A method and system for online quality analysis and feedback of magnetic resonance spectroscopy data

Abstract

This invention discloses an online quality analysis and feedback method and system for magnetic resonance spectroscopy (MRS) data. After a single MRS acquisition is completed, the MRS source data, acquisition parameters, and object identification information for the current inspection task are obtained through a PACS system or its associated gateway. A single acquisition analysis task is established, and the evaluation context is determined. Data integrity and resolvability checks, preprocessing, spectral reconstruction, and fitting are performed sequentially. Through two levels of gating (data-level and analysis-level), continuous quality indicators are extracted only after both levels of gating are passed. Based on the evaluation context, a threshold set is matched, and the indicators are subjected to directional unification, threshold normalization, and hierarchical fusion to obtain the quality judgment result. Before the end of the same inspection session, the judgment result, its basis, and the obtainable quantitative results are output. If the result is unqualified, the cause of the anomaly is located, and adjustment suggestions are output for parameter correction or supplementary acquisition. This invention can reduce the feedback delay caused by offline analysis and improve the reliability and traceability of MRS quality evaluation.

Description

Technical Field

[0001] This invention relates to the fields of medical image data processing, magnetic resonance spectroscopy quality control, and medical information systems, specifically to a method and system for online quality analysis and feedback of magnetic resonance spectroscopy data. Background Technology

[0002] Magnetic resonance spectroscopy (MRS) is a non-invasive detection technique that acquires information about the biochemical composition of tissues based on the principle of nuclear magnetic resonance. Unlike conventional magnetic resonance imaging, which primarily displays differences in tissue anatomical morphology and signal intensity, MRS utilizes the differences in chemical shifts produced by atomic nuclei in different chemical environments to acquire resonance signals within local voxels or multiple spatial locations. Proton magnetic resonance spectroscopy (MMR) is commonly used in clinical and research settings. 1 Taking H-MRS as an example, the acquisition device first obtains a time-domain complex signal that decays over time. After frequency correction, phase correction, residual water treatment, and Fourier transform, a frequency domain spectrum is formed. The chemical shift, morphology, amplitude, or area of ​​the spectral peaks in the spectrum can be used to estimate metabolites or complex components such as N-acetylaspartate, choline, creatine, glutamate, and γ-aminobutyric acid. MRS can thus provide quantitative or semi-quantitative information related to tissue metabolic status without tissue sampling and has been used as an auxiliary assessment for various central nervous system diseases (see Wilson M, Andronesi O, Barker PB, et al. Methodological consensus on clinical proton MRS of the brain: Review and recommendations[J].Magnetic Resonance in Medicine, 2019, 82(2): 527-550.).

[0003] MRS data quality is affected by both the acquisition and post-processing processes. During acquisition, improper voxel localization, non-uniform main magnetic field, insufficient shimming, center frequency shift, inadequate water suppression, external volume lipid contamination, abnormal radio frequency coil connection, subject movement, and improper sequence parameter settings can all lead to reduced effective metabolite signals, broadened peaks, and increased residual water or lipid signals. During post-processing, inaccurate frequency or phase correction, improper residual water removal, overfitting or underfitting of the background baseline, and mismatch between the metabolite basis set and the magnetic field strength or the acquired sequence can also result in seemingly good fit but unreliable quantitative results. Therefore, MRS quality management typically requires simultaneous checks of data integrity, peak quality, residual water and baseline, fitting residuals, and quantitative uncertainty, rather than relying solely on a single output value (see Pedrosa de Barros N, Slotboom J. Quality management in in vivo proton MRS[J].Analytical Biochemistry, 2017, 529: 98-116.).

[0004] Current MRS analysis workflows typically involve the operator exporting source data from the MRI machine or picture archiving and communication system (PACS) after the inspection, then transferring it to a separate workstation for offline preprocessing, spectral fitting, and result review. Data quality issues caused by improper voxel localization, shimming, water suppression, or incorrect acquisition parameter settings often only become apparent after the inspected object leaves the equipment. Even if the problem is identified at this point, it's difficult to adjust acquisition conditions or perform supplementary acquisitions during the current inspection, easily leading to invalid data, duplicate appointments, and additional inspection costs. Some PACS systems only store standard DICOM images; MRS time-domain signals, unsuppressed reference data, or vendor-specific source data may be provided by the associated data gateway or device interface. Therefore, relying solely on manual file export is not conducive to establishing a stable data link corresponding to the current inspection task.

[0005] Existing quality assessments suffer from fragmented judgment dimensions. For example, signal-to-noise ratio (SNR) and full width at half maximum (FWHM) can reflect the overall quality of the spectrum, but cannot fully characterize abnormal baselines or local fitting errors. The Cramér-Rao lower bound (CRLB) or relative standard deviation output by fitting software can characterize the parameter uncertainty under a specific fitting model, but the relative CRLB is also affected by factors such as the size of the metabolite estimate, and should not be used as the sole basis for data exclusion. Existing studies have specifically pointed out that quality screening based solely on the relative CRLB threshold may introduce bias (see Kreis R. Thetrouble with quality filtering based on relative Cramér-Rao lower bounds[J].Magnetic Resonance in Medicine, 2016, 75(1): 15-18.). In addition, the quality reference ranges corresponding to different magnetic field strengths, acquisition sequences, echo times, detection nuclei, voxel locations, and target metabolites are not entirely the same. If a fixed threshold is used to process all data, it may reduce the accuracy and interpretability of the judgment.

[0006] Therefore, existing technologies lack a complete technical solution that can communicate with a PACS system or its associated data gateway, directly acquire data for the current inspection task after an MRS acquisition, sequentially complete data-level inspection, preprocessing and fitting, continuous quality index extraction, context matching, and multi-dimensional fusion judgment, and provide feedback to the operator on the causes of anomalies and adjustment suggestions before the end of the same inspection session. Establishing an online closed loop that allows for direct analysis after acquisition, traceable results, and timely supplementary acquisition is a technical problem that needs to be solved in this field. Summary of the Invention

[0007] To address the problems of existing MRS quality analysis, such as reliance on offline processing after inspection, delayed discovery of quality issues, biased judgment based on single indicators, and difficulty in unifying thresholds under different acquisition conditions, this invention provides an online quality analysis and feedback method and system for magnetic resonance spectroscopy data. This invention directly acquires the source data corresponding to the current inspection task after a single MRS acquisition is completed through a PACS system or a data gateway associated with a PACS system; it controls subsequent processing entry points through data-level and analysis-level hard gating; it extracts continuous quality indicators only after both levels of gating have passed, and performs hierarchical fusion judgment based on a set of quality thresholds selected according to the evaluation context; and it outputs quality conclusions, causes of anomalies, and acquisition adjustment suggestions before the end of the same inspection session, enabling operators to correct processing parameters, adjust acquisition conditions, or supplement acquisitions based on feedback.

[0008] The "online quality analysis" referred to in this invention refers to a quality analysis system establishing a communication connection with the PACS system or a data gateway associated with the PACS system via the hospital's data communication network. After an MRS acquisition is completed and the corresponding data becomes accessible, the system directly acquires and analyzes the data corresponding to the current examination task, and outputs the quality judgment result and feedback information before the end of the same examination session corresponding to the task. The online quality analysis does not require synchronization with the MRI signal acquisition process, nor does it require automatic triggering; it can be initiated by data arrival events, acquisition completion status, or operator commands.

