Multi-mode sensor for detecting cerebral hemorrhage

Abstract

The invention relates to the technical field of medical detection, in particular to a multi-mode sensor for detecting cerebral hemorrhage, which has the advantages that non-invasive continuous monitoring is realized by adopting multi-mode technologies such as non-ionizing radiation low-power microwave and near-infrared light, and various physical signals are provided for high-precision early diagnosis of cerebral hemorrhage. Therefore, the problem that a single signal mode is insufficient is solved, and precious treatment time is won for a patient.

Description

[0001] This application is a divisional application of the invention application filed on March 12, 2025, with application number 202510290386.2 and invention title "A non-invasive radio frequency microwave fusion near-infrared light brain hemorrhage detection device". Technical Field

[0002] This disclosure relates to the field of medical testing equipment, and in particular to a device that utilizes non-invasive radio frequency microwave technology combined with near-infrared light for rapid and accurate detection of acute cerebral hemorrhage. Background Technology

[0003] Intracerebral hemorrhage (ICH) is a common cerebrovascular disease characterized by rapid onset and progression, resulting in high rates of disability and mortality. Early and accurate diagnosis and assessment of the location and extent of acute ICH are crucial for timely and effective treatment. However, current primary methods for detecting ICH (such as CT and MRI) have limitations: CT and MRI equipment is expensive and bulky, typically limited to use in medical institutions; the time from onset to diagnosis is usually considerable, especially in remote areas or regions with limited medical resources; CT involves ionizing radiation, and long-term use may negatively impact patient health. While CT imaging is relatively fast, it carries radiation hazards, and its sensitivity to small hemorrhages may be limited in the early stages; MRI, while offering high resolution for soft tissue, is time-consuming, expensive, and unsuitable for rapid emergency screening in some patients with metal implants.

[0004] Currently, determining the type of stroke, especially between stroke / non-stroke and ischemic / hemorrhagic stroke, often requires hospital-based computed tomography (CT) and fMRI scans for differentiation. This leads to poor pre-hospital triage and delayed treatment, impacting postoperative recovery. While CT and fMRI are considered the gold standard, their long timeframe from onset to diagnosis and poor equipment portability easily cause patients to miss the golden window for stroke treatment.

[0005] The shortcomings of existing stroke monitoring devices are as follows: (1) Long detection time. The time taken includes the time from the onset of the disease to treatment. For example, CT and fMRI are large and inconvenient to move, so patients can only undergo brain imaging diagnosis after admission, resulting in long detection time. The time taken also refers to the long detection process itself. For example, EEG to identify the type of stroke, the electrodes need to be moistened and placed, resulting in long detection time.

[0006] (2) The detection range is not wide. For example, volume impedance phase-shift spectroscopy can only be used to identify anterior circulation large vessel occlusion (LVO), and is not suitable for differentiating posterior circulation LVO and cerebral hemorrhage patients.

[0007] (3) There is a risk of secondary injury during the detection process. Existing microwave stroke monitoring methods are ring-shaped, requiring the patient to wear the device on their head. However, in reality, stroke patients should minimize movement and remain lying down. Furthermore, if the patient remains lying down, the existing device requires moving the patient's head and pressing it onto the device. If the stroke site is located in the cerebral cortex, there is a risk of secondary injury. Moreover, although microwaves can penetrate the skull and sensitively detect differences in electromagnetic properties between different tissues (such as blood and brain tissue), a single microwave signal may be affected by the complexity of the head structure, resulting in insufficient spatial resolution. Near-infrared light has strong tissue penetration capabilities and can sensitively detect changes in blood oxygen saturation. However, relying solely on near-infrared light may make it difficult to accurately locate the brain hemorrhage area because light scattering and absorption by the skull can affect signal quality. Summary of the Invention

[0008] To address the aforementioned problems in the existing technology, this invention discloses a non-invasive radiofrequency microwave fusion near-infrared light brain hemorrhage detection device. This device employs multimodal technology to effectively overcome the shortcomings of a single signal mode by integrating the advantages of multiple physical signals.

