CN110660464B

CN110660464B - Intelligent daily quality control method and system for LYSO crystal PET

Info

Publication number: CN110660464B
Application number: CN201911035279.6A
Authority: CN
Inventors: 叶宏伟; 王小状; 王瑶法
Original assignee: Zhejiang Mingfeng Intelligent Medical Technology Co ltd
Current assignee: Zhejiang Mingfeng Intelligent Medical Technology Co ltd
Priority date: 2019-10-29
Filing date: 2019-10-29
Publication date: 2020-10-27
Anticipated expiration: 2039-10-29
Also published as: CN110660464A

Abstract

The invention provides an intelligent daily quality control method of LYSO crystal PET, which relates to the technical field of PET quality control and comprises the following steps: data intelligent acquisition: acquiring parameter information according to preset data, and acquiring and storing LYSO background data; data intelligent analysis and processing: counting information in the data to obtain a three-dimensional heat statistical map, and sending the three-dimensional heat statistical map to a deep learning neural network trained in advance for diagnosis to obtain an abnormal diagnosis result and generate a DailyQC report; data decision and early warning: classifying the diagnosed abnormal conditions, judging whether the abnormal conditions affect the overall performance and reach a danger level, carrying out danger alarm on the abnormal conditions reaching the danger level, and recording the abnormal conditions which do not reach the danger level and do not affect the overall performance. The invention does not need additional radioactive sources and field operation of operators, automatically realizes data acquisition, processing and analysis, automatic identification and decision making and early warning of DailyQC.

Description

Intelligent daily quality control method and system for LYSO crystal PET

Technical Field

The invention relates to a PET quality control technology, in particular to an intelligent daily quality control method and system for LYSO crystal PET.

Background

Positron Emission Tomography (PET) is a nuclear medicine imaging device. A typical PET system is composed primarily of a detector system, electronics system, data correction system, and reconstruction system. The electronic system is the most important component of the whole PET hardware part and is used as a hardware main body to directly determine the basic performance parameters of PET. The PET signal processing process is as follows, the crystal in the detector system mainly collects the gamma photon information generated by radioactive tracer, converts the gamma photon information into a large amount of visible light, converts the visible light into a large amount of electrons by the photoelectric conversion device to form current signals, inputs the current signals into the front-end electronics module to perform waveform amplification and formation, performs noise filtering and discrimination logic judgment and selection and other processing, outputs the current signals to the rear-end electronics module to perform digital processing, finally obtains the position, time and energy physical information and the like of the incident photon, performs scale and correction by the energy time correction system to obtain coincidence case information, and transmits the coincidence case information to the rear end to perform data processing, image reconstruction and the like.

Daily quality control (DailyQC) is an integral part of the PET during its work and use. Because the PET structure contains crystal, electronic components, sensor, detector etc. and the structure is very complicated, the normal use of PET is all influenced to one of them some abnormality appears. Therefore, in order to check whether the machine is in good condition, monitor whether the functional indexes of the key components are qualified or not, ensure that the hardware performance and the software performance of the PET are normal, and regularly and strictly perform DailyQC in daily use.

The current DailyQC for the vast majority of PET cases is roughly as follows: the DailyQC begins by placing prepared radioactive sources (usually Na22, Ge68, etc.) within the data acquisition field of view of the PET machine, some requiring technicians to perform positioning and fixing operations, and some requiring additional phantoms. And then, acquiring PET starting data, wherein the process needs manual operation of technicians, and after the process is finished, the PET processes and analyzes the data, wherein the process comprises evaluation of the abnormal condition of the PET at present, calculation of crystal detector performance parameters such as counting rate, energy peak track value, energy resolution, time resolution and the like, record of temperature and humidity and the like, a diagnosis report of DailyQC is given by combining with a preset evaluation standard, evaluation and warning are given to abnormal parameters and indexes in the diagnosis report, and the abnormal parameters and indexes are fed back to an analysis processing system at the rear end of the PET. After the experiment is completed, technicians remove the radioactive source or the external mold bodies, instruments and the like. The generated diagnosis report needs to be manually evaluated by a technician, and if an abnormality occurs, the diagnosis report needs to be solved and processed according to relevant regulations.