[0009] To achieve the above objectives, this invention provides an online quality analysis and feedback method for magnetic resonance spectroscopy data, such as... Figure 1 As shown, it includes the following steps:

[0010] S101. After the magnetic resonance spectroscopy acquisition device completes an MRS data acquisition, it obtains the MRS source data, acquisition parameters, and object identification information corresponding to the current inspection task through the PACS system or a data gateway connected to the PACS system. The data is associated with a single acquisition analysis task based on the inspection identifier, sequence identifier, and acquisition time, and the evaluation context is determined based on the acquisition parameters.

[0011] S102, perform data integrity and resolvability checks on the single acquisition and analysis task; if the check fails, generate a data-level hard failure flag and generate a quality judgment result indicating that the acquisition is unqualified.

[0012] S103, when the data integrity and resolvability check passes, the parsed MRS source data is preprocessed, and the preprocessing includes at least one of frequency correction, phase correction and residual water treatment.

[0013] S104 performs spectral reconstruction and spectral fitting on the preprocessed MRS data; when the preprocessing, spectral reconstruction, or spectral fitting does not produce valid analysis results, an analysis-level hard failure flag is generated, and a quality judgment result indicating that the acquisition is unqualified is generated; continuous quality indicators are extracted from the preprocessed MRS data, reconstructed spectrum, and fitting results only when the data integrity and resolvability checks pass and valid analysis results are generated.

[0014] S105, select a matching quality threshold set based on the evaluation context, unify the direction of quality variation and normalize the thresholds for continuous quality indicators, and perform hierarchical fusion judgment to obtain the quality judgment result of this collection; when the quality judgment result is unqualified, locate the failure type, abnormal indicator combination and possible abnormal cause, and generate processing correction suggestions or collection adjustment suggestions.

[0015] S106 Before the end of the inspection session corresponding to this inspection task, output the quality judgment result, judgment basis, available reconstructed spectrum, fitting result and quantitative result to the operator terminal, and save the processing traceability information; when the quality judgment result is unqualified, send the possible abnormal reasons and acquisition adjustment suggestions to the operator terminal or MRS acquisition device, so that the operator can adjust the acquisition conditions or perform supplementary acquisition within the inspection session.

[0016] The MRS source data includes at least one of time-domain signal data, frequency-domain spectral data, MRS DICOM objects, raw data storage objects, and private DICOM objects containing MRS data, and can be associated with at least one of untreated reference data and calibration reference data. The quality analysis system can obtain the data through at least one of DICOM C-STORE, DICOMweb STOW-RS, shared directory, message queue, application programming interface, or active query method.

[0017] The data integrity and resolvability check includes checking the association between the data objects and the current check task, as well as checking at least one of the following: number of data objects, number of time-domain data points, complex data format, data length, missing values, non-numerical values, infinite values, data saturation, availability of reference data, and consistency of acquisition parameters. If at least one of the following occurs: incorrect association, missing necessary data objects, data corruption, incorrect data format, or inconsistent acquisition parameters, spectrum fitting and continuous quality index extraction will not be performed.

[0018] The continuous quality indicators include spectral quality indicators, residual water and baseline quality indicators, fitting reliability indicators, and quantitative reliability indicators. Spectral quality indicators include at least one of signal-to-noise ratio, full width at half maximum (FWHM), center frequency shift, phase shift, and lipid contamination indicators; residual water and baseline quality indicators include at least one of residual water band power, residual water power percentage, background baseline amplitude, the ratio of background baseline amplitude range to measured spectral amplitude range, and target metabolite band spectral shape indicators; fitting reliability indicators include at least one of the following: coefficient of determination after background subtraction, normalized root mean square error, full-spectrum fitting residual, and target peak region fitting residual; quantitative reliability indicators include at least one of the following: fitting uncertainty characterized by CRLB of the target metabolite and fitting stability indicators.

[0019] The evaluation context includes at least the magnetic field strength, acquisition sequence type, and echo time, and includes at least one of the following: repetition time, detection nucleus type, anatomical region of the voxel, voxel size, and target metabolite. For continuous quality indicators where higher values ​​indicate lower quality, indicator values ​​reaching the acceptable reference threshold are mapped to 0, indicator values ​​reaching the unacceptable reference threshold are mapped to 1, and indicator values ​​between the two thresholds are mapped to 0 to 1. For continuous quality indicators where lower values ​​indicate lower quality, the mapping is performed in reverse, thereby obtaining a dimensionless anomaly score with a consistent direction of quality deterioration.

[0020] The hierarchical fusion judgment includes hard failure judgment, key indicator or abnormal combination judgment, and category score fusion judgment. When a data-level hard failure flag or an analysis-level hard failure flag exists, the current acquisition is directly judged as unqualified. When no hard failure flag exists, the judgment is made based on whether the key indicators reach the severe anomaly threshold and whether the abnormal indicators form a preset combination. When there are no severely abnormal key indicators or abnormal combinations, category scores are generated for spectral quality, residual water and baseline quality, fitting reliability, and quantitative reliability, and the quality judgment result is determined based on the category scores. When CRLB reaches the corresponding threshold, it is preferable to combine at least one other quality indicator anomaly for judgment, rather than using relative CRLB as the sole basis for unqualification.

[0021] This invention also provides an online quality analysis and feedback system for magnetic resonance spectroscopy data, including a PACS / gateway communication interface, a module for checking task association, integrity verification and parsing, an MRS data preprocessing module, a spectrum reconstruction and fitting module, a quality index extraction module, a single acquisition quality judgment module, a failure cause identification module, an online feedback suggestion generation module, and a module for checking session status, display, result storage, and processing traceability. Each module is used to perform the corresponding steps in the above methods.

[0022] The present invention also provides a computer device, including a processor, a memory, a PACS / gateway communication interface, an input / output interface, and a system bus. The memory stores a computer program, which, when executed by the processor, implements the above-described online quality analysis and feedback method for magnetic resonance spectroscopy data.

[0023] The present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the above-described method; the present invention also provides a computer program product comprising a computer program, wherein the computer program, when executed by a processor, implements the above-described method.

[0024] The beneficial effects of this invention are:

[0025] First, this invention establishes a communication connection between the quality analysis system and the PACS system or its associated data gateway, and directly acquires the data corresponding to the current inspection task after a single MRS acquisition is completed, and outputs the results before the end of the same inspection session, which can reduce the feedback delay caused by manual export, transfer and offline analysis after inspection.

[0026] Second, this invention sets data-level hard gating and analysis-level hard gating; when data is missing, damaged, fails to correlate, or has an incorrect format, it will not proceed to fitting; when preprocessing or fitting does not produce effective results, it will not proceed to continuous index extraction, thus avoiding invalid data being mistakenly treated as quantifiable data for further processing.

[0027] Third, this invention combines spectral quality, residual water and baseline quality, fitting reliability and quantitative reliability indicators, and makes combined judgments on abnormal background baselines, apparent goodness of fit and CRLB. Compared with the method of using only signal-to-noise ratio, full width at half maximum or CRLB single threshold, it can provide more complete and interpretable quality basis.

[0028] Fourth, this invention selects a set of quality thresholds based on evaluation contexts such as magnetic field strength, sequence type, echo time, detection nucleus type, voxel location, and target metabolite, and normalizes quality indicators of different directions and dimensions, which is beneficial for compatibility with different devices, sequences, and application scenarios.

[0029] Fifth, this invention maps abnormal index combinations to possible causes such as voxel localization, shimming, center frequency, water suppression, external volume suppression, radio frequency coil, fixed state of the subject under test, and acquisition parameters, and generates processing-side or acquisition-side suggestions, enabling operators to adjust parameters or supplement acquisition before the subject under test leaves, thereby reducing the proportion of invalid data caused by improper operation or acquisition conditions.