[0009] In a first aspect, this disclosure proposes a non-invasive radiofrequency microwave fusion near-infrared light brain hemorrhage detection device. The device includes a multimodal sensor, a signal processing module, and a fusion analysis module. The multimodal sensor is configured to simultaneously acquire near-infrared spectral signals and radiofrequency microwave signals during brain hemorrhage monitoring. The signal processing module is configured to construct voxels using a Cartesian coordinate system based on the average volume of human head MRI data, match the position information of the multimodal sensor with the voxels, and acquire the near-infrared spectral signal and radiofrequency microwave signal corresponding to each voxel. The light intensity attenuation rate, average photon flight time, and photon count distribution variance extracted from the near-infrared spectral signal, along with the S21 scattering parameters, phase shift, and temporal impulse response peak value acquired from the radiofrequency microwave signal, are used as the signal feature vector corresponding to each voxel. The fusion analysis module is configured to generate a three-dimensional reconstruction map of the brain hemorrhage region based on the signal feature vector of the voxel and the spatial information of the voxel, using a trained deep learning fusion model. In one embodiment of the above technical solution, the deep learning fusion model includes a 3D convolutional neural network and a graph attention network; wherein: the 3D convolutional neural network is configured to extract the cerebral hemorrhage feature vector of all voxels based on the signal feature vector of all voxels; the graph attention network is configured to use voxels as graph nodes, and use the attention mechanism to learn the relationship between the cerebral hemorrhage features of the nodes to obtain the hemorrhage probability of each node, thereby generating a three-dimensional reconstruction effect map of the cerebral hemorrhage region.

[0010] In one embodiment of the above technical solution, the device further includes a reliability analysis module; the reliability analysis module is configured to calculate the confidence level of radio frequency data and near-infrared data of all multimodal sensors. If either confidence level is greater than a first set threshold, the signal data corresponding to that confidence level is resampled with a locally altered depth. If the sum of the two confidence levels is greater than a second set threshold, the radio frequency microwave signals and near-infrared spectral signals of all multimodal sensors are considered to have high confidence. The fusion analysis module is then used to generate a three-dimensional reconstruction map of the brain hemorrhage region based on the signal feature vector.

[0011] In one embodiment of the above technical solution, the confidence scores of the radio frequency data and near-infrared data of all multimodal sensors are calculated as follows:

[0012]

[0013] In the formula: For the confidence level of radio frequency data from n multimodal sensors, Let n be the confidence scores for near-infrared data from multiple multimodal sensors. Let i be the phase shift of the i-th multimodal sensor. This is the phase offset threshold; This represents the attenuation amount of the S21 parameter in the i-th channel. The threshold for the S21 scattering parameters; Let be the light intensity attenuation rate of the j-th multimodal sensor. This is the light intensity attenuation threshold; Let be the average time of flight of photons from the j-th multimodal sensor. This represents the average time-of-flight threshold for photons.

[0014] In one embodiment of the above technical solution, the voxel size is 16mm x 16mm x 16mm.

[0015] Secondly, this disclosure proposes a multimodal sensor for detecting cerebral hemorrhage. The multimodal sensor includes an MCU control module, a microwave signal transceiver module, a near-infrared light signal transceiver module, and a probe fusing near-infrared and microwave signals. The MCU control module includes an STM32 microcontroller. The microwave transceiver module includes an RF transmit link and an RF receive link. The RF transmit link generates radio frequency (RF) signals using an LMX2595, which are then transmitted from a patch electrode via a QPA2211 and an ADL5920. The RF receive link receives microwave signals from the patch electrode, which are then transmitted to the STM32 microcontroller after passing through a low-noise amplifier, a mixer, and an analog-to-digital converter. 32; The near-infrared signal transceiver module includes an optical signal generation optical path and an optical signal receiving optical path. The optical signal generation optical path drives an LMH6521 high-speed constant current source to excite a multi-channel LP785-SAV50 laser diode to generate near-infrared light, which is then emitted through a light source probe. The optical signal receiving optical path receives the light through the S13360-3050CS silicon photomultiplier of the light source probe, and sends it to the STM32 via a preamplifier OPA1612, a filter LTC1569, and an analog-to-digital converter. The microstrip of the microwave transceiver module and the silicon photomultiplier of the near-infrared signal transceiver module are coupled through a probe that integrates near-infrared and microwave signals.