As can be seen from the above, data acquisition of DailyQC requires a radioactive source, purchasing, using and storing of the radioactive source require a whole set of legal and normative procedures, which are very strict to technical requirements of operators (basic requirements must be professional radiological workers), and some links in data acquisition strongly depend on operations and experiences of the technicians, and the whole testing procedure is complicated and complex, and requires personnel to work in a radioactive environment, which is dangerous to some extent.

The present application was made based on this.

Disclosure of Invention

In order to overcome the defects in the prior art, the invention provides an intelligent daily quality control system for PET based on LYSO crystal, no additional radioactive source is needed, no field operation of operators is needed, data acquisition, processing and analysis of DailyQC are automatically realized, automatic identification of hardware conditions is realized, and decision and early warning are made.

In order to achieve the purpose, the technical scheme adopted by the invention is as follows:

the present invention states that all of the following is discussed in terms of LYSO crystal-based PET, and thus PET that is not LYSO crystals is not suitable for use in the context of the present invention.

The invention discloses an intelligent daily quality control method of LYSO crystal PET, which comprises the following steps:

data intelligent acquisition: acquiring parameter information according to preset data, and acquiring and storing LYSO background data;

data intelligent analysis and processing: counting information in the data to obtain a three-dimensional heat statistical map, and sending the three-dimensional heat statistical map to a deep learning neural network trained in advance for diagnosis to obtain an abnormal diagnosis result and generate a DailyQC report;

data decision and early warning: classifying the diagnosed abnormal conditions, judging whether the abnormal conditions affect the overall performance and reach a danger level, carrying out danger alarm on the abnormal conditions reaching the danger level, and recording the abnormal conditions which do not reach the danger level and do not affect the overall performance.

The invention relates to an intelligent daily quality control system of LYSO crystal PET, which comprises:

the intelligent data acquisition system is used for acquiring parameter information according to preset data, and acquiring and storing LYSO background data;

the data intelligent analysis processing system is used for counting the information in the data to obtain a three-dimensional heat statistical map, sending the three-dimensional heat statistical map to a deep learning neural network trained in advance for diagnosis, obtaining an abnormal diagnosis result and generating a DailyQC report;

and the decision and early warning system is used for classifying the diagnosed abnormal conditions, judging whether the abnormal conditions affect the overall performance and reach the danger level, carrying out danger alarm on the abnormal conditions reaching the danger level, and recording the abnormal conditions which do not reach the danger level and cannot affect the overall performance.

The invention utilizes the radioactivity background of natural radioactive lutetium 176 element (Lu-176) contained in the LYSO crystal, does not need to place additional radioactive sources, and does not need to be operated by workers. The flow of data acquisition is roughly as follows: a table of information relating to the parameters is prepared prior to data acquisition and stored in the PET system, and data acquisition is initiated by setting conditions for data acquisition in the PET electronics based on the parameters in the stored table, selecting an appropriate time (e.g., at night or during a time period when the PET is not undergoing a clinical scan) for data acquisition of the LYSO natural radioactivity background. The detailed process is described in the detailed description.

The data intelligent analysis processing system of the invention: and carrying out intelligent data analysis on various parameters in the acquired data by utilizing a deep learning network. The process is completed by fully utilizing the deep learning neural network, and the neural network required by various parameters is preset and stored in the PET system. The deep learning network based on the Unet is used for monitoring and analyzing the abnormal conditions and parameter indexes of the PET detector hardware, but the invention is not limited to other deep learning networks.