[0030] Sixth, this invention saves the input object identifier, algorithm version, metabolite matrix version, processing parameters, threshold set, intermediate results, judgment results, feedback messages and supplementary collection and association records, which can form a traceable, verifiable and auditable MRS quality analysis record. Attached Figure Description

[0031] Figure 1 This is an overall flowchart of the online quality analysis and feedback method for magnetic resonance spectroscopy data of the present invention.

[0032] Figure 2 This is a schematic diagram of the structure and external connections of the online quality analysis and feedback system for magnetic resonance spectroscopy data of the present invention.

[0033] Figure 3 This is a schematic diagram of the process of two-level gating, continuous quality index extraction, hierarchical fusion judgment and dual-path feedback of the present invention.

[0034] Figure 4 This is a schematic diagram of the overall interface for quality analysis of qualified sample A in one embodiment of the present invention.

[0035] Figure 5 This is a schematic diagram of the interface of all quantitative component results for qualified sample A in one embodiment of the present invention.

[0036] Figure 6 This is a schematic diagram of the overall quality analysis interface for non-conforming sample B in one embodiment of the present invention.

[0037] Figure 7 This is a schematic diagram of the interface of all quantitative component results for unqualified sample B in one embodiment of the present invention.

[0038] Figure 8 A schematic diagram illustrating the computer equipment structure for implementing the method of the present invention and its communication relationship with the PACS system, operator terminal, and MRS acquisition device. Detailed Implementation

[0039] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described below with reference to the accompanying drawings and specific embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the scope of protection of this invention.

[0040] In this embodiment of the invention, MRS includes at least one of monomorphic magnetic resonance spectroscopy and multimorphic magnetic resonance spectroscopy imaging; the detected nucleus can be a hydrogen proton, or a phosphorus, carbon, or other nucleus capable of magnetic resonance spectroscopy detection. For ease of explanation, the following mainly uses hydrogen proton monomorphic MRS as an example.

[0041] The "MRS source data" referred to in this invention refers to data objects that can be used to reconstruct the spectrum, perform spectrum fitting, or calculate quality indicators. These can be time-domain free-induction attenuation signals, echo signals, frequency-domain spectra, standard MRS DICOM objects, device raw data objects, or manufacturer-proprietary DICOM objects. MRS source data can simultaneously include water suppression data, unsuppressed water reference data, and calibration reference data, or it can include only the available data from these sources.

[0042] In this invention, an "inspection task" refers to a work unit for this MRS inspection, defined by at least two of the following: inspection identifier, sequence identifier, object identifier, and acquisition time. A "single acquisition analysis task" refers to a quality analysis record corresponding to a specific MRS acquisition within the inspection task. A supplementary acquisition forms a new single acquisition analysis task and is associated with the original single acquisition analysis task.

[0043] The term "same inspection session" in this invention refers to the time range from the start of the current inspection task to the time range during which the operator or inspection system marks the inspection as complete. Within this time range, the MRS acquisition device can continue to acquire data on the currently inspected object. Online feedback only needs to reach the operator's terminal before the inspection session ends; it is not required to be output synchronously during signal acquisition, nor is it required to be completed within a fixed millisecond or second time limit.

[0044] The "quality judgment result" referred to in this invention is used to characterize whether the results of this MRS collection and analysis meet the preset requirements for analyzability or quantification, and does not constitute a diagnostic conclusion on the disease state of the examined subject.

[0045] Example 1: Online Quality Analysis and Feedback Method After Data Acquisition

[0046] This embodiment provides a method for online quality analysis and feedback of magnetic resonance spectroscopy data, such as... Figure 1 As shown, it includes steps S101 to S106.

[0047] S101, acquire the data collected this time and establish a single collection and analysis task.

[0048] like Figure 1 As shown, after the MRS acquisition device completes a data acquisition, the quality analysis system obtains the MRS source data, acquisition parameters, and object identification information for this inspection task through the PACS system or a data gateway associated with the PACS system. The PACS system can directly save standard MRS DICOM objects; when the PACS system does not save complete time-domain signals, unsuppressed water reference data, or manufacturer-specific source data, the associated data gateway receives the corresponding data from the MRS acquisition device and retains inspection and sequence identifiers that are consistent with or can be mapped to the PACS data objects. Therefore, the quality analysis system does not require all source data to be actually stored in the PACS as a necessary condition.

[0049] Data acquisition can be done via push notifications or active querying. Push notifications include receiving data through DICOM C-STORE, DICOMweb STOW-RS, shared directory monitoring, message queues, or application programming interfaces (APIs). Active querying involves retrieving data from the PACS system or data gateway based on check identifiers, sequence identifiers, collection time ranges, or data object instance identifiers. A single data acquisition and analysis task can be initiated by a data arrival event, collection completion status, or operator instructions.

[0050] Object identification information may include at least one of the following: inspected object identifier, inspection number, inspection instance identifier StudyInstanceUID, sequence instance identifier SeriesInstanceUID, data object instance identifier SOPInstanceUID, and acquisition time. The quality analysis system establishes an association between the data object and the current inspection task based on at least two of these information and assigns a single acquisition analysis task identifier to a specific acquisition.

[0051] Acquisition parameters may include at least one of the following: magnetic field strength, acquisition sequence type, echo time, repetition time, detection nucleus type, number of sampling points, sampling interval or spectral width, center frequency, voxel position, voxel size, averaging count, water inhibition state, and external volume inhibition state. The quality analysis system establishes an evaluation context based on the acquisition parameters, used to select the metabolite basis set, data processing parameter set, and quality threshold set.

[0052] S102, perform data integrity and resolvability checks.

[0053] The quality analysis system first checks the correlation between the data objects and the current inspection task to prevent incorrect combination of data from different inspected objects, different sequences, or different collection times. It then checks whether the necessary data objects are complete, whether the number of data objects conforms to the preset structure, whether the number of time-domain data points and the data length are consistent, whether the real and imaginary parts can correctly form a complex signal, and checks for missing values, non-numerical values, infinite values, abnormal constant segments, data truncation, or saturation.

[0054] When this processing requires unsuppressed water reference data or calibration reference data, it also checks whether the corresponding reference data is available and whether the reference data matches the water suppression data in terms of identification, sequence parameters, and acquisition time. If the number of sampling points, spectral width, center frequency, or echo time in the DICOM tag is inconsistent with the actual data length or data header information, it indicates that the acquisition parameters are inconsistent.

[0055] When at least one of the following occurs: object association error, missing necessary data object, data corruption, unparseable format or byte order, missing necessary parameters that cannot be reliably estimated, incorrect data length, or conflicting acquisition parameters, a data-level hard failure flag is generated, directly determining the acquisition as unqualified and outputting data anomaly information. Data with a data-level hard failure is not included in spectrum fitting, nor are continuous quality indicators such as signal-to-noise ratio, full width at half maximum (FWHM), residual water ratio, fitting residual, or CRLB extracted.

[0056] S103, after the data-level check passes, perform MRS data preprocessing.

[0057] Preprocessing includes frequency correction, phase correction, and residual water treatment. For data containing multiple averages or dynamics, frequency and phase alignment can be performed on each dynamic before weighted averaging. For a single time-domain signal, the center frequency shift and zero-order phase shift can be estimated based on the unsuppressed water reference peak, residual water peak, or selected metabolite reference peak, and frequency axis and phase correction can be performed. Preprocessing may also include at least one of the following: abnormal dynamic identification, coil channel merging, windowing, zero-fill, and eddy current correction.

[0058] In a residual water treatment method, signal components are decomposed within a preset data segment of the time-domain signal. Let the time-domain signal be s[n], which can be expressed as the sum of several exponentially decaying complex components and noise:

[0059] s[n] = Σ A_k · exp{(-d_k + j2πf_k)nΔt + jφ_k} + ε[n].

[0060] In the formula, A_k, d_k, f_k, and φ_k represent the amplitude, attenuation parameter, frequency, and phase of the k-th signal component, respectively; Δt represents the sampling interval; and ε[n] represents the unmodeled signal and noise. The quality analysis system determines the residual water center frequency within a preset water peak search interval, selects components located in the neighborhood of the center frequency and satisfying the attenuation constraint from each signal component, reconstructs the residual water signal according to the complete data length, and subtracts the reconstruction result from the original time-domain signal. The component decomposition can be achieved through Hankel singular value decomposition, matrix pencil method, linear prediction, or other methods capable of estimating exponentially decaying components.

[0061] To prevent residual water treatment from damaging adjacent metabolite peaks, the system compares the peak power, target metabolite spectral shape, spectral correlation, and local amplitude changes before and after treatment. If the peak power does not meet the preset reduction condition, or the target metabolite spectral shape distortion exceeds the preset range, the system rejects the residual water treatment result, retains the data before treatment, or reverts to the previous available treatment result, and generates suggestions for adjusting the peak search range, component selection conditions, or treatment order.

[0062] S104 is used for spectrum reconstruction, fitting, and extraction of continuous quality indicators.

[0063] The quality analysis system selects a metabolite basis set that matches the magnetic field strength, acquisition sequence, echo time, and detection nucleus species based on the evaluation context. The metabolite basis set can be obtained from experimental model solutions, numerical simulations, or validated existing basis sets, and the basis set name, version, applicable field strength, sequence, and echo time are recorded.

[0064] The preprocessed time-domain signal was subjected to Fourier transform to obtain the measured frequency-domain spectrum, and spectrum fitting was performed within the target chemical shift range. The measured spectrum y(δ) can be expressed as the sum of the base spectrum of each metabolite or complex component, the background baseline, and the fitting residual:

[0065] y(δ) = Σ c_m · b_m(δ; θ_m) + B(δ) + e(δ).

[0066] In the formula, δ represents the chemical shift, c_m represents the concentration or amplitude estimate of the m-th quantitative component, b_m represents the base spectrum adjusted by the frequency, phase, linewidth, or line shape parameter θ_m, B(δ) represents the background baseline, and e(δ) represents the fitting residual. Spectrum fitting can employ frequency domain linear combination fitting, time domain fitting, nonlinear least squares fitting, Bayesian fitting, or other methods capable of outputting metabolite estimates and fitting errors.

[0067] A valid analysis result must meet at least two of the following conditions: a finite numerical spectrum array corresponding to the preset chemical shift range is generated; the fitting process terminates normally and returns to a valid state; the lengths of the fitted spectrum, background baseline, and fitting residuals are consistent with the measured spectrum; a finite numerical estimate of at least one target metabolite or complex component is output; and resolvable fitting uncertainty or fitting quality information is output. If preprocessing fails, the spectrum array is empty or contains unprocessable non-finite values, the metabolite basis set is mismatched, the fitting fails to converge, the result file structure is incorrect, or necessary fields are not returned, an analysis-level hard failure flag is generated, directly determining the acquisition as unqualified, and continuous quality indicators will no longer be extracted.

[0068] The quality analysis system extracts continuous quality indicators only when data integrity and resolvability checks pass and valid analytical results are produced. For example... Figure 3 As shown, continuous quality indicators are divided into four categories: spectral quality, residual water and baseline quality, fitting reliability, and quantitative reliability.

[0069] Spectral quality metrics include signal-to-noise ratio (S / N), full width at half maximum (FWHM), center frequency shift, phase shift, and lipid contamination metrics. S / N can be defined as the ratio of the representative signal amplitude of the target peak or fitted spectrum to the standard deviation of the metabolite-free noise region, or it can use the S / N definition output by validated fitting software. FWHM can be expressed in Hz or ppm using the water peak, reference metabolite peak, or the common linewidth obtained from the fitting. The calculation definitions used are stored together with the algorithm version, and the threshold set matches this definition.

[0070] Residual water and baseline quality indicators include water peak power, residual water power percentage, background baseline amplitude, the ratio of background baseline amplitude range to measured spectral amplitude range, and the target metabolite band spectral shape. The residual water power percentage can be defined as the ratio of water peak band power to the total analysis band power, reference band power, or target metabolite band power. The ratio of background baseline amplitude range to measured spectral amplitude range can be expressed as:

[0071] R_B = [max(B)-min(B)] / [max(y)-min(y)].

[0072] The reliability metrics for the fit include the coefficient of determination (R²) after background subtraction, the normalized root mean square error (NRMSE), the full-spectrum fitting residual, and the target peak region fitting residual. Let the measured spectrum after background subtraction be y′=yB, and the fitted spectrum after background subtraction be ŷ′. Then the coefficient of determination and the normalized root mean square error after background subtraction can be calculated using the following formulas:

[0073] R² = 1 - Σ(y′-ŷ′)² / Σ[y′-mean(y′)]²;

[0074] NRMSE = sqrt{mean[(y′-ŷ′)²]} / [max(y′)-min(y′)].

[0075] By calculating R², NRMSE, and residuals after background baseline subtraction, and combining these with the background baseline amplitude range ratio R_B, we can identify the apparent goodness of fit increase that occurs when the background baseline dominates. For example, the original fitted curve may follow the slow trend of the measured data due to the high degree of freedom of the baseline, but the decrease in R², increase in NRMSE, and increase in R_B after background subtraction indicate that the true metabolite peaks are not reliably fitted.

[0076] Quantitative reliability metrics include the fitting uncertainty and fitting stability characterized by the CRLB of the target metabolite. The %SD relative to the CRLB or fitting software output describes the uncertainty of the estimate under given model and noise conditions; a higher value indicates that the corresponding component cannot be stably estimated, but does not mean that the true concentration of that component is zero. For components with an output of 0 and a relative CRLB value of 999% or other upper limit codes, the system interprets this as the fitting software failing to reliably estimate the value, rather than a definitive biological zero.

[0077] S105 performs context threshold matching, hierarchical fusion determination, and anomaly cause localization.

[0078] The evaluation context includes at least the magnetic field strength, acquisition sequence type, and echo time, and may further include repetition time, detection nucleus type, anatomical region of the voxel, voxel size, average number of repetitions, target metabolite, preprocessing algorithm version, fitting software version, and metabolite basis set version. The quality threshold set can be established from equipment acceptance data, model solution data, manually verified reference samples, or historical qualified data within the same evaluation context.

[0079] In one approach to establishing a threshold set, for indicators where higher values ​​indicate lower quality, a preset high quantile of the distribution of acceptable samples is used as the acceptable reference threshold T_g, and an expert-confirmed distribution of unacceptable samples, a higher quantile, or a preset upper limit is used as the unacceptable reference threshold T_b. For indicators where lower values ​​indicate lower quality, T_g and T_b are determined in reverse. Alternatively, the thresholds can be determined based on the relationship between quality labels and indicators on the validation set, using receiver operating characteristic (ROC) curves, sensitivity and specificity constraints, or cost functions.

[0080] For continuous quality indices where higher values ​​indicate worse quality and T_b is greater than T_g, the dimensionless outlier score q is obtained according to the following formula:

[0081] q = clip[(x-T_g) / (T_b-T_g), 0, 1].

[0082] For continuous quality indices where lower values ​​indicate worse quality and T_g is greater than T_b, the dimensionless outlier score q is obtained according to the following formula:

[0083] q = clip[(T_g-x) / (T_g-T_b), 0, 1].

[0084] In the formula, x represents the measured value of the indicator, and clip indicates that values ​​less than 0 are truncated to 0, and values ​​greater than 1 are truncated to 1. Thus, each indicator is converted to a score in the same direction where 0 represents no abnormality and 1 represents reaching the unqualified reference level. For indicators without a naturally monotonic direction, their deviation from the target range or target value can be calculated first, and then normalized as described above.

[0085] The stratified fusion judgment first performs a hard failure judgment: if a data-level hard failure flag or an analysis-level hard failure flag is present, the data is directly deemed unqualified. If no hard failure flag is present, the critical indicator gating is applied. Critical indicators can be selected from at least two of the following: S / N, FWHM, residual water power ratio, R_B, NRMSE, and CRLB of the target metabolite. If any critical indicator other than CRLB reaches a severe anomaly threshold, the data can be directly deemed unqualified. If CRLB reaches a severe anomaly threshold, it is preferable that at least one other critical indicator is also abnormal to avoid data rejection based solely on a relative CRLB value.

[0086] When no critically abnormal indicators or pre-defined abnormal combinations exist, category scores are calculated for each of the four quality indicators. The category score C_c for the c-th indicator can be the maximum value of all abnormal scores in that category, or it can be calculated as C_c = Σw_iq_i / Σw_i, where w_i is a pre-defined weight. The system can determine non-compliance when any category score reaches the category threshold, or it can take the maximum value or a weighted average of the category scores again to form an overall score and compare it with the overall threshold. The definitions, thresholds, weights, and judgment rules of each indicator are saved as a single versioned threshold set.

[0087] Anomaly localization is based on anomaly indicators and their combinations. An abnormally high FWHM indicates an abnormal shimming quality as a possible cause; an abnormal center frequency shift indicates a main magnetic field drift or abnormal center frequency calibration as possible causes; an abnormal residual water band power or residual water power ratio indicates an abnormal water suppression calibration or abnormal residual water treatment as possible causes; lipid contamination or abnormal background baseline amplitude indicates an abnormal voxel localization or external volume suppression as possible causes; a decreased S / N ratio accompanied by an increased CRLB for the target metabolite indicates insufficient effective metabolite signal as a possible cause; a decreased R² after background subtraction, an increased NRMSE accompanied by an increased R_B, indicates unreliable fitting dominated by an abnormal background baseline as a possible cause.

[0088] S106 outputs results, provides online feedback, and generates a traceability of the processing.

[0089] When the quality assessment result is satisfactory, the system outputs the available reconstructed spectrum, fitted spectrum, fitted residuals, continuous quality indicators, quantitative results, and a quality report to the operator's terminal, and marks the single acquisition and analysis task as successful. Quantitative results may include estimated concentrations of metabolites or complex components, relative ratios to reference metabolites, concentration ratios between target metabolites, and fitting uncertainty. The units of the concentration estimates depend on whether unsuppressed water references, tissue composition corrections, and relaxation corrections are used, and can be absolute concentration units, mechanical units, or relative units defined by the fitting software; the system simultaneously displays or records the corresponding quantitative method and units.

[0090] When the quality assessment result is unqualified, the system generates processing correction suggestions or acquisition adjustment suggestions based on possible causes of the anomaly. Processing correction suggestions include adjusting at least one of the following: residual water content search interval or selection conditions, frequency correction parameters, phase correction parameters, baseline processing parameters, and fitting parameters; and re-performing the corresponding processing on the same source data. Acquisition adjustment suggestions include checking or adjusting at least one of the following: voxel localization, shimming, center frequency, water suppression, external volume suppression, RF coil connection, subject fixation status, and acquisition sequence parameters.

[0091] Before the current inspection session ends, the system sends a feedback message to the operator's terminal containing abnormal indicators, possible causes of the abnormality, and suggestions for data collection adjustments. When the MRS data collection device provides a permitted message or interface, it can also send parameter adjustment or supplementary data collection suggestions to the MRS data collection device. The system only sends data collection-side feedback when the quality judgment result is unqualified and a data collection adjustment suggestion is generated; it does not directly bypass the operator's control of the data collection device.

[0092] After the operator performs supplementary data collection, the system establishes the new data as a new single data collection and analysis task, and saves the association between the new task and the original task. The association information includes at least one of the following: original task identifier, new task identifier, supplementary data collection time, feedback reason, operator confirmation status, and before-and-after quality results, thereby enabling comparison of quality changes before and after the adjustment.

[0093] The processing of source tracing information includes at least five of the following: input data object identifier, algorithm version, metabolite matrix version, data processing parameter set, quality threshold set, intermediate processing results, quality judgment results, feedback messages, reanalysis records, and supplementary collection records. This source tracing information can be stored in a database, file system, PACS-linked structured reports, or hospital information system.

[0094] Example 2: Two-level gating and dual-path feedback

[0095] This embodiment combines Figure 3 Further explanation of the quality analysis process. Figure 3 The main process, from top to bottom, consists of the following steps: input of the collected data and evaluation context, data integrity and resolvability check, preprocessing and fitting, extraction of continuous quality indicators, threshold normalization, hierarchical fusion judgment, and result output.

[0096] The first level is data-level gating. When the data is complete and parseable, the process proceeds to preprocessing, spectrum reconstruction, and fitting; when the data is incomplete or unparseable, a data-level hard failure is generated, directly resulting in an unacceptable outcome, without further preprocessing, fitting, or continuous quality index calculation.

[0097] The second level is analytical-level gating. When preprocessing, spectrum reconstruction, and fitting are successful and produce valid analytical results, the process proceeds to continuous quality index extraction; if no valid analytical results are produced, an analytical-level hard failure is generated, directly entering the unacceptable result stage, and continuous quality indices are no longer calculated. Therefore, Figure 3 The “continuous quality index extraction” has a single entry point, which is passed by both levels of gating.

[0098] After both levels of gating are passed, the system extracts spectral quality, residual water and baseline quality, fitting reliability, and quantitative reliability indicators. Then, based on the evaluation context, it selects a threshold set and performs hierarchical fusion judgment. If the judgment is qualified, it outputs quantitative results, a quality report, and processing traceability information; if the judgment is unqualified, it first locates the failure type, abnormal combination, and possible causes, and then selects a feedback path based on whether the problem is located on the processing side or the acquisition side.

[0099] The processing-side path is used to handle situations where parameters can be corrected but the source data still has analytical value, such as an inappropriate residual water treatment search range, frequency or phase correction deviations, or mismatched baseline or fitted parameters. After adjusting the relevant parameters, the system re-analyzes the same source data, and the re-analyzed records are correlated with the original records.

[0100] The acquisition-side path is used for situations where data quality cannot be recovered through reprocessing alone, such as significant voxel deviation from the target area, insufficient shimming, failed water suppression calibration, external volume lipid contamination, abnormal RF coil connection, significant subject movement, or insufficient acquisition attempts. The system provides feedback to the operator on inspection items and suggestions for supplementary acquisitions within the same inspection session.

[0101] Example 3: Quality Result Display and Case Verification

[0102] This embodiment combines Figures 4 to 7 Explain the display method of the quality analysis results and the results of the stratified fusion judgment. Figure 4 and Figure 5 Sample A is a qualified sample. Figure 6 and Figure 7 This is sample B, which is not qualified. The values ​​in the figure are used to illustrate the indicators and results that the system can display and fuse, and do not constitute fixed thresholds applicable to all magnetic field strengths, sequences, and voxel locations.

[0103] like Figure 4 As shown, the MRS source data for qualified sample A was obtained via PACS, the data integrity check passed, and the analysis status is "completed." The quality analysis overview interface simultaneously displays the measured spectrum, the fitted spectrum, and the fitted residuals, and shows that S / N is 9, FWHM is 0.049 ppm, GABA CRLB is 7%, Glu CRLB is 6%, R² after background subtraction is 0.947, NRMSE is 4.17%, and the target metabolite ratio GABA / Glu is 0.501.

[0104] Within the threshold set matching the evaluation context of this sample, S / N, FWHM, R² after background subtraction, and NRMSE did not reach the severe anomaly threshold. CRLB for GABA and Glu was within acceptable range, and the peak structure showed good consistency with both the measured and fitted spectra. Therefore, the multidimensional fusion result was deemed acceptable. The system output a suggestion: "No need to re-acquire; save the quantitative results and generate an analysis report."

[0105] like Figure 5 As shown, the operator can access the full quantitative component results interface from the overall quality overview interface. This interface displays the name of each quantitative component, its estimated concentration, the %SD output by the fitting software, and the relative Cr+PCr ratio. For components with a %SD of 999% or higher, the interface indicates that the quantification of that component is unreliable and clarifies that this output does not represent a true concentration of zero; a high %SD for a single non-target component does not change the overall quality conclusion formed by the preset multidimensional rules.

[0106] like Figure 6 As shown, the data integrity check of the non-compliant sample B also passed, and the spectral fitting generated a formally displayable result. Therefore, this sample did not fall under the category of hard failure at the data or analysis level, but instead entered the continuous quality index and stratified fusion judgment. Its S / N was 2, FWHM was 0.044 ppm, GABA CRLB was 31%, Glu CRLB was 61%, R² after background subtraction was 0.267, NRMSE was 12.85%, and the target metabolite ratio GABA / Glu was 0.708.

[0107] Although the FWHM of the substandard sample B did not show a significant broadening, and the apparent fit curve could follow the slow trend of the measured data, the S / N ratio was significantly reduced, the R² after background subtraction was reduced, and the NRMSE was increased, accompanied by an increase in the CRLB of GABA and Glu. This abnormal combination suggests that the background baseline is dominant and the true target spectral peak information is insufficient, making the target metabolite ratio unreliable. Based on this, the system determined that the acquisition was substandard and generated suggestions for supplementary acquisition after checking voxel localization, shimming, and water suppression settings.

[0108] like Figure 7 As shown, even with a non-compliant sample B, all quantitative component results can still be displayed, allowing the operator to review the raw output of the fitting software. However, the interface retains a prominent non-compliant status and indicates that the results cannot be directly used for subsequent quantitative statistics. This design distinguishes between "whether the results can be generated" and "whether the results meet the quality requirements," avoiding the misconception that data is acceptable simply because a quantitative table exists.

[0109] The two examples above illustrate that the present invention can distinguish between three states: unparseable data, failed analysis process, and completed analysis but substandard quality; it can avoid making one-sided judgments based on single indicators such as FWHM or CRLB; and it can simultaneously provide the operator with the overall quality conclusion, judgment basis, and recommended actions.

[0110] Example 4: Online Quality Analysis and Feedback System

[0111] This embodiment provides an online magnetic resonance spectroscopy data quality analysis and feedback system 200, such as... Figure 2 As shown. System 200 is communicatively connected to PACS system or associated data gateway 100, operator terminal 300 and MRS acquisition device 400.

[0112] The PACS / gateway communication interface 210 is used to obtain the MRS source data, acquisition parameters, and object identification information corresponding to the current inspection task from the PACS system or the associated data gateway 100 after the MRS acquisition device 400 completes a data acquisition.

[0113] The task association, integrity verification and parsing module 220 is used to associate the acquired data with a single acquisition and analysis task, determine the evaluation context, perform data integrity and parsing checks, and generate a data-level hard failure flag when the check fails.

[0114] The MRS data preprocessing module 230 is used to perform frequency correction, phase correction, residual water treatment, and other selected preprocessing on the parsed MRS source data after the data integrity and resolvability checks have passed.

[0115] The spectrum reconstruction and fitting module 240 is used to perform spectrum reconstruction and spectrum fitting on the preprocessed MRS data, generate the measured spectrum, fitted spectrum, background baseline, fitting residual and quantitative results, and generate an analysis-level hard failure flag when no valid analysis results are generated.

[0116] The quality indicator extraction module 250 is used to extract spectral quality, residual water and baseline quality, fit reliability and quantitative reliability indicators only when the data integrity and resolvability checks pass and valid analysis results are produced.

[0117] The single-acquisition quality judgment module 260 is used to directly generate unqualified results based on two types of hard failure indicators, or, when there is no hard failure, select a quality threshold set based on the evaluation context to perform directional unification, threshold normalization and hierarchical fusion judgment on continuous quality indicators.

[0118] The failure cause identification module 270 is used to locate the failure type, abnormal indicator combination, and possible abnormal cause when the quality judgment result is unqualified. The online feedback suggestion generation module 280 is used to generate processing correction suggestions or collection adjustment suggestions based on possible abnormal causes, and send the corresponding feedback before the end of the inspection session corresponding to this inspection task.

[0119] The session status check, display, result storage and processing traceability module 290 is used to record whether the check task is still in a state where it can continue to collect data. It displays the quality judgment results and judgment basis on the operator terminal 300, and saves the available reconstructed spectrum, fitting results, quantitative results, feedback information and processing traceability information.

[0120] The operator terminal 300 can be an MRS acquisition console, a medical imaging workstation, an in-hospital computer terminal, or an authorized mobile terminal. The MRS acquisition device 400 is used to perform parameter adjustments or supplementary acquisitions after operator confirmation. Figure 2 Solid lines in the diagram represent data or result transmission, while dashed lines represent feedback sent only when the quality assessment fails.

[0121] Example 5: Computer devices, storage media, and computer program products

[0122] This embodiment provides a computer device 500, such as... Figure 8 As shown, the system includes a processor 510, a memory 520, a PACS / gateway communication interface 530, an input / output interface 540, and a system bus 560. The processor 510, memory 520, PACS / gateway communication interface 530, and input / output interface 540 are connected via the system bus 560.

[0123] The PACS / gateway communication interface 530 is used to receive and associate the data acquired in this session from the PACS system or the associated data gateway 100. The input / output interface 540 is used to output quality results and feedback to the operator terminal 300 within the same inspection session, and can output parameter adjustment or supplementary acquisition suggestions to the MRS acquisition device 400 when the quality results are unsatisfactory.

[0124] Memory 520 is used to store runtime data, check session status, data processing parameters, metabolite basis set, quality threshold set, and process traceability information. Computer-readable storage medium 570 stores program instructions 571, which, when executed by processor 510, implement... Figure 1 The method shown.

[0125] The processor 510 may be a central processing unit, a graphics processing unit, a digital signal processor, or other processing unit; the memory 520 and the computer-readable storage medium 570 may be random access memory, read-only memory, solid-state memory, disk, optical disk, or other non-transitory storage medium.

[0126] The computer device 500 can be deployed on an in-hospital server, a medical imaging workstation, or multiple computing nodes that communicate via a controlled network; each module can be implemented by software, hardware, or a combination of both, and functions can be combined or split.

[0127] The present invention also provides a computer-readable storage medium storing a computer program, and a computer program product including the computer program; the computer program, when executed by a processor, implements the aforementioned method.

[0128] Although specific embodiments, processing flows, quality indicators, threshold settings, experimental results, and accompanying drawings of the present invention have been disclosed for illustrative purposes to aid in understanding the technical content of the invention and its implementation, the present invention is not limited to the above-described embodiments. Those skilled in the art will understand that substitutions, adjustments, or combinations of data access methods, spectral data formats, preprocessing and fitting methods, quality indicators and their thresholds, hierarchical fusion rules, feedback methods, communication interfaces, and functional module combinations, without departing from the spirit and scope defined by the present invention and its claims, should all fall within the protection scope of the present invention.

Claims

1. A method for online quality analysis and feedback of magnetic resonance spectroscopy data, characterized in that, The method is applied to a quality analysis system, which is communicatively connected to a PACS system or a data gateway associated with the PACS system, and also communicatively connected to an operator terminal; the method includes the following steps: 1) After the magnetic resonance spectroscopy acquisition device completes a magnetic resonance spectroscopy data acquisition, the magnetic resonance spectroscopy source data, acquisition parameters and object identification information corresponding to this inspection task are obtained through the PACS system or the data gateway; the acquired data are associated with a single acquisition analysis task based on the inspection identifier, sequence identifier and acquisition time, and the evaluation context of the single acquisition analysis task is determined based on the acquisition parameters. 2) Perform data integrity and resolvability checks on the single data acquisition and analysis task; if the check fails, generate a data-level hard failure flag and generate a quality judgment result indicating that the data acquisition is unqualified; 3) When the data integrity and resolvability checks pass, the resolved magnetic resonance spectral source data is preprocessed, reconstructed, and fitted. When the preprocessing, reconstruction, or fitting does not produce valid analysis results, an analysis-level hard failure flag is generated, and a quality judgment result indicating that the acquisition is unqualified is generated. 4) Only when the data integrity and resolvability checks pass and the valid analysis results are generated, continuous quality indicators are extracted from the preprocessed magnetic resonance spectral data, reconstructed spectrum, and fitting results; a matching quality threshold set is selected according to the evaluation context, the continuous quality indicators are unified in terms of quality variation direction and threshold normalized, and a hierarchical fusion judgment is performed to obtain the quality judgment result of this collection. 5) When the quality judgment result is qualified, generate quantitative results and quality report; when the quality judgment result is unqualified, locate the failure type, abnormal indicator combination and possible abnormal cause, and generate processing correction suggestions or collection adjustment suggestions. 6) Before the end of the inspection session corresponding to this inspection task, output the quality judgment result, judgment basis, available reconstructed spectrum, fitting result and quantitative result to the operator terminal, and save the processing traceability information; when the quality judgment result is unqualified, also send the possible abnormal reasons and the acquisition adjustment suggestions to the operator terminal or the magnetic resonance spectroscopy acquisition device, so that the operator can adjust the acquisition conditions or perform supplementary acquisition within the inspection session.

2. The method as described in claim 1, characterized in that, In step 1): The magnetic resonance spectral source data includes at least one of time-domain signal data, frequency-domain spectral data, magnetic resonance spectral DICOM object, raw data storage object, and private DICOM object containing magnetic resonance spectral data. The magnetic resonance spectral source data is also associated with at least one of unsuppressed water reference data and calibration reference data; The quality analysis system receives the magnetic resonance spectral source data through data push or active query acquisition, and initiates the single acquisition and analysis task based on any one of the following: data arrival event, acquisition completion status, or operator instruction.

3. The method as described in claim 1, characterized in that, In step 2), the data integrity and resolvability check includes: Based on at least two of the object identifier, inspection instance identifier, sequence instance identifier, and data object instance identifier, check the association between each data object and this inspection task; Check at least one of the following: number of data objects, number of time-domain data points, complex data format, data length, missing values, non-numerical values, infinite values, data saturation status, availability of reference data, and consistency of acquisition parameters. When at least one of the following occurs: incorrect association, missing necessary data objects, corrupted data, incorrect data format, or inconsistent acquisition parameters, the data integrity and resolvability check is deemed to have failed. Without performing spectrum fitting and continuous quality index extraction, the data-level hard failure flag and corresponding data anomaly information are output.

4. The method as described in claim 1, characterized in that, In step 3), the preprocessing includes frequency correction, phase correction, and residual water treatment; when the magnetic resonance spectral source data includes time-domain signal data, the residual water treatment includes: Within a preset data segment of the time-domain signal data, signal component decomposition is performed to obtain the frequency, attenuation parameter, amplitude, and phase of each signal component; Within a preset water peak search interval, the residual water center frequency is determined, and residual water components that are located in a preset neighborhood of the residual water center frequency and satisfy the attenuation constraint are selected from the signal components. The residual water content is reconstructed according to the complete data length of the time-domain signal data, and the reconstructed residual water content is subtracted from the time-domain signal data; Compare the peak power of water and the spectrum of target metabolites before and after residual water treatment; if the peak power of water does not meet the preset reduction condition or the spectrum of target metabolites is distorted beyond the preset range, reject the results of this residual water treatment and generate corresponding treatment correction suggestions.

5. The method as described in claim 1, characterized in that, In step 3), the spectrum reconstruction and spectrum fitting include: Based on the evaluation context, select a metabolite basis set that matches the magnetic field strength, acquisition sequence, echo time, and detection nuclei. The preprocessed magnetic resonance spectral data is subjected to frequency domain transformation to obtain the measured spectrum within the target chemical shift range. The measured spectrum is then decomposed into metabolite signal, background baseline and fitting residual to obtain the fitted spectrum. Based on the fitting results, at least one of the following is obtained: concentration estimate of the quantitative component, fitting uncertainty, relative reference metabolite ratio, and concentration ratio between target metabolites; When the preprocessing, spectrum reconstruction, or spectrum fitting fails to produce results that meet the preset data structure requirements, it is determined that no valid analysis results have been generated, and without performing continuous quality index extraction, the analysis-level hard failure flag and the corresponding analysis anomaly information are output.

6. The method as described in claim 1, characterized in that, In step 4): The continuous quality indicators include spectral quality indicators, residual water and baseline quality indicators, fitting reliability indicators, and quantitative reliability indicators. The spectral quality metrics include at least one of the following: signal-to-noise ratio, full width at half maximum (FWHM), center frequency offset, phase offset, and lipid contamination metrics. The residual water and baseline quality indicators include at least one of the following: residual water bandwidth power, residual water power ratio, background baseline amplitude, ratio of background baseline amplitude range to measured spectrum amplitude range, and target metabolite bandwidth spectral shape indicator. The fitting reliability index includes at least one of the following: the determination coefficient after background subtraction, the normalized root mean square error, the full spectrum fitting residual, and the target peak region fitting residual. The quantitative reliability index includes at least one of the fitting uncertainty and fitting stability index characterized by the lower bound of the target metabolite, Cramer-Robb.

7. The method as described in claim 6, characterized in that, The acquisition of the fitting reliability index includes: Remove the background baseline from the measured spectrum to obtain the background-subtracted measured spectrum, and remove the corresponding background baseline component from the fitted spectrum to obtain the background-subtracted fitted spectrum; Based on the measured spectrum after background subtraction and the fitted spectrum after background subtraction, calculate the determination coefficient after background subtraction, the normalized root mean square error, and the fitting residual; Calculate the ratio of the amplitude range of the background baseline to the amplitude range of the measured spectrum; The determination coefficient after background subtraction, the normalized root mean square error, and the ratio of the fitting residual to the amplitude range are used in combination to identify an increase in apparent goodness of fit caused by an anomalous background baseline.

8. The method as described in claim 1, characterized in that, In step 4): The evaluation context includes at least the magnetic field strength, the type of acquisition sequence, and the echo time, and includes at least one of the following: repetition time, the type of nucleus detected, the anatomical region where the voxel is located, the voxel size, and the target metabolite. For continuous quality indicators where higher values ​​indicate lower quality, the indicator value that reaches the qualified reference threshold is mapped to 0, the indicator value that reaches the unqualified reference threshold is mapped to 1, and the indicator value between the qualified reference threshold and the unqualified reference threshold is linearly mapped to 0 to 1. For continuous quality indicators where lower values ​​indicate worse quality, the mapping is performed in reverse; the indicator values ​​outside the mapping range are truncated to 0 or 1 to obtain dimensionless anomaly scores that are consistent with the direction of quality deterioration.

9. The method as described in claim 1, characterized in that, In step 4), the layered fusion determination includes: When the data-level hard failure flag or the analysis-level hard failure flag is present, the current data collection is directly judged as unqualified. When there is no hard failure indicator, at least two of the following are selected as key indicators: signal-to-noise ratio, full width at half maximum (FWHM), residual water power ratio, ratio of background baseline amplitude range to measured spectrum amplitude range, normalized root mean square error, and lower bound of the target metabolite (cramere-roh). If any key indicator other than the lower bound of the cramere-roh reaches the corresponding severe anomaly threshold, or if the lower bound of the cramere-roh reaches the corresponding threshold and at least one other key indicator is abnormal at the same time, the acquisition is judged as unqualified. When there are no key indicators or abnormal indicator combinations that reach the severe abnormality threshold, the corresponding category score is formed by using the maximum value or weighted average value of the dimensionless abnormal scores of the spectral quality index, the residual water and baseline quality index, the fitting reliability index and the quantitative reliability index, respectively. The quality judgment result of this collection is determined based on the comparison result of the category score and the preset category threshold.

10. The method as described in claim 1, characterized in that, In step 5), locating the possible cause of the anomaly includes at least one of the following methods: When the full width at half maximum (FWHM) exceeds the corresponding threshold, the abnormal shimming quality is identified as a possible cause of the abnormality. When the center frequency deviation exceeds the corresponding threshold, the main magnetic field drift or center frequency calibration abnormality will be identified as a possible cause of the abnormality. When the residual water frequency band power or residual water power ratio exceeds the corresponding threshold, water suppression calibration abnormality or residual water treatment abnormality will be identified as possible abnormal causes. When the lipid pollution index or background baseline amplitude exceeds the corresponding threshold, voxel localization abnormality or external volume inhibition abnormality is identified as a possible abnormal cause. When the signal-to-noise ratio is lower than the corresponding threshold and the lower bound of the target metabolite's clammet-roline ratio is higher than the corresponding threshold, insufficient effective metabolite signal is identified as a possible cause of abnormality. When the determination coefficient after background subtraction is lower than the corresponding threshold, the normalized root mean square error is higher than the corresponding threshold, and the ratio of the amplitude range of the background baseline to the amplitude range of the measured spectrum is higher than the corresponding threshold, the unreliable fitting dominated by the abnormal background baseline is identified as a possible abnormal cause.

11. The method as described in claim 1, characterized in that, In steps 5) and 6): The processing correction suggestions include adjusting at least one of the following: the search range or selection conditions for residual moisture, frequency correction parameters, phase correction parameters, baseline processing parameters, and fitting parameters; and re-performing the corresponding processing on the same magnetic resonance spectral source data. The acquisition adjustment suggestions include checking or adjusting at least one of the following: voxel localization, shimming, center frequency, water suppression, external volume suppression, radio frequency coil connection, subject fixation status, and acquisition sequence parameters. Before the inspection session ends, the quality analysis system will send a feedback message containing abnormal indicators, possible causes of abnormality, and the acquisition adjustment suggestions to the operator terminal or the magnetic resonance spectroscopy acquisition device. When new magnetic resonance spectral source data is received from supplementary acquisitions, a new single acquisition and analysis task is established for the new magnetic resonance spectral source data, and the correlation between the new single acquisition and analysis task and the original single acquisition and analysis task is saved.

12. The method as described in claim 1, characterized in that, In step 6): The quality analysis overview interface displays the measured spectrum, the fitted spectrum, the fitted residual, various quality indicators, qualified or unqualified status, possible abnormal causes and suggested actions, and provides an operation entry point to access the full quantitative component results interface. The interface for all quantitative component results displays the name, estimated concentration, fitting uncertainty, and relative reference metabolite ratio of each quantitative component, as well as the concentration ratio between target metabolites. Save at least five of the following: input data object identifier, algorithm version, metabolite matrix version, data processing parameter set, quality threshold set, intermediate processing results, quality judgment results, feedback messages, reanalysis records, and supplementary collection records, to form the processing traceability information.

13. An online quality analysis and feedback system for magnetic resonance spectroscopy data, characterized in that, include: The data communication interface is used to obtain magnetic resonance spectral source data, acquisition parameters and object identification information corresponding to the current inspection task from the PACS system or the data gateway associated with the PACS system after the magnetic resonance spectral acquisition device completes a magnetic resonance spectral data acquisition. The task association, integrity verification and parsing module is used to associate the acquired data with a single collection and analysis task, determine the evaluation context, perform data integrity and parsability checks, and generate a data-level hard failure flag when the checks fail. The magnetic resonance spectroscopy data preprocessing module is used to preprocess the resolved magnetic resonance spectroscopy source data after the data integrity and resolvability checks have passed. The spectrum reconstruction and fitting module is used to perform spectrum reconstruction and spectrum fitting on the preprocessed magnetic resonance spectrum data, and generate an analysis-level hard failure flag when no valid analysis results are produced. The quality indicator extraction module is used to extract continuous quality indicators only when the data integrity and resolvability checks pass and the valid analysis results are generated. The single acquisition quality judgment module is used to generate a quality judgment result indicating that the acquisition is unqualified when the data-level hard failure flag or the analysis-level hard failure flag is present. And it is used to select a quality threshold set according to the evaluation context when there is no hard failure flag, to unify the direction of quality variation and normalize the threshold of the continuous quality index, and to obtain the quality judgment result of this collection through hierarchical fusion judgment. The failure cause identification module is used to locate the failure type, abnormal indicator combination, and possible abnormal cause when the quality judgment result is unqualified. The online feedback suggestion generation module is used to generate processing correction suggestions or acquisition adjustment suggestions based on the possible abnormal reasons, and send the acquisition adjustment suggestions to the operator terminal or the magnetic resonance spectroscopy acquisition device before the end of the inspection session corresponding to this inspection task. The module for checking session status, displaying, storing results, and tracing the source of processing is used to record the status of the checking session, display the quality judgment results and judgment basis before the end of the checking session, and save the available reconstructed spectrum, fitting results, quantitative results, feedback information, and processing traceability information.

14. The system as described in claim 13, characterized in that: The data communication interface is configured to receive data objects pushed by the PACS system or the data gateway via C-STORE or STOW-RS, or to obtain the data objects through active querying. The system also includes an input / output interface, which is used to communicate with at least one of the magnetic resonance spectroscopy acquisition device and the operator terminal. The input / output interface outputs parameter adjustment or supplementary acquisition suggestions to the magnetic resonance spectroscopy acquisition device only when the quality judgment result is unqualified and the acquisition adjustment suggestion is generated. The input / output interface is also used to output the quality judgment result, the quantitative result, and the parameter adjustment or supplementary collection suggestions to the operator terminal, and to receive re-analysis instructions, feedback confirmation information, or supplementary collection confirmation information.

15. A computer device, characterized in that, It includes a processor, a memory, a data communication interface, an input / output interface, and a system bus, wherein the processor, the memory, the data communication interface, and the input / output interface are connected through the system bus; The memory stores a computer program, which, when executed by the processor, implements the online quality analysis and feedback method for magnetic resonance spectral data as described in any one of claims 1 to 12.

16. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the online quality analysis and feedback method for magnetic resonance spectral data as described in any one of claims 1 to 12.

17. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the online quality analysis and feedback method for magnetic resonance spectral data as described in any one of claims 1 to 12.

Images