, new radiopharmaceuticals, or changes in patient population characteristics are introduced, the system does not require manual remodeling and can automatically adapt through continuous learning; 3. Seamless workflow integration: Fully automated data flow and parameter loading greatly improve technicians' work efficiency and reduce human error; 4. Model portability and security: The optimized model parameters can be exported in an encrypted manner and, with authorization, can be securely applied to other similar devices, enabling the rapid replication of high-quality experience while protecting the data assets of users of this invention. Attached Figure Description
[0015] Figure 1 This is a schematic diagram illustrating the overall architecture of the system of the present invention and its relationship with the external environment; Figure 2 A flowchart illustrating the internal working principle of the adaptive optimization engine; Figure 3 This is a timing diagram of the workflow of the system of the present invention integrated with PET / CT equipment. In the diagram, "→" represents input and "←" represents output. Detailed Implementation
[0016] The present invention will be further illustrated below with reference to specific embodiments. It should be understood that these embodiments are for illustrative purposes only and are not intended to limit the scope of the invention. Furthermore, it should be understood that after reading the teachings of this invention, those skilled in the art can make various alterations or modifications to the invention, and these equivalent forms also fall within the scope defined by the appended claims.
[0017] like Figure 1 As shown in the figure, a personalized PET / CT parameter calculation system with adaptive model optimization function disclosed in this embodiment of the invention includes: 1) Patient Feature Data Extraction Module The patient feature data extraction module acquires patient feature data from different dimensions and then sends the integrated patient feature data to the personalized parameter calculation module. In a preferred embodiment of the invention, the patient feature data extraction module is deeply integrated with the hospital information collection system, enabling automatic retrieval of patient information while also supporting manual entry of patient information. In an optional embodiment of the invention, the hospital information collection system includes: a Hospital Information System (HIS) providing a database containing basic patient information; an Electronic Medical Record System (EMR) providing a database containing patient medical history, diagnosis, and laboratory test data; and a Medical Information System (HIS) providing a database containing patient imaging data and reports. The patient feature data extraction module automatically acquires structured data of the same patient from the Hospital Information System, Electronic Medical Record System, and Medical Information System through standardized data interfaces, including but not limited to height, weight, lean body mass, and BMI.
[0018] II) Personalized Parameter Calculation Module The personalized parameter calculation module maintains a parameterized model library, which stores adjustable parameter calculation models for different radiopharmaceuticals. Each adjustable parameter calculation model is not a fixed formula, but rather a calculation framework defined by a set of learnable model parameters. For example, the parameterized model library stores models for... 18 Model A of F-FDG, targeting 68 Model B of Ga-DOTATATE, targeting 68 Ga-FAPI model C, targeting 68 Ga-PSMA model D, for 68 Models E, A, B, C, D, and E in Ga-Pentixafor represent different computational frameworks defined by their respective learnable model parameters. After obtaining patient characteristic data, the personalized parameter calculation module selects a corresponding adjustable parameter calculation model from the parametric model library based on the radiopharmaceutical used by the patient. Using the patient characteristic data as model input, the adjustable parameter calculation model calculates personalized PET / CT clinical scan recommended parameters suitable for the current patient, including but not limited to injection activity and acquisition time. The personalized parameter calculation module then pushes these personalized PET / CT clinical scan recommended parameters to the PET / CT equipment control system through the system integration interface.
[0019] (iii) System Integration Interface The system integration interface is used to directly and securely push the personalized PET / CT clinical scan recommendation parameters generated by the personalized parameter calculation module to the PET / CT equipment control system. It is also used to read the equipment status of the PET / CT equipment and automatically load the scanning protocol of the PET / CT equipment to achieve "two-way dialogue" with the PET / CT equipment control system.
[0020] IV) Image Quality Feedback Acquisition Module After the PET / CT scanner completes the scan, the image quality feedback acquisition module automatically analyzes the image quality based on pre-set objective evaluation parameters and generates quantified quality feedback data. In a preferred embodiment of the present invention, the objective evaluation parameters may include at least one of signal-to-noise ratio, noise, and physician subjective evaluation score (5-likert score).
[0021] (v) Adaptive Optimization Engine The adaptive optimization engine is the core innovation of this invention. It is used to compare the quality feedback data generated by the image quality feedback acquisition module with the preset target. If there is a significant deviation, the optimization algorithm (e.g., gradient descent, reinforcement learning, etc.) is activated to fine-tune the model parameters of the corresponding adjustable parameter calculation model in the parameterized model library, so as to optimize the adjustable parameter calculation model.
[0022] In one embodiment of the present invention, an optional implementation is that, during the initialization of the entire system, the model parameters of the adjustable parameter calculation model in the parameterized model library are either general model parameters or initial model parameters set based on relevant literature.
[0023] Combination Figure 2 In a preferred embodiment of the present invention, the adaptive optimization engine optimizes the adjustable parameter calculation model using the following steps: Step 1: The adaptive optimization engine obtains quality feedback data, extracts the corresponding target indicators, and then calculates the deviation between the quality feedback data and the corresponding target indicators. ; Step 2: Determine the deviation value Is it within the preset allowable range? If so, the current optimization process ends, and the corresponding adjustable parameter calculation model in the parameterized model library remains unchanged; If not, proceed to step 3; Step 3: Based on the deviation value Obtain the corresponding deviation patterns (e.g., low signal-to-noise ratio and / or high noise), and obtain the corresponding optimization strategies based on different deviation patterns; Step 4: Directly use the optimization strategy obtained in the previous step to perform optimization calculations and obtain the optimized model parameters; Or, for the deviation value The root causes of the deviations are evaluated, and the optimization strategy obtained in the previous step is adaptively adjusted based on the evaluation results (e.g., using different optimization methods or setting different optimization objectives). The adjusted optimization strategy is then used to perform optimization calculations to obtain the optimized model parameters. In this invention, the adaptive optimization engine uses each completed case after the system is put into use as a training sample when optimizing the adjustable parameter calculation model. The label of the training sample is the actual image quality. After all the training samples form a training dataset, the adjustable parameter calculation model is trained based on this training dataset using optimization strategies to obtain optimized model parameters. In a preferred embodiment of the invention, the adaptive optimization engine continuously uses newly acquired training samples, with the goal of optimizing image quality, to iteratively update the model parameters, so that the adjustable parameter calculation model is continuously optimized. After a period of operation (such as hundreds of cases), the system will "learn" a calculation model with personalized model parameters that is most suitable for the current use environment and the specific patient population in the current use environment, forming a unique "digital asset".
[0024] Step 5: After optimizing and updating the adjustable parameter calculation model using the optimized model parameters obtained in the previous step, perform security verification and model performance verification on the adjusted adjustable parameter calculation model: If the verification is successful, the optimized adjustable parameter calculation model will be updated to the parameterized model library. If the verification fails, adjust the optimization strategy and return to step 4 to recalculate the optimization until the verification is passed.
[0025] Figure 3 This illustration shows the workflow between the PET / CT personalized parameter calculation system with adaptive model optimization function disclosed in the embodiments of the present invention and the PET / CT equipment, which will not be described in detail here.
[0026] The following specific examples further illustrate a personalized PET / CT parameter calculation system disclosed in this invention.
[0027] Example 1 A hospital recently installed a uEXPLORER PET / CT scanner and deployed the system disclosed in this embodiment of the invention. During system initialization, the parameterized model library contains parameters for... 18The adjustable parameter calculation model for F-FDG uses parameters based on population data from publicly available literature, defined as the initial calculation model. For the first 100 patients, the system uses this initial calculation model to calculate the dose and time for PET / CT scans. After each patient's scan, the image quality feedback module automatically measures the liver SNR. The adaptive optimization engine found that the actual average liver SNR for the first 50 patients was 13.5, lower than the system's preset target value of 15.0, and the deviation was statistically significant. The adaptive optimization engine then activated a reinforcement learning algorithm as an optimization strategy, using the training dataset formed by the newly obtained training samples to train the initial calculation model, achieving minor adjustments to the model parameters and completing the adjustment of the initial calculation model. In patients 51-100, the system uses the adjusted calculation to obtain the dose and time for PET / CT scans, and the average liver SNR reported by the image quality feedback module improved to 14.8, closer to the target value. Through continuous learning, the system becomes increasingly stable during localization.
[0028] Example 2 For use 68 PET / CT examinations using Ga-PSMA, creating models targeting Ga-PSMA in a parametric model library. 68 The Ga-PSMA "PSMA model" initially referenced the model for a similar drug, DOTATATE. Due to the different biological distribution of PSMA compared to DOTATATE, the image quality feedback from the image quality feedback module for the first 20 examinations (primarily based on prostate cancer lesion contrast) was unstable. The adaptive optimization engine rapidly adjusted the PSMA model parameters based on the feedback from these 20 cases, enabling it to quickly reach and stabilize at the ideal image quality level in subsequent cases, demonstrating the system's rapid adaptability to new drugs.
[0029] Example 3 The PET / CT personalized parameter calculation system disclosed in this invention has been running for one year in a certain center A. Its parameterized model library contains parameters specific to... 18 The adjustable parameter calculation model for F-FDG has been highly optimized. Center B introduced the same model of equipment as Center A. Center A can encrypt and export the model parameters (not the original patient data) from its optimized parametric model library, and import them into Center B's PET / CT personalized parameter calculation system's parametric model library as the initial model after authorization. Based on this, Center B's PET / CT personalized parameter calculation system only needs minor adjustments with a small number of cases to achieve an image quality level comparable to Center A, greatly shortening the "break-in period" of the new equipment and ensuring the consistency of image quality across multiple centers.
Claims
1. A personalized parameter calculation system for PET / CT with adaptive model optimization function, characterized in that, include: The patient feature data extraction module is used to obtain multiple patient feature data and then send the integrated patient feature data to the personalized parameter calculation module. A personalized parameter calculation module maintains a parameterized model library, which stores different adjustable parameter calculation models for different radiopharmaceuticals. Each adjustable parameter calculation model is a calculation framework defined by a set of learnable model parameters. After obtaining the patient characteristic data, the personalized parameter calculation module selects a corresponding adjustable parameter calculation model from the parameterized model library based on the radiopharmaceutical currently used by the patient. The patient characteristic data is then used as the input of the adjustable parameter calculation model, which calculates and outputs personalized PET / CT clinical scan recommended parameters suitable for the current patient. The system integration interface is used to directly and securely push the personalized PET / CT clinical scan recommendation parameters generated by the personalized parameter calculation module to the PET / CT equipment control system. It is also used to read the equipment status of the PET / CT equipment and automatically load the scanning protocol of the PET / CT equipment. The image quality feedback acquisition module is used to analyze the image quality after the PET / CT device completes the scan of the current patient based on the personalized PET / CT clinical scan recommendation parameters, and obtain quantitative quality feedback data composed of objective evaluation parameter values generated based on the analysis results. An adaptive optimization engine is used to compare the quality feedback data generated by the image quality feedback acquisition module with a preset target value; If there is a significant deviation between the quality feedback data and the target value, then the corresponding adjustable parameter calculation model is trained using the corresponding optimization algorithm, with each completed case after the PET / CT personalized parameter calculation system is put into use as a training sample, to complete the fine-tuning of the model parameters, and the optimized adjustable parameter calculation model is updated to the parameterized model library.
2. The PET / CT personalized parameter calculation system with adaptive model optimization function as described in claim 1, characterized in that, The patient feature data extraction module is integrated with the existing hospital information collection system and automatically retrieves the feature data from the hospital information collection system.
3. The PET / CT personalized parameter calculation system with adaptive model optimization function as described in claim 1, characterized in that, The personalized PET / CT clinical scan recommended parameters calculated and output by the adjustable parameter calculation model include injection activity and acquisition time.
4. The PET / CT personalized parameter calculation system with adaptive model optimization function as described in claim 1, characterized in that, When the PET / CT personalized parameter calculation system is initialized, the model parameters of the adjustable parameter calculation model in the parameterized model library are either general model parameters or initial model parameters set based on relevant literature.
5. The PET / CT personalized parameter calculation system with adaptive model optimization function as described in claim 1, characterized in that, The adaptive optimization engine optimizes the adjustable parameter calculation model using the following steps: Step 1: The adaptive optimization engine obtains the quality feedback data, extracts the corresponding target indicators, and then calculates the deviation between the quality feedback data and the corresponding target indicators. ; Step 2: Determine the deviation value Is it within the preset allowable range? If so, the current optimization process ends, and the corresponding adjustable parameter calculation model in the parameterized model library remains unchanged. If not, proceed to step 3; Step 3: Based on the deviation value Obtain the corresponding deviation pattern, and obtain the corresponding optimization strategy based on the different deviation patterns; Step 4: Directly use the optimization strategy obtained in the previous step to perform optimization calculations and obtain the optimized model parameters; Or, for the deviation value The root causes of the deviation are evaluated, and the optimization strategy obtained in the previous step is adaptively adjusted based on the evaluation results. The adjusted optimization strategy is then used to perform optimization calculations to obtain the optimized model parameters. Step 5: After optimizing and updating the adjustable parameter calculation model using the optimized model parameters obtained in the previous step, perform security verification and model performance verification on the adjusted adjustable parameter calculation model: If the verification is successful, the optimized adjustable parameter calculation model will be updated to the parameterized model library. If the verification fails, adjust the optimization strategy and return to step 4 to recalculate the optimization until the verification is passed.
6. The PET / CT personalized parameter calculation system with adaptive model optimization function as described in claim 4, characterized in that, In step 4, when the adaptive optimization engine optimizes the adjustable parameter calculation model, it uses each completed case after the PET / CT personalized parameter calculation system is put into use as a training sample. The label of the training sample is the actual image quality. After all the training samples constitute the training dataset, the adjustable parameter calculation model is trained based on the training dataset using the optimization strategy to obtain the optimized model parameters.
7. A PET / CT personalized parameter calculation system with adaptive model optimization function as described in claim 6, characterized in that, The adaptive optimization engine continuously utilizes newly acquired training samples, aiming at optimizing image quality, and iteratively updates the model parameters, thereby continuously optimizing the adjustable parameter calculation model. After a period of operation, it obtains a calculation model with personalized model parameters that is most suitable for the current usage environment and the specific patient population in the current usage environment.