The decision and early warning system of the invention comprises: the preset abnormal danger level information is stored in the PET system, and after the results obtained from the intelligent analysis processing system are input into the process, the decision and processing are carried out one by one according to the set level analysis rules, and the condition of high danger level is pre-warned.

The principle and the beneficial technical effects of the invention are as follows: the background generated by the natural radioactive lutetium 176 element in the LYSO crystal is used as the radioactive source of the DailyQC, other radioactive sources, die bodies, instruments and the like do not need to be arranged externally, data can be automatically collected and analyzed, the deep learning network is used for detecting the abnormality of the detector in the DailyQC, the parameter result of the data is intelligently evaluated, and a diagnosis report is given. The whole process does not need technicians to participate in the operation on site, and a large amount of time and resource cost are saved.

The invention provides an intelligent LYSO-based PET intelligent daily control method and system, which automatically and intelligently realize the process, do not need a radioactive source, a die body, instruments and the like, do not need the participation of technicians, automatically acquire experimental data, intelligently detect and evaluate by a deep learning method, and analyze and process results.

Drawings

FIG. 1 is a block diagram of an intelligent daily quality control system for LYSO crystal PET according to the present embodiment;

FIG. 2 is a flow chart of the operation of the intelligent data acquisition system of the present embodiment;

FIG. 3 is a flowchart of the operation of the intelligent data analysis and processing system of the present embodiment;

fig. 4 is a net neural network diagram for predicting whether the three-dimensional heat statistics histogram is abnormal or not, which is adopted in the present embodiment;

FIG. 5 is a three-dimensional heat statistical histogram of a PET anomalous count condition for LYSO background data acquisition as provided in this example.

FIG. 6 is a position reference diagram of PET anomaly information obtained by deep learning network analysis and prediction in the data intelligent analysis and processing system according to the present embodiment, and a white area in the diagram is a crystal coordinate position of a detector anomaly;

fig. 7 is a flowchart illustrating the operation of the decision-making and early-warning system according to the present embodiment.

Detailed Description

In order to make the technical means of the present invention and the technical effects achieved thereby clearer and more complete, an embodiment is provided, and the following detailed description is made with reference to the accompanying drawings:

as shown in fig. 1, the present embodiment provides an intelligent daily quality control system based on LYSO crystal PET, which mainly comprises three parts: the system comprises a data intelligent acquisition system 1, a data intelligent analysis processing system 2 and a decision and early warning system 3.

Wherein the hardware parts and devices of the intelligent data acquisition system 1 are substantially all from the PET data acquisition system.

Fig. 2 is a schematic view of a work flow of the intelligent data acquisition system in this embodiment, where the work flow includes:

step S101, starting data acquisition;

step S102, reading preset data acquisition parameter information; note that in the present solution, in order to implement intelligent automation, parameter information and the like of data acquisition need to be stored in the PET system in advance. The parameter information of data acquisition includes, but is not limited to, parameter information conditions of general data acquisition, such as energy window, time window, scanning field of view (FOV), and the like;

step S103, writing the parameter information read in the step S102 into a hardware register of PET electronics, so as to facilitate subsequent data acquisition;

step S104, sending a data acquisition starting instruction to a PET electronic system;

s105, the electronic system collects, discriminates and selects LYSO background data meeting the conditions according to the conditions in the register;

step S106, judging whether the current sample data is required by a preset condition, if not, turning to S104; if the answer is yes, continuing to the next step;

step S107, sending an instruction for terminating data acquisition to the electronics;

step S108, storing the collected data in a storable medium;

after the above process is completed, background data of LYSO satisfying the condition can be obtained. And (4) transmitting the data to a data intelligent analysis processing system for intelligent analysis processing to obtain a basic diagnosis result of DailyQC. The intelligent data analysis and processing system mainly utilizes a deep learning neural network which is trained and tested to be qualified in advance to carry out anomaly detection and processing, statistics is carried out on an anomaly detection unit, and important parameters such as energy peak positions, energy resolution and the like are identified.