[0016] The aforementioned multimodal sensor technology employs non-ionizing radiation low-power microwaves and near-infrared light, among other multimodal technologies, to achieve non-invasive continuous monitoring. This provides the advantage of multiple physical signals for high-precision early diagnosis of cerebral hemorrhage, overcoming the limitations of a single signal mode and saving valuable treatment time for patients.

[0017] In one embodiment of the above technical solution, the STM32 generates periodic TTL pulses while the microwave transceiver module starts the radio frequency signal transmission and the near-infrared light source modulation of the near-infrared signal transceiver module. The analog-to-digital converter of the radio frequency receiving link and the analog-to-digital converter of the near-infrared signal share the same trigger signal.

[0018] In one embodiment of the above technical solution, the substrate of the probe that integrates near-infrared and microwave is U-shaped, with dimensions of 10mm×10mm×0.6mm. The outer ring of the U-shape is a ring-shaped ground line GND with a line width of 1mm, used to constrain the radio frequency microwave signal. The middle ring has a side length of 7mm and a line width of 1mm, and is connected to the feed position. Holes are opened below the optical end of the transmitting end for inserting light-guiding multimode optical fiber, and holes are opened below the SiPM of the receiving end for inserting the light source probe SiPM. An IPEX socket is used on the back of the probe, and radio frequency microwave signals are transmitted using a coaxial cable.

[0019] In one embodiment of the above technical solution, the wavelength range of the near-infrared signal transceiver module during operation is 650-950 nm.

[0020] In one embodiment of the above technical solution, the microwave transceiver module operates in a frequency range of 0.5-10 GHz.

[0021] The beneficial effects of this disclosure include its applicability in the detection of acute cerebral hemorrhage. Because microwave / radio frequency signals are highly sensitive to the electromagnetic properties of brain tissue (such as dielectric constant and conductivity), they can detect the nature and extent of cerebral hemorrhage. Near-infrared light is sensitive to changes in oxyhemoglobin and deoxyhemoglobin concentrations, allowing for the assessment of hemorrhage-related metabolic and physiological changes. Therefore, the complementary signals of radio frequency microwaves and near-infrared light can be utilized to achieve measurements from superficial to deep. Since both low-power microwaves and infrared light are non-ionizing radiation, they are harmless to human tissue. The device design allows for non-invasive, continuous, and safe monitoring, avoiding patient discomfort. The device can be designed as a portable device, suitable for pre-hospital emergency care, screening and monitoring patients in primary hospitals or remote areas. Signal processing algorithms and intelligent analysis modules enable real-time diagnosis. Furthermore, this disclosed device achieves low-cost, high-precision early diagnosis of cerebral hemorrhage through multimodal fusion of radio frequency microwaves and near-infrared light, saving patients valuable treatment time. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 , one A schematic diagram of the multimode sensor structure in one implementation method.

[0024] Figure 2 , one A schematic diagram of the probe in one implementation method.

[0025] Figure 3 , one A schematic diagram of the data fusion method in one implementation.

[0026] Figure 4 , one A schematic diagram of the three-dimensional reconstruction of the bleeding area in one implementation method. Detailed Implementation

[0027] The following will describe clearly and completely how the technical solution of this case is implemented, with reference to the accompanying drawings. Obviously, the described implementation methods are only a part of the implementation methods of this case, and not all of the implementation methods.

