CN-122019981-A - Sparse data reconstruction method and related device

CN122019981ACN 122019981 ACN122019981 ACN 122019981ACN-122019981-A

Abstract

The invention discloses a sparse data reconstruction method and a related device, which relate to the field of data processing, and are characterized in that an intracranial pressure sparse data stream acquired at a sampling rate within a preset low sampling rate range in a sliding time window is acquired, an intracranial pressure average value is calculated, the intracranial pressure average value is used as an initial value, an intermediate intracranial pressure data sequence obtained by preprocessing the intracranial pressure sparse data stream is subjected to dynamic characteristic analysis, a step mark corresponding to a pathological step is added to obtain a target intracranial pressure data sequence, a segmented fitting result is obtained by segment fitting the target intracranial pressure data sequence, a multicomponent mathematical model set is obtained by signal reconstruction, the sampling rate within the preset high sampling rate range is set, and a reconstructed intracranial pressure change waveform is generated. According to the invention, through preprocessing the intracranial pressure sparse data stream, and then carrying out step detection and model reconstruction, the intracranial pressure signal and the variation trend waveform thereof which are equivalent to those obtained by high sampling rate acquisition are reconstructed from the intracranial pressure sparse data stream with low sampling rate.

Inventors

Request for anonymity
Request for anonymity
Request for anonymity
Request for anonymity

Assignees

中国科学院微电子研究所

Dates

Publication Date: 20260512
Application Date: 20260203

Claims (10)