Referring to fig. 3, a schematic diagram of a work flow of the data intelligent analysis processing system of the present invention is shown, wherein the work flow includes:

step S201, background data of LYSO stored in the intelligent data acquisition system is read;

and S202, classifying information in the read data according to PET geometric arrangement information, and performing three-dimensional statistical processing to obtain a three-dimensional heat statistical histogram. In the present embodiment, for the explanation, only the PET abnormal count condition is described, but the information of the data is not limited thereto, and other information such as energy peak position, energy resolution, etc. may be processed similarly;

step S203, the obtained heat degree statistical histogram is taken as input and is transmitted to a neural network for prediction;

for explanation, referring to fig. 4, a pnet neural network used in the present embodiment to predict whether the three-dimensional heat statistic histogram is abnormal or not is illustrated, but the present invention is not limited to using only such a neural network. It should be noted that the network parameters need to be trained in advance until the network parameters are stable and reliable, and the embodiment only uses the network to predict here, and prediction and parameter debugging of the network are not elaborated in the present invention. The processing flow of the network is roughly as follows: and inputting the heat statistical histogram, continuously performing 4 times of downsampling and calculating a 3x3 two-dimensional convolution and a linear rectification function by doubling the feature layer, continuously performing upsampling, splicing the feature layers with the same size before downsampling, and finally outputting a prediction graph with the original size.

Reference is now made to FIG. 5, which is a three-dimensional heat statistical histogram of abnormal counts of PET acquired from LYSO background data provided in the present example.

Referring to fig. 6, a position reference diagram of PET anomaly information predicted by deep learning network analysis is shown, and a white area in the diagram is a crystal coordinate position of a detector anomaly;

step S204, obtaining and storing the result of the abnormality diagnosis;

step S205, generating a DailyQC diagnosis report;

after the above process is completed, the abnormal diagnosis result obtained in S204 is sent to the decision and early warning system 300, and the result is analyzed and the decision is issued to the central console;

referring to fig. 7, a schematic flowchart of the decision-making and early-warning system 300 according to the present invention is shown, wherein the workflow includes:

step S301, inputting an intelligently diagnosed abnormal result;

step S302, classifying and deciding the diagnosed abnormal result;

step S303, judging whether an abnormality is met, if the answer is negative, directly jumping to step S313, and if the answer is positive, turning to step S304;

step S304, judging whether the point of the PET crystal is abnormal or not, if the answer is affirmative, directly jumping to step S309, and if the answer is negative, turning to step S305;

step S305, judging whether the line of the PET crystal is abnormal or not, if the answer is positive, directly jumping to step S309, and if the answer is negative, turning to step S306;

step S306, judging whether the condition that the crystal submodule of the PET crystal is abnormal or not, if the answer is positive, directly jumping to the step S309, and if the answer is negative, turning to the step S307;

step S307, judging whether the condition that the detector module of the PET crystal is abnormal is judged, if yes, directly jumping to step S309, and if not, turning to step S308;

step S308, the other abnormal conditions are qualified as other unknown types of abnormal conditions;

step S309, judging whether the number and distribution of the abnormalities can affect the subsequent data acquisition, the parameter index performance of the PET and the like, if the answer is negative, directly jumping to step S312, and if the answer is positive, turning to step S310;

step S310, judging whether the abnormity constitutes danger, if yes, turning to step S311, if no, turning to step S312;

step S311, sending an alarm to the central control system;

step S312, recording and saving the abnormal condition into a report;

in step S313, the entire flow ends.

The above description is only an example of the present invention and should not be taken as limiting the present disclosure, and any modifications, equivalents, improvements and the like made within the principle and spirit of the present invention should be included in the scope of the present invention.