[0028] (a) Multimodal sensors A brain hemorrhage detection device integrating near-infrared and microwave technologies is being implemented. By combining near-infrared spectroscopy (NIRS) and microwave detection techniques, it utilizes the sensitivity of near-infrared light to hemoglobin and the response characteristics of microwaves to changes in tissue conductivity to achieve highly sensitive, non-invasive detection of brain hemorrhage. Specifically, a multi-modal sensor array collaboratively acquires near-infrared spectral (NIRS) and microwave signals, and combines this with machine learning algorithms to achieve high-precision detection and localization of brain hemorrhage.

[0029] The multimodal sensor includes an MCU main control module, a microwave signal transceiver module, and a near-infrared light signal transceiver module.

[0030] See Figure 1 The MCU main control module includes an STM32 microcontroller, a power supply unit, an RXO7050 crystal oscillator, and an LMK00304. The power supply unit provides the STM32 with the necessary power for operation. An external crystal oscillator, the RXO7050, provides a stable 10MHz reference clock, which is used for data distribution and level conversion via a low-jitter buffer, the LMK00304.

[0031] The microwave signal transceiver module includes an RF transmitting link and an RF receiving link. The RF transmitting link generates RF signals using an LMX2595, which are then transmitted from the patch electrode via a QPA2211 and an ADL5920. The RF receiving link receives microwave signals from the patch electrode, which are then transmitted to the STM32 via a low-noise amplifier, a mixer, and an analog-to-digital converter.

[0032] The microwave signal transceiver module can generate, adjust the frequency, and transmit and receive microwave signals. It can sense the change in dielectric constant of the brain hemorrhage area through ultra-wideband (UWB) signals in the 0.5-10GHz frequency band.

[0033] The near-infrared optical signal transceiver module includes an optical signal generation optical path and an optical signal receiving optical path. The optical signal generation optical path drives an LMH6521 high-speed constant current source to excite a multi-channel LP785-SAV50 laser diode to generate near-infrared light, which is then emitted by the light source probe. The optical signal receiving optical path receives the light from the silicon photomultiplier (SiPM) of the S13360-3050CS of the light source probe, and then sends it to the STM32 via a preamplifier OPA1612, a filter LTC1569, and an analog-to-digital converter.

[0034] The near-infrared module obtains brain tissue blood oxygenation parameters by transmitting light signals of a specific wavelength through the transmitting unit and receiving unit. The wavelength range of the near-infrared signal transceiver module is 650-950 nm.

[0035] The microstrip of the microwave transceiver module and the silicon photomultiplier of the near-infrared signal transceiver module are coupled by a probe that integrates near-infrared and microwave rectangular microstrip.

[0036] See Figure 2 The probe, which integrates near-infrared and microwave rectangular microstrip sensors, has a U-shaped substrate measuring 10mm × 10mm × 0.6mm. The outer ring is a 1mm wide circular ground line (GND) used to confine the RF microwave signal. The middle ring, with a side length of 7mm and a line width of 1mm, connects to the feed position. Two 2mm diameter holes are located below the feed position: one below the optical transmitter and the other below the SiPM receiver. The hole below the optical transmitter is used to insert a light-guiding multimode fiber, while the hole below the SiPM receiver is used to insert the light source probe SiPM. Since the photoelectric signals do not interfere with each other at the physical layer, a fused arrangement effect can be achieved. Figure 2 The front of the probe is shown, and an IPEX socket is used on the back of the probe for radio frequency microwave signal transmission via a coaxial cable.

[0037] The probe substrate is made of RO4350B with a dielectric constant of 3.66 and a loss tangent of 0.0037.

[0038] (II) Detection of cerebral hemorrhage Taking the use of eight multimodal sensors as an example, the eight multimodal sensors are arranged in a circular array, covering the frontal lobe, temporal lobe, and parietal lobe. During the detection of cerebral hemorrhage, near-infrared spectral signals and radio frequency microwave signals are acquired simultaneously.

[0039] (2.1) Changes in dielectric constant of the cerebral hemorrhage region This enables a multimodal sensor to acquire radio frequency microwave signals that detect changes in the dielectric constant of the brain hemorrhage area.

[0040] Specifically, the STM32 MCU module controls the LMX2595 wideband frequency synthesizer for parameter writing and control. An external crystal oscillator RXO7050 provides a stable 10MHz reference clock signal. This clock signal is distributed to the REFIN pin of the LMX2595 through a low-jitter buffer LMK00304, and phase-locked loop (PLL) frequency synthesis is performed based on the reference clock input. The frequency divider parameters are configured by setting the integer division ratio (N) and fractional division ratio (F) of the PLL via the SPI interface, calculated as follows:

[0041] In the formula: For output frequency, This is the reference clock frequency.

[0042] Based on the above formula and the fact that the LMX2595 integrates a multi-core voltage-controlled oscillator (VCO), it can provide frequency output from 300MHz to 10GHz. Simultaneously, by enabling the LMX2595's low-noise mode and optimizing the loop filter bandwidth, high-frequency spurious signals are suppressed. The target power and waveform output are then achieved through power amplification and filtering. This device is designed with a pre-stage driver, a final amplification stage, and a bandpass filter. In the LMX2595 pre-stage driver, the power is boosted to +20 dBm after driver amplification. With the QPA2211 microwave power amplifier receiving the drive signal, the gain is amplified to +45 dBm, while an isolator is connected to the output to prevent reflection damage. A cavity bandpass filter is used to filter out spurious signals outside the frequency band. The device can also switch operating frequency bands according to the scanning depth. Its operating logic is that the LMX2595 selects the path selection via an RF switch. When the detection depth is >3cm, the LMX2595 direct frequency division is selected; when the detection depth is ≤3cm, the high-frequency path is enabled. The RF switch is controlled by a TTL level output from the STM32 MCU.

[0043] In addition, this device also includes VSWR detection and temperature protection. The ADL5920 monitors the forward / reflected power in real time, and when the VSWR is too high, the STM32 cuts off the power amplifier power. Furthermore, by monitoring the heat dissipation temperature, when the temperature is too high, the power is reduced until the power is cut off.

[0044] The receiving signal link includes low-noise amplification, down-conversion processing, and digital demodulation. The received signal first passes through a QPL9547 low-noise amplifier, and an HMC8193 mixer mixes the signal with the local oscillator (LO) to output an intermediate frequency (IF) signal. An AD7768 analog-to-digital converter (ADC) samples the signal, and an STM32 MCU demodulates the I / Q components, calculating the amplitude and phase.

[0045] (2.2) Detection of brain tissue blood oxygenation parameters (HbO2, Hb) The multimodal sensor acquires near-infrared spectral signals to detect blood oxygenation parameters (HbO2, Hb) in brain tissue.

[0046] Specifically, the pulse generator of the STM32 MCU is used to drive the LMH6521 high-speed constant current source to excite the multi-channel LP785-SAV50 laser diode to emit 785 nm, 50 mW near-infrared light, realizing the optical signal transmission based on STM32 MCU control, as follows.

[0047] After the laser diode is powered on, the temperature is stabilized to 25±0.1°C by the TCLDM9 temperature control module, ensuring wavelength drift <0.1 nm. An integrated photodiode (PD) feedback regulates the drive current, stabilizing the output optical power at 50 mW. Using time-division multiplexing and frequency coding, multiple light sources are sequentially lit at 1 ms intervals to avoid crosstalk. Each light source is modulated with a 1 kHz sine wave via the voltage-controlled input of the LMH6521 for subsequent frequency domain separation, resulting in an output parameter of 20% pulse width duty cycle and an instantaneous peak power of 250 mW. The light source is transmitted to a probe integrating near-infrared and microwave rectangular microstrip optical fibers via an optical fiber bundle.

[0048] At the probe, light source reception is accomplished by the photodetector S13360-3050CS (SiPM). The optical signal then passes through a preamplifier OPA1612 → filter LTC1569 → analog-to-digital converter AD7768. After undergoing multiple scatterings within the brain tissue, a portion of the optical signal reaches the photodetector module S13360-3050CS.

[0049] The multimodal sensor front end integrates a spherical lens to focus scattered light onto the SiPM photosensitive surface of the S13360-3050CS, and then suppresses and eliminates ambient light through a 785BP10 filter.

[0050] A DE1 reference signal synchronized with the light source modulation frequency was generated using an STM32 microcontroller, and the effective signal was extracted using lock-in amplification (LAP). This signal was input to a preamplifier OPA1612, which converted the SiPM output current into a voltage signal. An AD7768 was used to acquire the signal at a rate of 256 kSPS, with multiple resolution settings, simultaneously capturing all multi-detector channels. The signal was stored in a buffer for optical signal analysis, which included time-domain and frequency-domain analysis. The time-domain analysis recorded photon arrival times using time-correlated single-photon counting, characterized by a longer optical path length and a decreased photon count rate at the hemorrhage location. The frequency-domain analysis performed a Fast Fourier Transform on the 1 kHz modulated signal to extract the fundamental amplitude reflecting light intensity attenuation and the phase shift reflecting optical path changes. Finally, based on Monte Carlo simulation, a pre-built brain tissue light transmission model was constructed, and the absorption coefficient distribution of the hemorrhage region was inverted using an iterative optimization algorithm.

[0051] (III) Signal Processing The average volume of the human head MRI data is obtained. A Cartesian coordinate system is established based on the average volume, and the data is divided into a uniform 16 x 16 x 16 grid, with each grid cell called a voxel. The position information of the multimodal sensor is matched with the voxels to obtain the near-infrared spectral signal and radio frequency microwave signal corresponding to each voxel.

[0052] Next, data preprocessing will be performed, including: S21 scattering parameters (amplitude attenuation), phase shift (Δφ), and peak time-domain impulse response (TDR) are extracted from the radio frequency microwave signal, and high-frequency noise is removed using fourth-order wavelet transform.

[0053] Light intensity attenuation rate, average photon flight time, and photon count distribution variance were extracted from near-infrared spectral signals.

[0054] The above signal processing is accomplished by constructing a signal processing module.

[0055] (iv) Reliability Analysis Based on the above, it can be seen that the near-infrared module monitors the dynamic changes of HbO2 / Hb, and the microwave module captures the abnormal dielectric constant caused by hemorrhage.

[0056] First, multimodal data is synchronized. Since the synchronization mechanism includes hardware time synchronization and software time calibration, a low-jitter temperature-compensated crystal oscillator (RXO7050) is used to provide synchronization clocks for the LMX2595, the light source, and the ADC. Furthermore, a clock buffer (LMK00304) distributes the clock to each subsystem, ensuring a global clock deviation of <1 ns. Additionally, the device uses an STM32 to generate periodic TTL pulses to simultaneously initiate RF signal transmission and NIR light source modulation. The ADC in the RF receiving link shares the same trigger signal with the NIR light source's ADC, and the data frame header is marked with a synchronization timestamp, with error accuracy controlled within ±10 ns. The RF signal board transmission delay is compensated in software using a pre-calibrated fixed delay (≈50 ns) in the MCU. The optical signal transmission delay is dynamically corrected based on a time-of-flight (TOF) model. Spatial alignment is performed based on coordinate system calibration using the RF, optical, and SiPM sensors.

[0057] Secondly, data reliability is assessed by calculating the confidence levels of radio frequency data and near-infrared data from all multimodal sensors. The calculations are as follows:

[0058]

[0059] In the formula: For the confidence level of radio frequency data from n multimodal sensors, Let n be the confidence scores for near-infrared data from multiple multimodal sensors. Let i be the phase shift of the i-th multimodal sensor. This is the phase offset threshold; This represents the attenuation amount of the S21 parameter in the i-th channel. The threshold for the S21 scattering parameters; Let be the light intensity attenuation rate of the j-th multimodal sensor. This is the light intensity attenuation threshold; Let be the average time of flight of photons from the j-th multimodal sensor. This represents the average time-of-flight threshold for photons.

[0060] If the confidence level of the radio frequency data is greater than the first set threshold, or the confidence level of the near-infrared data is greater than the first set threshold, then resampling with a localized change in detection depth is performed. For example, if the initial detection depth is greater than 3 cm, then high-frequency resampling with a depth of 3 cm or less is used.

[0061] like When the second threshold is set, it is judged as high confidence, the signal feature vector is constructed and input into the fusion analysis module, and the bleeding location is output.

[0062] (v) Fusion Analysis Module The fusion analysis module uses a convolutional neural network (CNN) to extract and fuse features from multi-source data, generating a heatmap of bleeding probability and a 3D reconstructed image. By using multi-source data, data fusion overcomes the limitations of single-modality datasets.

[0063] In this section, the brain tissue blood oxygenation parameters obtained by the near-infrared module and the dielectric constant changes of the hemorrhage area obtained by the microwave module are spatiotemporally registered and feature-level fused through the fusion analysis module to achieve high-precision hemorrhage detection and localization.

[0064] First, construct the feature vector. The feature dimensions of the radio frequency data are: 3 (S21, Δφ, TDR peak) × 8-channel antenna = 24 dimensions The feature dimensions of near-infrared data are: 3 (ΔI / I0, TOF, variance) × 8-channel optical path = 24-dimensional Radio frequency data features and near-infrared data features are spliced ​​and fused into a signal feature vector (48 dimensions), and spatial grid information is added (see [link]). Figure 3 Feature dimensions after fusion: 48-dimensional vector (24+24) + spatial grid information (X, Y, Z) = 51-dimensional Among them, the spatial grid information is the center position information of a voxel.

[0065] Then, normalization is performed using the Z-score model, with each feature dimension being normalized. Processing. (Among them) It is a parameter value. It is the average value. That is the standard deviation.

[0066] Next, a fusion model of a 3D convolutional neural network and a graph attention network (GAT) is set up to fuse 51-dimensional feature vectors into a spatial grid. Specifically, the 3D convolutional layers are as follows: Layer 1, 3×3×3 kernel, 1→32 input channels, stride 1, padding 1. Layer 2, 3×3×3 kernel, 32→64 input channels, stride 2, downsampling. Layer 3, 3×3×3 kernel, 64→128 input channels, stride 2. The graph attention layer (GAT) maps voxels to graph nodes, and the node features include radio frequency and near-infrared parameters. The graph attention weights are calculated as follows:

[0067] in, , Let W represent the node features, and W be the learnable weight matrix.

[0068] The output layer is a fully connected layer (128→64→1) and a sigmoid activation function, outputting a heatmap of bleeding probability.

[0069] In one embodiment, the three-dimensional reconstruction effect of the bleeding area is as follows: Figure 4 As shown in the figure, the fusion model extracts and fuses features from multiple data sources to generate a three-dimensional reconstructed image of the brain hemorrhage area, enabling high-precision hemorrhage detection and localization.

[0070] The resolution of a 3D reconstructed image of a brain hemorrhage region is related to voxel segmentation. For example, if a voxel is 16mm x 16mm x 16mm in size and the brain region at the hemorrhage site is 100mm, then the resolution of the hemorrhage site is 6mm³.

[0071] (VI) Summary This disclosure addresses the shortcomings of existing stroke monitoring devices, such as long detection times, limited detection range, and the risk of secondary injury. It proposes a non-invasive radiofrequency microwave fusion near-infrared light brain hemorrhage detection device. This device utilizes multimodal fusion of radiofrequency microwave and near-infrared light to achieve low-cost, high-precision early diagnosis of brain hemorrhage, thus buying valuable treatment time for patients.

[0072] From the above description of the embodiments, those skilled in the art will clearly understand that the device disclosed herein can be implemented using software plus necessary general-purpose hardware, or it can be implemented using dedicated hardware including dedicated integrated circuits, dedicated CPUs, dedicated memory, dedicated components, etc. Generally, any function performed by a computer program can be easily implemented using corresponding hardware, and the specific hardware structure used to implement the same function can be diverse, such as analog circuits, digital circuits, or dedicated circuits. However, for the purposes of this disclosure, software implementation is often a preferred implementation method.

[0073] Although the embodiments of this disclosure have been described above in conjunction with the accompanying drawings, this disclosure is not limited to the specific embodiments and application fields described above. The specific embodiments described above are merely illustrative and instructive, and not restrictive. Those skilled in the art can make many other forms based on the guidance of this specification and without departing from the scope of protection of the claims of this disclosure, and all of these are within the scope of protection of this disclosure.

Claims

1. A multimodal sensor for detecting cerebral hemorrhage, characterized in that, The multimodal sensor includes an MCU control module, a microwave signal transceiver module, a near-infrared optical signal transceiver module, and a probe that fuses near-infrared and microwave signals; wherein: The MCU control module includes an STM32; The microwave transceiver module includes an RF transmitting link and an RF receiving link. The RF transmitting link generates RF signals from an LMX2595, which are then transmitted from a patch electrode via a QPA2211 and an ADL5920. The RF receiving link receives microwave signals from a patch electrode, which are then transmitted to an STM32 via a low-noise amplifier, a mixer, and an analog-to-digital converter. The near-infrared signal transceiver module includes an optical signal generation optical path and an optical signal receiving optical path. The optical signal generation optical path drives an LMH6521 high-speed constant current source to excite a multi-channel LP785-SAV50 laser diode to generate near-infrared light, which is then emitted by the light source probe. The optical signal receiving optical path receives the light from the silicon photomultiplier S13360-3050CS of the light source probe, and then sends it to the STM32 via a preamplifier OPA1612, a filter LTC1569, and an analog-to-digital converter. The microstrip of the microwave transceiver module and the silicon photomultiplier of the near-infrared signal transceiver module are coupled through a probe that integrates near-infrared and microwave signals.

2. The multimodal sensor according to claim 1, characterized in that, The STM32 generates periodic TTL pulses, while the microwave transceiver module initiates RF signal transmission and the near-infrared light source modulation of the near-infrared signal transceiver module. The analog-to-digital converter of the RF receiving link and the analog-to-digital converter of the near-infrared link share the same trigger signal.

3. The multimodal sensor according to claim 1, characterized in that, The substrate of the probe that integrates near-infrared and microwave signals is U-shaped, with a ring-shaped ground line GND on the outer edge of the U-shape to constrain the radio frequency microwave signal. It is connected to the feed position. There are holes below the optical end of the transmitting end for inserting the light guide multimode fiber, and holes below the SiPM of the receiving end for inserting the light source probe SiPM. An IPEX socket is used on the back of the probe to transmit the radio frequency microwave signal using a coaxial cable.

4. The multimodal sensor according to claim 1, characterized in that, The wavelength range of the near-infrared signal transceiver module is 650-950 nm.

5. The multimodal sensor according to claim 1, characterized in that, The microwave transceiver module operates in a frequency range of 0.5-10 GHz.

6. The multimodal sensor according to claim 3, characterized in that, The dimensions of the U-shaped substrate are 10mm×10mm×0.6mm, with an outer ring line width of 1mm, a middle ring side length of 7mm, and a line width of 1mm.

7. The multimodal sensor according to claim 3, characterized in that, The substrate is made of RO4350B with a dielectric constant of 3.66 and a loss tangent of 0.0037.

8. A brain hemorrhage monitoring device, characterized in that, The device comprises a multimodal sensor array consisting of the multimodal sensors described in any one of claims 1-7.

9. The apparatus according to claim 8, characterized in that, The multimodal sensor array is arranged in a ring array, covering the frontal lobe, temporal lobe, and parietal lobe.

10. The apparatus according to claim 8, characterized in that, The device switches the operating frequency band according to the scanning depth. When the detection depth is greater than 3cm, LMX2595 direct frequency division is selected; when the detection depth is less than or equal to 3cm, the high-frequency path is activated.

Images