1. A sparse data reconstruction method, comprising: Acquiring an intracranial pressure sparse data stream acquired at a sampling rate within a preset low sampling rate range in a sliding time window, and calculating an intracranial pressure average value of the intracranial pressure sparse data stream; preprocessing the intracranial pressure sparse data stream to obtain a continuous and effective intermediate intracranial pressure data sequence; Taking the intracranial pressure average value as an initial value, carrying out dynamic characteristic analysis on the intermediate intracranial pressure data sequence, and adding a step mark aiming at a pressure step change section corresponding to a pathological step to obtain a target intracranial pressure data sequence; Performing segment fitting on the target intracranial pressure data sequence to obtain a segment fitting result, wherein the segment fitting result comprises a plurality of segment data and a best fitting model corresponding to each segment data; Reconstructing signals aiming at the segmented fitting result to obtain a multi-component mathematical model set for describing intracranial pressure dynamic change; the sampling rate of the set of multicomponent mathematical models is set to be in a preset high sampling rate range to generate a reconstructed intracranial pressure variation waveform.
2. The sparse data reconstruction method of claim 1, wherein the step-by-step signature is added to the intermediate intracranial pressure data sequence with the intracranial pressure mean as an initial value, for a pressure step-by-step segment corresponding to a pathological step, to obtain a target intracranial pressure data sequence, comprising: using the intracranial pressure average value as an initial value, and adopting a variable point detection algorithm based on statistical significance test to identify a pressure step change section in the intermediate intracranial pressure data sequence; Adding a step mark for each pressure step change section, and recording the step mark into a step mark candidate list, wherein the step mark comprises a step occurrence time point and a step change amplitude; Identifying each step marker in the step marker candidate list based on an adaptive threshold, wherein each step marker is identified to represent physiological fluctuation or pathological step; and deleting the step marks representing physiological fluctuation in the step mark candidate list, and adding the rest step marks to the corresponding pressure step change sections to obtain the target intracranial pressure data sequence.
3. The sparse data reconstruction method of claim 1, wherein performing a piecewise fit to the target intracranial pressure data sequence results in a piecewise fit result, comprising: dividing the target intracranial pressure data sequence into a plurality of segment data by taking a step mark as a boundary; for each piece of segmented data, adopting a plurality of fitting models of different types to perform dynamic curve fitting to obtain fitting errors corresponding to the fitting models; selecting a fitting model with the smallest fitting error as a best fitting model of the segmented data; and forming the segment fitting result by using all the segment data and the best fitting model corresponding to each segment data.
4. A sparse data reconstruction method according to any one of claims 1-3, wherein signal reconstruction is performed on the segmented fitting result to obtain a set of multicomponent mathematical models describing dynamic changes in intracranial pressure, comprising: for each piece of segment data, determining the best fit model corresponding to the piece of segment data as a trend term; Processing the segmented data by adopting a non-linear least square algorithm with constraint to obtain a periodic coefficient; determining a primary periodic fluctuation term and a secondary periodic fluctuation term according to the periodic coefficient; constructing a multi-component mathematical model describing intracranial pressure dynamic change according to the trend item, the primary periodic fluctuation item and the secondary periodic fluctuation item corresponding to each piece of segmented data; and carrying out aggregation processing on the multi-component mathematical models corresponding to all the segmented data to obtain the multi-component mathematical model set.
5. A sparse data reconstruction method according to any one of claims 1-3, wherein setting the sampling rate of the set of multicomponent mathematical models to be in a preset high sampling rate range to generate a reconstructed intracranial pressure variation waveform comprises: setting the sampling rate of the multi-component mathematical model set to be in a preset high sampling rate range to generate a reconstructed original intracranial pressure variation waveform; And carrying out moving average filtering treatment on the original intracranial pressure change waveform in the sliding time window to obtain a smooth and continuous intracranial pressure change waveform.
6. The method of claim 1, wherein obtaining the intracranial pressure sparse data stream acquired at a sampling rate within a predetermined low sampling rate range over a sliding time window comprises: acquiring intracranial pressure sparse data streams acquired in a sliding time window at a sampling rate within a preset low sampling rate range by adopting a sliding time window mechanism; Equal-length segmentation is carried out on the intracranial pressure sparse data stream to obtain a plurality of data segments; Calculating the average value of each data segment; Performing difference detection on each mean value by adopting t detection; when the inspection results show that the differences exist, averaging the average value of all the data segments to obtain a total average value; When the total mean value is smaller than a preset mean value threshold value, calculating the variance of the mean value of all the data segments; and when the variance is smaller than a preset variance threshold, determining that the intracranial pressure sparse data stream is steady state data, and finishing the initialization operation of the intracranial pressure sparse data stream.
7. The sparse data reconstruction method of claim 1 or 6, wherein preprocessing the intracranial pressure sparse data stream results in a continuous and efficient intermediate intracranial pressure data sequence, comprising: classifying each intracranial pressure data in the intracranial pressure sparse data stream into available data or unavailable data one by adopting Unet neural network, wherein the input of the Unet neural network is a one-dimensional array; Calculating the average value of two adjacent available data aiming at each unavailable data, and replacing the unavailable data by the average value; when all the unavailable data are completely replaced, a continuous and effective intermediate intracranial pressure data sequence is obtained.
8. A sparse data reconstruction device, comprising: The data acquisition unit is used for acquiring an intracranial pressure sparse data stream acquired at a sampling rate within a preset low sampling rate range in a sliding time window and calculating an intracranial pressure average value of the intracranial pressure sparse data stream; The preprocessing unit is used for preprocessing the intracranial pressure sparse data stream to obtain a continuous and effective intermediate intracranial pressure data sequence; the marking unit is used for carrying out dynamic characteristic analysis on the intermediate intracranial pressure data sequence by taking the intracranial pressure average value as an initial value, and adding a step mark aiming at a pressure step change section corresponding to a pathological step to obtain a target intracranial pressure data sequence; the fitting unit is used for carrying out piecewise fitting on the target intracranial pressure data sequence to obtain a piecewise fitting result, wherein the piecewise fitting result comprises a plurality of pieces of piecewise data and a best fitting model corresponding to each piece of piecewise data; The reconstruction model determining unit is used for carrying out signal reconstruction on the segmentation fitting result to obtain a multi-component mathematical model set for describing intracranial pressure dynamic change; And the reconstruction unit is used for setting the sampling rate of the multi-component mathematical model set to be in a preset high sampling rate range so as to generate a reconstructed intracranial pressure change waveform.
9. A computer storage medium storing at least one instruction which when executed by a processor implements a sparse data reconstruction method as claimed in any one of claims 1 to 7.
10. An electronic device, comprising a memory and a processor; The memory is used for storing at least one instruction; The processor is configured to execute the at least one instruction to implement the sparse data reconstruction method according to any one of claims 1-7.

Description

Sparse data reconstruction method and related device Technical Field The present invention relates to the field of data processing technologies, and in particular, to a sparse data reconstruction method and a related device. Background Intracranial pressure (INTRACRANIAL PRESSURE, ICP) is a key monitoring indicator in the treatment of critical diseases such as craniocerebral injury, cerebral hemorrhage, brain tumor, etc. Intracranial pressure can dynamically fluctuate along with heart beat and respiratory rhythm, and when the intracranial pressure suddenly rises when abnormal conditions such as edema and hemorrhage occur, the life of a patient is endangered. Conventional wired intracranial pressure monitoring devices acquire pressure data using a sampling rate that is several times higher than the frequency of the heart fluctuations (typically ≡100 Hz), followed by a running average process over a time window (e.g. 6 seconds) containing several fluctuation periods, to obtain a stable intracranial pressure average. However, since the implanted portion of the wired sensor is connected to an external monitoring device by a cable, the risk of infection is high during long-term monitoring, and thus the clinical routine monitoring period is generally not longer than 7 days. However, clinical studies indicate that patients with craniocerebral injury are at risk for life threatening due to elevated intracranial pressure within 30 days, and thus long Cheng Lu internal pressure monitoring is of great clinical significance. To achieve long Cheng Lu internal pressure monitoring, wireless intracranial pressure monitoring devices have been developed to achieve the goal of wireless implanted intracranial pressure monitoring. The wireless intracranial pressure monitoring equipment has small volume, limited capacity of a built-in battery and needs to realize long-time continuous pressure monitoring in a very small size space, so the requirement on power consumption is very high. To minimize power consumption and extend device operating time, wireless intracranial pressure monitoring devices are typically monitored at a low sampling rate (e.g., 1Hz or 0.5 Hz). However, at low sampling rates, often an intracranial pressure sparse data stream is acquired. To obtain a stable intracranial pressure average from these intracranial pressure sparse data streams, the low sampling rate data is typically calculated directly by a sliding average method. However, if a stable average is obtained, a long moving average window is required (e.g., tens of seconds or even minutes of data are required). When the intracranial pressure has pathological instant abrupt change (step rise), the method can not timely and accurately reflect the real change condition of the pressure, so that the early warning has serious hysteresis condition, which obviously does not meet the clinical requirement of real-time monitoring of critical patients. Furthermore, a simple moving average process smoothes out key dynamic features in the signal, especially physiological fluctuations (such as heart beat harmonics) with frequencies above the nyquist frequency (half the sampling rate), so that the reconstructed trend information is incomplete and cannot be exactly identical to the trend acquired by the high sampling rate device. Therefore, how to provide a sparse data reconstruction method, accurately reconstruct a real intracranial pressure signal and a variation trend thereof from an intracranial pressure sparse data stream with a low sampling rate, so as to provide a reliable and accurate diagnosis basis for a clinician, and become a technical problem to be solved by those skilled in the art. Disclosure of Invention In view of the above, the invention discloses a sparse data reconstruction method and a related device, so as to accurately reconstruct a real intracranial pressure signal and a change trend thereof from an intracranial pressure sparse data stream with a low sampling rate, and provide a reliable and accurate diagnosis basis for a clinician. A sparse data reconstruction method comprising: Acquiring an intracranial pressure sparse data stream acquired at a sampling rate within a preset low sampling rate range in a sliding time window, and calculating an intracranial pressure average value of the intracranial pressure sparse data stream; preprocessing the intracranial pressure sparse data stream to obtain a continuous and effective intermediate intracranial pressure data sequence; Taking the intracranial pressure average value as an initial value, carrying out dynamic characteristic analysis on the intermediate intracranial pressure data sequence, and adding a step mark aiming at a pressure step change section corresponding to a pathological step to obtain a target intracranial pressure data sequence; Performing segment fitting on the target intracranial pressure data sequence to obtain a segment fitting result, wherein the segment fitting result comprises