The above description is provided for the purpose of further elaboration of the technical solutions provided in connection with the preferred embodiments of the present invention, and it should not be understood that the embodiments of the present invention are limited to the above description, and it should be understood that various simple deductions or substitutions can be made by those skilled in the art without departing from the spirit of the present invention, and all such alternatives are included in the scope of the present invention.

Claims

1. An intelligent daily quality control method of LYSO crystal PET is characterized by comprising the following steps:

s100, intelligently acquiring data: acquiring parameter information according to preset data, and acquiring and storing LYSO background data;

s200, intelligent data analysis and processing: counting information in the data to obtain a three-dimensional heat statistical map, and sending the three-dimensional heat statistical map to a deep learning neural network trained in advance for diagnosis to obtain an abnormal diagnosis result and generate a DailyQC report;

s300, data decision and early warning: classifying the diagnosed abnormal conditions, judging whether the abnormal conditions affect the overall performance and reach a danger level, performing danger alarm on the abnormal conditions reaching the danger level, and recording the abnormal conditions which do not reach the danger level and do not affect the overall performance;

the processing flow of the deep learning neural network comprises the following steps: inputting the heat statistical histogram, continuously performing 4 times of down-sampling and feature layer doubling 3x3 two-dimensional convolution and linear rectification function calculation, continuously performing up-sampling, splicing feature layers with the same size before down-sampling, and finally outputting a prediction graph with the original size;

the method specifically comprises the following steps of classifying the diagnosed abnormal conditions, judging whether the abnormal conditions affect the overall performance and reach a danger level, and specifically judging whether the abnormal results are the following conditions in sequence: the method comprises the steps of determining the abnormal conditions except the point abnormal condition of the PET crystal, the line abnormal condition of the PET crystal, the crystal sub-module abnormal condition of the PET crystal and the detector module abnormal condition of the PET crystal, determining the abnormal conditions except the above conditions as other unknown types of abnormal conditions, judging whether the number and distribution of the abnormal conditions influence the subsequent data acquisition and the parameter index performance of the PET, recording and storing the abnormal conditions into a report if the abnormal conditions do not influence the subsequent data acquisition and the parameter index performance of the PET, and further judging whether the abnormal conditions reach a danger level if the abnormal conditions do not influence the parameter index performance of the PET crystal.

2. The method of claim 1, wherein the method comprises the steps of: in step S100, the preset data acquisition information includes an access condition, a time window, an energy window, and a scanning field of view.

3. The method of claim 1, wherein the method comprises the steps of: the step S100 specifically includes the following:

step S101, starting data acquisition;

step S102, reading preset data acquisition parameter information;

step S103, writing the parameter information read in the step S102 into a hardware register of the PET electronics;

step S106, judging whether the current case data is required by a preset condition, if not, turning to step S104; if the answer is yes, continuing to the next step;

and step S108, storing the acquired data into a storable medium.

4. The method of claim 1, wherein the method comprises the steps of: the step S200 specifically includes the following steps:

step S202, information in read data is classified according to PET geometric arrangement information, and three-dimensional statistical processing is carried out to obtain a three-dimensional heat statistical histogram;

step S203, the obtained three-dimensional heat degree statistical histogram is taken as input and is transmitted to a deep learning neural network which is trained in advance for prediction;

step S204, obtaining and storing the result of the abnormality diagnosis;

and step S205, generating a DailyQC diagnosis report.

5. The method of claim 1, wherein the method comprises the steps of: the step S300 specifically includes the following steps:

step S301, inputting an intelligently diagnosed abnormal result;

step S302, classifying and deciding the diagnosed abnormal result;

step S309, judging whether the number and distribution of the abnormalities can affect the subsequent data acquisition and the parameter index performance of the PET, if the answer is negative, directly jumping to step S312, and if the answer is positive, turning to step S310;

step S311, sending an alarm to the central control system;

step S312, recording and saving the abnormal condition into a report;

in step S313, the entire flow ends.

6. An intelligent daily quality control system for LYSO crystal PET, comprising: