A non-contact tumor cell in-vitro screening device and method

Abstract

The application belongs to the technical field of medical apparatus and instruments, and discloses a non-contact tumor cell in-vitro screening device and method, which comprises a wireless radio frequency probe built-in wideband radio frequency transceiver antenna unit, a multi-stage adjustable impedance matching network, a near-field coupling radiation structure and a radio frequency detection circuit, which are used for transmitting 1MHz-100GHz frequency band electromagnetic waves and synchronously receiving echo signals and near-field scattering signals; a signal processing system is connected with the wireless radio frequency probe and is used for pre-processing and feature extraction of the dual-mode signals; a control unit is internally provided with an AI fusion system, which is used for feature fusion analysis of the pre-processed dual-mode signals and outputs a benign and malignant determination result; the AI fusion system comprises an IC-CF-DMAS image reconstruction module, a one-dimensional convolutional neural network module, a cross-attention fusion module, a weighted voting decision module and a classification head, realizes deep interaction and accurate determination of cross-modal features, and realizes low cost, no radiation, rapidness, non-invasiveness, portability and high accuracy of early tumor screening.

Description

Technical Field

[0001] This application belongs to the field of medical device technology, specifically relating to a non-contact in vitro tumor cell screening device and method based on low-power radio frequency technology. Background Technology

[0002] Early cancer screening is a crucial step in improving cancer prevention and control. However, current clinically used cancer diagnostic methods have many limitations and cannot meet the needs of large-scale, widespread early screening. PET-CT scans not only cost thousands of yuan per examination, but the equipment purchase cost can reach millions of yuan. Furthermore, they rely on radioactive tracers, posing radiation risks to the human body and potentially increasing the probability of secondary cancer, making them unsuitable for frequent early screening. Magnetic resonance elastography (MRI) technology has a signal acquisition time of about 30 minutes or even longer, resulting in extremely low detection efficiency and making it difficult to meet the large-scale population screening needs of high-traffic medical centers or public health projects. Invasive diagnostic methods such as gastroscopy and biopsy cause significant physical pain to patients and carry risks of complications such as bleeding and infection, leading to low patient acceptance. Many high-risk individuals delay early intervention due to resistance to screening.

[0003] Meanwhile, current mainstream tumor detection equipment is all large-scale, with MRI machines occupying tens of square meters and weighing several tons, making them extremely difficult to carry. This hinders their rapid deployment in disaster relief sites, remote rural areas, and other locations lacking specialized medical equipment, thus limiting the coverage of early tumor screening. Furthermore, the signal processing and judgment methods of existing detection technologies have low levels of intelligence, making it difficult to accurately identify tumor tissue and further impacting screening accuracy.

[0004] Therefore, there is a need to design a non-contact in vitro tumor cell screening device based on low-power radio frequency technology to improve the above problems.

[0005] Invention / Invention Content To address the problems of existing technologies, this invention provides a non-contact in vitro tumor cell screening device, comprising: This wireless radio frequency (RF) probe is a handheld, portable design. It integrates a wideband RF transceiver antenna unit, a multi-level adjustable impedance matching network, a near-field coupled radiation structure, and an RF detection circuit. The RF detection circuit includes a microcontroller and an RF transceiver module. The multi-level adjustable impedance matching network is connected in series between the RF transceiver module and the wideband RF transceiver antenna unit, and is electrically connected to the microcontroller. The near-field coupled radiation structure is coupled to the wideband RF transceiver antenna unit to concentrate and confine the emitted electromagnetic waves to the near-field region at the probe's front end. It is used to emit multiple electromagnetic waves within the 1MHz-100GHz frequency band that are sensitive to the state of biological cells and tissues, and to receive echo signals from normal and diseased biological tissues, as well as to collect near-field scattered signals from the tumor target region. The signal processing system, connected to the wireless radio frequency probe, includes a general imaging unit and a data recognition and analysis unit, used for image enhancement and feature extraction of near-field scattered signals, and for filtering, analyzing and screening key features of echo signals. The control unit is connected to the signal processing system and has a built-in AI fusion system. The AI ​​fusion system is based on Maxwell's equations and combined with a physical information neural network framework. It is used to perform feature fusion analysis on the preprocessed echo signal and near-field scattering signal, and output the probability judgment result and confidence level of the good or bad of the test area. A visual interface, connected to the control unit via an HDMI interface, is used to display near-field scattering images, judgment results, and working status information of the wireless radio frequency probe, mark abnormal areas, and generate and export standardized screening reports. The power management unit is electrically connected to the wireless radio frequency probe, signal processing system, control unit, and visual interface, and is used to supply power to the various modules of the device.

[0006] Furthermore, the wideband RF transceiver antenna unit is a multi-band resonant antenna structure adapted to the 1MHz-100GHz frequency band; The near-field coupled radiation structure is a near-field coupled antenna structure that constrains the propagation range of electromagnetic waves; The multi-stage adjustable impedance matching network is composed of multiple adjustable reactive elements; The radio frequency detection circuit also includes a signal calibration module. The microcontroller integrates an analog-to-digital converter and a digital-to-analog converter. The microcontroller is electrically connected to the radio frequency transceiver module, the signal calibration module, and the multi-level adjustable impedance matching network, respectively, to coordinate the synchronous operation of each module. The radio frequency detection circuit is used to provide feedback on the impedance matching status. The multi-level adjustable impedance matching network adjusts the impedance parameters in real time within a wide frequency range of 1MHz-100GHz to ensure the stability of electromagnetic wave transmission and reception. The signal calibration module adopts a closed-loop feedback mechanism to automatically adjust the frequency deviation and maintain the stability of electromagnetic wave transmission and reception within a preset error range.

[0007] Furthermore, the general imaging unit integrates an electromagnetic imaging module for noise suppression, edge enhancement, and structural feature extraction of near-field scattering signals, extracting structural and texture features of the detection area to enhance near-field scattering images; the data recognition and analysis unit is equipped with a signal amplification and filtering module for preprocessing, phase analysis, amplitude analysis, categorized noise filtering, feature selection, and standardization of echo signals, ultimately extracting key features for subsequent diagnostic analysis.

[0008] Furthermore, the AI ​​fusion system includes: The IC-CF-DMAS image reconstruction module is used to process near-field scattering signals. It optimizes data preprocessing by grouping scattering parameters and transforming the time domain, and embeds a coherence factor weighting and iterative correction mechanism into the DMAS basic operation to extract the scattering feature map and location information of the detection area. A one-dimensional convolutional neural network module is used to process echo signals. It uses convolutional kernels of different scales to process radio frequency echo signal sequences in parallel, capture their time-frequency domain dynamic characteristics, and output high-dimensional feature vectors of the echo signals. The first classification sub-network is connected to the IC-CF-DMAS image reconstruction module and is used to classify the scattering feature map output by the IC-CF-DMAS image reconstruction module to generate a first preliminary classification result. The second classification sub-network is connected to the one-dimensional convolutional neural network module and is used to classify the high-dimensional feature vector of the echo signal output by the one-dimensional convolutional neural network module to generate a second preliminary classification result. The cross-attention fusion module, connected to the IC-CF-DMAS image reconstruction module and the one-dimensional convolutional neural network module, is used to output a fused feature vector by constructing a mutual attention correlation between the scattering image features and the radio frequency echo signal features, and obtain a third preliminary classification result after processing by the built-in classifier. The weighted voting decision module is connected to the first classification sub-network, the second classification sub-network, and the cross-attention fusion module, respectively. It is used to dynamically allocate weights to the first, second, and third preliminary classification results based on the confidence of each channel on the validation set, and generate the final comprehensive judgment result through weighted voting. The classification head, connected to the weighted voting decision module, uses a fully connected layer and a Softmax function to output the probability judgment result and confidence level of the benign or malignant region to be tested.

[0009] Furthermore, the power management unit includes a rechargeable lithium-ion battery, an overload / short circuit protection circuit, a multi-channel voltage regulator output module, and a power monitoring module; The rechargeable lithium-ion battery has a capacity of 10000mAh and is used for energy storage and long-lasting power supply. The multi-channel voltage regulator output module is used to output stable DC voltage at multiple levels to adapt to the power supply requirements of different modules in the device. The power monitoring module is used to display the remaining power and estimated working time in real time, and trigger a prompt when the power is low; The overload / short circuit has a response time of less than 50ms and quickly cuts off the power supply in case of abnormal current or short circuit.

[0010] A non-contact in vitro tumor cell screening method, implemented based on any of the screening devices described above, includes the following steps: Step 1: Use a wireless radio frequency probe to emit electromagnetic waves in the 1MHz-100GHz frequency band into biological tissue, and use a wideband radio frequency transceiver antenna unit to simultaneously receive the echo signal and the near-field scattered signal of the detection area; use a multi-level adjustable impedance matching network to adjust the impedance parameters in real time to ensure the stability of electromagnetic wave transmission and reception. Step 2: Image enhancement and feature extraction of the near-field scattering signal are performed using a general imaging unit, and the echo signal is filtered, analyzed, and key features are selected using a data recognition and analysis unit; Step 3: Input the preprocessed echo signal and near-field scattering signal into the AI ​​fusion system. After feature extraction, cross-modal fusion, and weighted voting, the system outputs the probability determination result and confidence level of the benign or malignant region to be tested. Step 4: The visual interface displays the judgment results in multiple dimensions, marks abnormal areas, verifies them by combining the status data of the wireless radio frequency probe, and automatically generates and exports a standardized screening report.

[0011] Furthermore, in step 2, the general imaging unit performs noise suppression, edge enhancement, and structural feature extraction on the near-field scattering signal through the electromagnetic imaging module, extracting the structural and texture features of the detection area; the data recognition and analysis unit performs preprocessing, phase analysis, amplitude analysis, type-specific noise filtering, feature screening, and standardization on the echo signal through the signal amplification and filtering module, extracting key features for subsequent diagnostic analysis.

[0012] Furthermore, step 3 specifically includes: Step 3.1: The near-field scattering signal is processed by the IC-CF-DMAS image reconstruction module. The data preprocessing is optimized by scattering parameter grouping and time-domain transformation. The coherence factor weighting and iterative correction mechanism is embedded in the DMAS basic operation to extract the scattering feature map and location information of the detection area. Step 3.2: Process the echo signal using a one-dimensional convolutional neural network module. Utilize convolutional kernels of different scales to process the radio frequency echo signal sequence in parallel, capture its dynamic characteristics in the time and frequency domain, and output a high-dimensional feature vector of the echo signal. Step 3.3: Classify the scattering feature map output from Step 3.1 using the first classification sub-network to generate a first preliminary classification result; classify the high-dimensional feature vector of the echo signal output from Step 3.2 using the second classification sub-network to generate a second preliminary classification result. Step 3.4: Construct the mutual attention relationship between the scattering image features and the radio frequency echo signal features through the cross-attention fusion module, output the fused feature vector, and obtain the third preliminary classification result after processing by the built-in classifier; Step 3.5: The weighted voting decision module dynamically assigns weights to the first, second, and third preliminary classification results based on the confidence level of each channel on the validation set, and generates the final comprehensive judgment result through weighted voting. Step 3.6: Using a fully connected layer and the Softmax function in the classification head, output the probability judgment result and confidence level of the benign or malignant region to be tested. If the confidence level is lower than the preset threshold, trigger a suspected warning.

[0013] The beneficial effects that this application can produce include: 1) Low cost: Through the integrated design of the handheld portable wireless radio frequency probe, the complex mechanical and optical structure of large equipment is eliminated. The core hardware consists of modular components such as wireless radio frequency probe, signal processing system, and control unit, which greatly reduces the overall purchase and maintenance costs of the equipment, far lower than traditional equipment such as PET-CT and MRI. At the same time, the screening device of this application does not require the use of consumables such as radioactive tracers and contrast agents, and there is no additional consumable cost per test. The test cost is less than 1 / 10 of that of traditional technologies. It is suitable for the clinical testing needs of medical institutions at all levels and can also meet the large-scale population screening requirements of public health projects. It breaks down the barriers to the popularization of early cancer screening from the cost perspective and achieves the technical effect of low cost and wide applicability. 2) Radiation-free / Non-invasive: Based on low-power radio frequency technology, this non-contact, radiation-free detection method eliminates the need to insert any detection components into the human body. The detection is completed simply by placing the handheld radio frequency probe against the skin of the area being examined. The entire process is non-invasive and painless, completely eliminating the risk of complications associated with invasive procedures. Furthermore, the low-power radio frequency electromagnetic waves emitted by the device are non-ionizing radiation, causing no radioactive pollution and posing no harm to the human body. This allows for regular and frequent screening of high-risk groups, significantly improving patient acceptance and cooperation, achieving a non-invasive, radiation-free, and highly acceptable technical outcome. 3) Fast and efficient: The handheld portable wireless radio frequency probe is small in size and light in weight, and can be operated with one hand. All modules of the entire screening device are integrated and miniaturized, and can be stored in a dedicated protective case. A single person can complete the transportation and deployment of the equipment. At the same time, the device is equipped with a 10000mAh high-capacity rechargeable lithium-ion battery, which can work continuously for 8 hours on a single full charge. No external power supply is required, and it can be used normally in remote rural areas and field rescue sites without power supply. This solves the problem of traditional large equipment being "difficult to move and limited in scene", and achieves the technical effect of high portability, long battery life and full scene adaptability. 4) High portability: The handheld portable wireless radio frequency probe is small in size and light in weight, and can be operated with one hand. All modules of the entire screening device are integrated and miniaturized, and can be stored in a dedicated protective case. A single person can complete the transportation and deployment of the equipment. At the same time, the device is equipped with a 10000mAh high-capacity rechargeable lithium-ion battery, which can work continuously for 8 hours on a single full charge without the need for an external power source. It can be used normally in remote rural areas and field rescue sites without power supply, solving the problems of traditional large equipment being difficult to move and limited in scenarios. It achieves the technical effects of high portability, long battery life and full scenario adaptability. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall structure of the device of the present invention; Figure 2 This is a block diagram of the internal radio frequency detection circuit of the wireless radio frequency probe of the present invention; Figure 3 This is a flowchart of the non-contact in vitro tumor cell screening method of the present invention; Figure 4 This is a schematic diagram of the echo signal received by the wireless radio frequency probe of the present invention; Figure 5 This is a schematic diagram of the electromagnetic wave signal emitted by the wireless radio frequency probe of the present invention; Figure 6 The image of the detection area is extracted using the IC-CF-DMAS image reconstruction module in this embodiment of the invention. Detailed Implementation

[0015] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0016] Example 1: Device Structure Please see Figure 1-2 The present invention provides a non-contact in vitro tumor cell screening device, comprising: This wireless radio frequency (RF) probe is a handheld, portable design. It integrates a wideband RF transceiver antenna unit, a multi-level adjustable impedance matching network, a near-field coupled radiation structure, and an RF detection circuit. The RF detection circuit includes a microcontroller and an RF transceiver module. The multi-level adjustable impedance matching network is connected in series between the RF transceiver module and the wideband RF transceiver antenna unit, and is electrically connected to the microcontroller. The near-field coupled radiation structure is coupled to the wideband RF transceiver antenna unit to concentrate and confine the emitted electromagnetic waves to the near-field region at the probe's front end. It is used to emit multiple electromagnetic waves within the 1MHz-100GHz frequency band that are sensitive to the state of biological cells and tissues, and to receive echo signals from normal and diseased biological tissues, as well as to collect near-field scattered signals from the tumor target region. The signal processing system, connected to the wireless radio frequency probe, includes a general imaging unit and a data recognition and analysis unit, used for image enhancement and feature extraction of near-field scattered signals, and for filtering, analyzing and screening key features of echo signals. The control unit is connected to the signal processing system and has a built-in AI fusion system. The AI ​​fusion system is based on Maxwell's equations and combined with a physical information neural network framework. It is used to perform feature fusion analysis on the preprocessed echo signal and near-field scattering signal, and output the probability judgment result and confidence level of the good or bad of the test area. The visual interface is connected to the control unit via a high-definition multimedia interface (HDMI) to display near-field scattering images, judgment results, and working status information of the wireless radio frequency probe, mark abnormal areas, and generate and export standardized screening reports. The power management unit is electrically connected to the wireless radio frequency probe, signal processing system, control unit, and visual interface, and is used to supply power to the various modules of the device.

[0017] It should be noted that the wireless radio frequency probe achieves wideband near-field radiation function through the coordinated cooperation of the multi-level adjustable impedance matching network, the broadband radio frequency transceiver antenna unit, and the near-field coupling radiation structure. Furthermore, the wideband RF transceiver antenna unit is a multi-band resonant antenna structure adapted to the 1MHz-100GHz frequency band; The near-field coupled radiation structure is a near-field coupled antenna structure that constrains the propagation range of electromagnetic waves; The multi-stage adjustable impedance matching network is composed of multiple adjustable reactive elements; The radio frequency detection circuit also includes a signal calibration module. The microcontroller integrates an analog-to-digital converter (ADC) and a digital-to-analog converter (DAC). The microcontroller is electrically connected to the radio frequency transceiver module, the signal calibration module, and the multi-level adjustable impedance matching network to coordinate the synchronous operation of each module. The radio frequency detection circuit is used to provide feedback on the impedance matching status. The multi-level adjustable impedance matching network adjusts the impedance parameters in real time within a wide frequency range of 1MHz-100GHz to ensure the stability of electromagnetic wave transmission and reception. The signal calibration module adopts a closed-loop feedback mechanism to automatically adjust the frequency deviation and maintain the stability of electromagnetic wave transmission and reception within a preset error range. Furthermore, the general imaging unit integrates an electromagnetic imaging module for noise suppression, edge enhancement, and structural feature extraction of near-field scattering signals, extracting structural and texture features of the detection area to enhance near-field scattering images; the data recognition and analysis unit is equipped with a signal amplification and filtering module for preprocessing, phase analysis, amplitude analysis, categorized noise filtering, feature selection, and standardization of echo signals, ultimately extracting key features for subsequent diagnostic analysis.

[0018] It should be noted that the electromagnetic imaging module integrated into the general imaging unit systematically processes the raw near-field scattering data through a multi-level algorithm processing flow, significantly improving data feature parameters and optimizing subsequent analysis efficiency. First, the scattering parameters are transformed and grouped in the time domain, using inverse Fourier transform technology to convert the frequency domain signal into a time domain signal, providing a time dimension for delay matching. Second, based on the different propagation speeds of electromagnetic waves in different tissues, spatial distance modeling and delay time formulas are used to compensate for the calculated delay time in the time domain signal, ensuring that scattering signals from different paths are synchronously superimposed at the reconstruction point, avoiding image blurring caused by delay. The general imaging unit integrates an electromagnetic imaging module, which is a dedicated imaging chip TITMS320C6748. The electromagnetic imaging module is configured to perform noise suppression, edge enhancement, and structural feature extraction processing of near-field scattering signals, ultimately realizing the transformation of the original near-field scattering signal into a high-definition, high-feature-recognition near-field scattering signal, providing reliable image feature data for subsequent AI fusion system analysis. The electromagnetic imaging module processes near-field scattered signals in the following steps: scattering parameter grouping, frequency-to-time domain conversion, propagation delay compensation, signal alignment and fusion, and scattering intensity calculation. The collected frequency domain scattering parameters S(f,rx,φ) (where f is the frequency, rx is the receiving antenna number, and φ is the antenna array angle) are divided into two groups according to the antenna array angle φ. The odd-numbered angle group S odd (f,rx,φ odd )(φ odd =1,3,5,...N φ -1) As the initial illumination signal, it provides basic scattering information; even-numbered angle group S even (f,rx,φ even )(φ even =2,4,6,...N φ () is used as an offset lighting signal to compensate for signal errors caused by angular deviation; An inverse Fourier transform is applied to the grouped scattering parameters to convert the frequency-domain scattering signal into a time-domain signal Γ(t,rx,φodd), obtaining the time-dimensional characteristics of electromagnetic wave propagation in tissue and providing a reference for subsequent delay matching. The time-domain signal more directly reflects the propagation delay of electromagnetic waves in tissue, providing a time-dimensional reference for subsequent delay matching. Due to the differences in the dielectric constant εr of human tissues, electromagnetic waves propagate at different speeds in different tissues, resulting in propagation delays. By accurately calculating the delay time, it is ensured that the signals received by different antennas are aligned in the time dimension. Define the three-dimensional coordinate matrix of any reconstructed point C within the imaging domain, and calculate the coordinates of this reconstructed point relative to the transmitting antenna A. Tx Receiving antenna A Rx Spatial distance PC-Txφ odd (i,rx,l) and PC-Rxφ odd (i,rx,l), where i is the index of the reconstruction point and l is the propagation path length, based on the formula for the propagation speed of electromagnetic waves in a medium: v=c / εr 1 / 2 (where c is the speed of light in a vacuum, and εr is the dielectric constant of human tissue), derive the delay time τ(i,rx,φ) odd ):τ(i,rx,φ odd )=(εr 1 / 2 (PC-Rxφ odd (i,rx,l))) / c, and compensate the calculated delay time into the time domain signal to ensure that the scattering signals of different paths are synchronously superimposed at the reconstruction point, avoiding image blurring caused by delay; Time axis alignment is performed on the time domain signal after compensation for delay to ensure that the timestamps of the signals collected by all receiving antennas are consistent at the same reconstruction point i. Then, point-by-point multiplication is performed on the aligned multi-antenna time domain signal to enhance the correlation between signals. The multiplied signals are integrated over the entire time domain and summed to obtain the preliminary scattering intensity of each reconstructed point in the imaging domain. This completes the conversion of near-field scattering signals into scattering intensity data, providing a data foundation for subsequent image generation and feature extraction. The signal amplification and filtering module of the data recognition and analysis unit processes the echo signal by sequentially performing preprocessing, phase analysis, amplitude analysis, type-specific noise filtering, feature selection, and standardization operations, ultimately extracting key features for subsequent diagnostic analysis. The specific process is as follows: Signal preprocessing: The acquired raw radio frequency echo signal is amplified to enhance the weak signal strength and improve detection sensitivity; filtering is performed simultaneously to initially filter out basic background noise and irrelevant interference signals in the echo signal, thus completing the basic optimization of the signal.

[0019] Phase analysis: First, a high-pass filter with a cutoff frequency of 10kHz is used to filter out the DC component in the preprocessed echo signal; then, the core frequency bands of 0.01GHz-10GHz and 40GHz-80GHz are focused and calculated, the echo phase difference of specific narrowband frequency points in the band is calculated, and phase consistency detection is performed. The phase difference quantitative index reflects the difference in tissue characteristics, providing a phase dimension basis for tissue identification. Amplitude analysis: First, a moving average filter with a 10-sampling-point window is used to smooth the original amplitude signal; second, according to the formula α=20log∣A0 / A n | / d (where A0 is the amplitude of the transmitted signal, A n The amplitude attenuation coefficient α is calculated for the received signal amplitude (d is the detection depth). This coefficient reflects the tissue's absorption and scattering characteristics of electromagnetic energy. Finally, a 128-point Fast Fourier Transform (FFT) is performed on the amplitude signal to extract key spectral features such as peak frequency, spectral width, and number of spectral peaks, thereby comprehensively quantifying the energy distribution characteristics of the signal.

[0020] Classified noise filtering: A multi-strategy joint filtering method is adopted, and the db4 wavelet basis soft threshold filtering is used to effectively suppress Gaussian noise; adaptive median filtering is used to suppress impulse noise, and adaptive linear prediction filtering is used to suppress clutter interference originating from biological tissues, so as to achieve precise and targeted removal of different types of noise in the echo signal and ensure signal purity.

[0021] Feature selection: The feature selection algorithm (Relief-F) is used to select 12-15 dimensional features with strong organizational discrimination from the 28-dimensional original feature space obtained after the above processing, eliminate redundant and low-value features, and retain core effective features. Feature standardization: The core features obtained by screening are normalized by standard score (Z-score) to eliminate scale differences and potential interference between features, and obtain key features with uniform feature dimensions and stable data features. These features are then transmitted to the control unit as the signal feature basis for subsequent AI fusion system diagnostic analysis. Furthermore, the AI ​​fusion system includes: The IC-CF-DMAS image reconstruction module is used to process near-field scattering signals. It optimizes data preprocessing by grouping scattering parameters and transforming time domains. It also embeds a coherence factor weighting and iterative correction mechanism into the basic operation of the nonlinear beamforming and image reconstruction algorithm (DMAS, Delay Multiply and Sum) to extract the scattering features and location information of the detection area. A one-dimensional convolutional neural network module is used to process echo signals. It uses convolutional kernels of different scales to process radio frequency echo signal sequences in parallel, capture their time-frequency domain dynamic characteristics, and output high-dimensional feature vectors of the echo signals. The first classification sub-network is connected to the IC-CF-DMAS image reconstruction module and is used to classify the scattering feature map output by the IC-CF-DMAS image reconstruction module to generate a first preliminary classification result. The second classification sub-network is connected to the one-dimensional convolutional neural network module and is used to classify the high-dimensional feature vector of the echo signal output by the one-dimensional convolutional neural network module to generate a second preliminary classification result. The cross-attention fusion module, connected to the IC-CF-DMAS image reconstruction module and the one-dimensional convolutional neural network module, is used to output a fused feature vector by constructing a mutual attention correlation between the scattering image features and the radio frequency echo signal features, and obtain a third preliminary classification result after processing by the built-in classifier. The weighted voting decision module is connected to the first classification sub-network, the second classification sub-network, and the cross-attention fusion module, respectively. It is used to dynamically allocate weights to the first, second, and third preliminary classification results based on the confidence of each channel on the validation set, and generate the final comprehensive judgment result through weighted voting. The classification head is connected to the weighted voting decision module. It uses a fully connected layer and a Softmax function to output the probability judgment result and confidence level of the good or bad nature of the test area. It should be noted that the IC-CF-DMAS image reconstruction module is an optimized design based on the DMAS algorithm. The DMAS algorithm initially extracts the scattering intensity of the reconstructed points through "delay alignment-signal multiplication-integral summation". It introduces a coherence factor to weight the reliability of different antenna signals, and uses coherence factor calculation and weighted optimization methods to amplify the effective signal with high coherence, suppress noise or interference signals with low coherence, and improve the image signal-to-noise ratio. It introduces an iterative correction mechanism to gradually adjust the scattering intensity through multiple rounds of iteration, and uses inverse distance weighting update and convergence judgment to output the final scattering intensity distribution map.

[0022] Furthermore, the power management unit includes a rechargeable lithium-ion battery, an overload / short circuit protection circuit, a multi-channel voltage regulator output module, and a power monitoring module; The rechargeable lithium-ion battery has a capacity of 10000mAh and is used for energy storage and long-lasting power supply. The multi-channel voltage regulator output module is used to output stable DC voltage at multiple levels to adapt to the power supply requirements of different modules in the device. The power monitoring module is used to display the remaining power and estimated working time in real time, and trigger a prompt when the power is low; The overload / short circuit has a response time of less than 50ms and quickly cuts off the power supply when there is an abnormal current or short circuit. It should be noted that the power management unit provides stable power to all modules of the device, including a rechargeable lithium-ion battery, an overload / short circuit protection circuit, a multi-channel voltage regulator output module, and a power monitoring module. The rechargeable lithium-ion battery has a large capacity and long battery life, and a single charge can meet the device's continuous operation requirements for a long time. The multi-channel voltage regulator output module can output stable DC voltage at multiple levels to adapt to the power supply requirements of different modules. The overload / short circuit has a fast response speed and quickly cuts off the power supply in case of abnormal current or short circuit to protect the safety of system components. Power supply can be automatically restored after the fault is cleared. The power monitoring module collects battery status data in real time, displays the remaining power and estimated working time on a visual interface, and provides a low battery warning to avoid affecting the screening work. Example 2: Signal Processing Method like Figure 3 As shown, a non-contact in vitro tumor cell screening method based on the screening device includes the following steps: Step 1: Use a wireless radio frequency probe to emit electromagnetic waves in the 1MHz-100GHz frequency band into biological tissue, and use a wideband radio frequency transceiver antenna unit to simultaneously receive the echo signal and the near-field scattered signal of the detection area; use a multi-level adjustable impedance matching network to adjust the impedance parameters in real time to ensure the stability of electromagnetic wave transmission and reception. It should be noted that dual-modal data acquisition is used. Based on the biological tissue location to be screened, the corresponding preset mode is selected on the visualization interface. The device automatically loads appropriate parameters such as detection depth and radio frequency band; it then activates the radio frequency transceiver module to transmit electromagnetic waves within the 1MHz-100GHz frequency band. The wideband radio frequency transceiver antenna unit simultaneously receives the echo signal and the near-field scattered signal from the detection area; a multi-level adjustable impedance matching network adjusts the impedance parameters in real time to ensure the stability of electromagnetic wave transmission and reception. Step 2: Image enhancement and feature extraction of the near-field scattering signal are performed using a general imaging unit, and the echo signal is filtered, analyzed, and key features are selected using a data recognition and analysis unit; It should be noted that in signal preprocessing and feature extraction, the signal processing system receives the raw dual-modal data acquired by the wireless radio frequency probe. The general imaging unit uses the electromagnetic imaging module to perform noise suppression, edge enhancement, and structural feature extraction on the near-field scattering signal, extracting the structural and texture features of the detection area. The data recognition and analysis unit uses the signal amplification and filtering module to preprocess the echo signal, perform phase analysis, amplitude analysis, categorized noise filtering, feature selection, and standardization, ultimately extracting key features for subsequent diagnostic analysis. Step 3: Input the preprocessed echo signal and near-field scattering signal into the AI ​​fusion system. After feature extraction, cross-modal fusion, and weighted voting, the system outputs the probability determination result and confidence level of the benign or malignant region under test. The specific process of this step is as follows: Step 3.1: The near-field scattering signal is processed by the IC-CF-DMAS image reconstruction module. The data preprocessing is optimized by scattering parameter grouping and time-domain transformation. The coherence factor weighting and iterative correction mechanism is embedded in the DMAS basic operation to extract the scattering feature map and location information of the detection area. Step 3.2: Process the echo signal using a one-dimensional convolutional neural network module. Utilize convolutional kernels of different scales to process the radio frequency echo signal sequence in parallel, capture its dynamic characteristics in the time and frequency domain, and output a high-dimensional feature vector of the echo signal. Step 3.3: Classify the scattering feature map output from Step 3.1 using the first classification sub-network to generate a first preliminary classification result; classify the high-dimensional feature vector of the echo signal output from Step 3.2 using the second classification sub-network to generate a second preliminary classification result. Step 3.4: Construct the mutual attention relationship between the scattering image features and the radio frequency echo signal features through the cross-attention fusion module, output the fused feature vector, and obtain the third preliminary classification result after processing by the built-in classifier; Step 3.5: The weighted voting decision module dynamically assigns weights to the first, second, and third preliminary classification results based on the confidence level of each channel on the validation set, and generates the final comprehensive judgment result through weighted voting. Step 3.6: Using a fully connected layer and the Softmax function in the classification head, output the probability judgment result and confidence level of the benign or malignant region to be tested. If the confidence level is lower than a preset threshold, a suspected alert is triggered. Step 4: The visual interface displays the judgment results in multiple dimensions, marks abnormal areas, verifies them in conjunction with the status data of the wireless radio frequency probe, and automatically generates and exports a standardized screening report; Example 3: Application Case of Breast Tumor Screening This embodiment uses early screening for breast tumors as a specific application scenario to explain the implementation process of the present invention in detail.

[0023] Device Preparation: The screening device is powered on and completes self-calibration. The remaining power supply is ≥85%. The signal calibration module sets the signal level error threshold to ±0.05dB, and the multi-level adjustable impedance matching network completes initial impedance adjustment. Select the preset breast cancer screening mode on the visual interface. The device automatically loads the appropriate parameters: detection depth 1-4cm, RF core operating frequency bands focused f1=2GHz, f2=5GHz, f3=8GHz, and AI fusion system confidence threshold 90%.

[0024] Signal transmission and acquisition: The operator holds the wireless radio frequency probe and places the tip of the probe against the upper outer quadrant of the breast (a high-incidence area for tumors). Transmission is initiated, and the broadband radio frequency transceiver antenna unit is activated. Figure 5 It features multi-frequency electromagnetic wave emission (f1=2GHz, amplitude 20dB; f2=5GHz, amplitude 28dB; f3=8GHz, amplitude 22dB). Simultaneously, it receives near-field scattered and echo signals from breast tissue, with the echo signal forming at the three core frequencies of f1, f2, and f3. Figure 4 Consistent signal characteristics.

[0025] Signal processing: Data recognition and analysis unit Figure 4 The corresponding echo signals were preprocessed, and the following calculations were made: attenuation coefficient α1 = 8 dB / cm and phase difference 8.5° at frequency f1; attenuation coefficient α2 = 16 dB / cm and phase difference 18.2° at frequency f2; and attenuation coefficient α3 = 10.4 dB / cm and phase difference 11.3° at frequency f3. The attenuation coefficient at frequency f2 is significantly higher than that of normal breast tissue (α ≤ 7 dB / cm for normal tissue), and the phase shift is also significantly greater than that of normal tissue (phase difference ≤ 5° for normal tissue). Simultaneously, the general imaging unit processes the near-field scattered signals, such as... Figure 6 As shown, the scattering characteristics and location information of the suspected lesion area in the upper outer quadrant of the breast were extracted (approximately 1.2cm × 0.9cm in size).

[0026] AI fusion judgment: The IC-CF-DMAS image reconstruction module processes the scattering signal and outputs a scattering feature map. The one-dimensional convolutional neural network module processes echo signal features and outputs a 256-dimensional high-dimensional feature vector. The first sub-classification network classifies the scattering feature map and obtains a malignancy probability of 95%. The second classification subnetwork classifies the echo feature vectors, obtaining a malignancy probability of 96%. The cross-attention fusion module combines two features, and the malignancy probability is 97% after passing through the built-in classifier. The weighted voting decision module assigns weights based on the confidence level of the validation set (fusion feature 0.5, echo feature 0.35, scattering feature 0.15), and calculates the overall probability of malignancy using the weighted average: 97%×0.5+96%×0.35+95%×0.15=96.35%; Classification head output: The probability of malignancy in the upper outer quadrant of the breast is 96.35%, with a confidence level of 98% (above the 90% threshold, no suspected indication).

[0027] Results Display: The visual interface simultaneously displays the signal feature comparison spectrum, near-field scattering enhancement image, core parameter data table and final judgment result, and automatically generates a standardized screening report.

[0028] This embodiment is only for... Figure 4 , Figure 5 The signal characteristics are a typical application in tumor screening. This signal analysis logic is also applicable to tumor screening in other parts of the body, such as the thyroid, liver, and lymph nodes. It is worth noting that this application, through the integrated design of a handheld portable wireless radio frequency probe, abandons the complex mechanical and optical structures of large equipment. The core hardware consists of modular components such as a wireless radio frequency probe, a signal processing system, and a control unit, significantly reducing the overall purchase and maintenance costs of the equipment, which are far lower than those of traditional equipment such as PET-CT and MRI. At the same time, the screening device of this application does not require the use of consumables such as radioactive tracers and contrast agents, and there are no additional consumable costs per test. The testing cost is less than 1 / 10 of that of traditional technologies. It is suitable for the clinical testing needs of medical institutions at all levels and can also meet the requirements of large-scale population screening in public health projects. From a cost perspective, it breaks down the barriers to the popularization of early cancer screening and achieves the technical effect of low cost and wide applicability. This application utilizes low-power radio frequency technology to achieve a non-contact, radiation-free detection method. No detection components need to be inserted into the human body; the detection is completed simply by placing the handheld radio frequency probe against the skin of the area to be examined. The entire process is non-invasive and painless, completely eliminating the risk of complications associated with invasive procedures. Furthermore, the low-power radio frequency electromagnetic waves emitted by the device are non-ionizing radiation, causing no radioactive pollution and posing no harm to the human body. This allows for regular and frequent screening of high-risk groups, significantly improving patient acceptance and cooperation, and achieving a non-invasive, radiation-free, and highly acceptable technical effect. This application, through a fully automated technical design, reduces the overall time for tumor screening to less than one minute. The feature extraction of the signal processing system and the benign / malignant determination of the AI ​​fusion system are both performed at the millisecond level. The entire process from signal acquisition to report generation requires no manual intervention. Operators only need to perform simple operations such as probe fitting, mode selection, and report export. A single device can complete hundreds or even thousands of screenings per day, with detection efficiency hundreds of times higher than traditional magnetic resonance imaging technology. It is perfectly adapted to the large-scale population screening needs of high-traffic medical centers and grassroots screening points, achieving rapid, efficient, and automated technical results, and significantly improving the coverage efficiency of early tumor screening. This application presents a handheld portable wireless radio frequency probe. The probe is small in size and lightweight, and can be operated with one hand. All modules of the entire screening device are integrated and miniaturized, and can be stored in a dedicated protective case. A single person can complete the transportation and deployment of the device. At the same time, the device is equipped with a 10000mAh high-capacity rechargeable lithium-ion battery, which can work continuously for 8 hours on a single full charge without the need for an external power source. It can be used normally in remote rural areas and field rescue sites without power supply, solving the problem of traditional large equipment being "difficult to move and limited in scenarios". It achieves the technical effect of high portability, long battery life and full scenario adaptability.

[0029] The above description is merely a few embodiments of this application and is not intended to limit this application in any way. Although this application discloses preferred embodiments as described above, it is not intended to limit this application. Any changes or modifications made by those skilled in the art without departing from the scope of the technical solution of this application using the disclosed technical content are equivalent to equivalent implementation cases and fall within the scope of the technical solution.

Claims

1. A non-contact in vitro tumor cell screening device, characterized in that, include: This wireless radio frequency (RF) probe is a handheld, portable design. It integrates a wideband RF transceiver antenna unit, a multi-level adjustable impedance matching network, a near-field coupled radiation structure, and an RF detection circuit. The RF detection circuit includes a microcontroller and an RF transceiver module. The multi-level adjustable impedance matching network is connected in series between the RF transceiver module and the wideband RF transceiver antenna unit, and is electrically connected to the microcontroller. The near-field coupled radiation structure is coupled to the wideband RF transceiver antenna unit to concentrate and confine the emitted electromagnetic waves to the near-field region at the probe's front end. It is used to emit multiple electromagnetic waves within the 1MHz-100GHz frequency band that are sensitive to the state of biological cells and tissues, and to receive echo signals from normal and diseased biological tissues, as well as to collect near-field scattered signals from the tumor target region. The signal processing system, connected to the wireless radio frequency probe, includes a general imaging unit and a data recognition and analysis unit, used for image enhancement and feature extraction of near-field scattered signals, and for filtering, analyzing and screening key features of echo signals. The control unit is connected to the signal processing system and has a built-in AI fusion system. The AI ​​fusion system is based on Maxwell's equations and combined with a physical information neural network framework. It is used to perform feature fusion analysis on the preprocessed echo signal and near-field scattering signal, and output the probability judgment result and confidence level of the good or bad of the test area. A visual interface, connected to the control unit via an HDMI interface, is used to display near-field scattering images, judgment results, and working status information of the wireless radio frequency probe, mark abnormal areas, and generate and export standardized screening reports. The power management unit is electrically connected to the wireless radio frequency probe, signal processing system, control unit, and visual interface, and is used to supply power to the various modules of the device.

2. The non-contact in vitro tumor cell screening device according to claim 1, characterized in that, The wideband RF transceiver antenna unit is a multi-band resonant antenna structure adapted to the 1MHz-100GHz frequency band. The near-field coupled radiation structure is a near-field coupled antenna structure that constrains the propagation range of electromagnetic waves; The multi-stage adjustable impedance matching network is composed of multiple adjustable reactive elements; The radio frequency detection circuit also includes a signal calibration module. The microcontroller integrates an analog-to-digital converter and a digital-to-analog converter. The microcontroller is electrically connected to the radio frequency transceiver module, the signal calibration module, and the multi-level adjustable impedance matching network, respectively, to coordinate the synchronous operation of each module. The radio frequency detection circuit is used to provide feedback on the impedance matching status. The multi-level adjustable impedance matching network adjusts the impedance parameters in real time within a wide frequency range of 1MHz-100GHz to ensure the stability of electromagnetic wave transmission and reception. The signal calibration module adopts a closed-loop feedback mechanism to automatically adjust the frequency deviation and maintain the stability of electromagnetic wave transmission and reception within a preset error range.

3. The non-contact in vitro tumor cell screening device according to claim 1, characterized in that, The general imaging unit integrates an electromagnetic imaging module for noise suppression, edge enhancement, and structural feature extraction of near-field scattering signals, extracting structural and texture features of the detection area to enhance near-field scattering images. The data recognition and analysis unit is equipped with a signal amplification and filtering module for preprocessing, phase analysis, amplitude analysis, categorized noise filtering, feature selection, and standardization of echo signals, ultimately extracting key features for subsequent diagnostic analysis.

4. The non-contact in vitro tumor cell screening device according to claim 1, characterized in that, The AI ​​fusion system includes: The IC-CF-DMAS image reconstruction module is used to process near-field scattering signals. It optimizes data preprocessing by grouping scattering parameters and transforming the time domain, and embeds a coherence factor weighting and iterative correction mechanism into the DMAS basic operation to extract the scattering feature map and location information of the detection area. A one-dimensional convolutional neural network module is used to process echo signals. It uses convolutional kernels of different scales to process radio frequency echo signal sequences in parallel, capture their time-frequency domain dynamic characteristics, and output high-dimensional feature vectors of the echo signals. The first classification sub-network is connected to the IC-CF-DMAS image reconstruction module and is used to classify the scattering feature map output by the IC-CF-DMAS image reconstruction module to generate a first preliminary classification result. The second classification sub-network is connected to the one-dimensional convolutional neural network module and is used to classify the high-dimensional feature vector of the echo signal output by the one-dimensional convolutional neural network module to generate a second preliminary classification result. The cross-attention fusion module, connected to the IC-CF-DMAS image reconstruction module and the one-dimensional convolutional neural network module, is used to output a fused feature vector by constructing a mutual attention correlation between the scattering image features and the radio frequency echo signal features, and obtain a third preliminary classification result after processing by the built-in classifier. The weighted voting decision module is connected to the first classification sub-network, the second classification sub-network, and the cross-attention fusion module, respectively. It is used to dynamically allocate weights to the first, second, and third preliminary classification results based on the confidence of each channel on the validation set, and generate the final comprehensive judgment result through weighted voting. The classification head, connected to the weighted voting decision module, uses a fully connected layer and a Softmax function to output the probability judgment result and confidence level of the benign or malignant region to be tested.

5. The non-contact in vitro tumor cell screening device according to claim 1, characterized in that, The power management unit includes a rechargeable lithium-ion battery, an overload / short circuit protection circuit, a multi-channel voltage regulator output module, and a power monitoring module. The rechargeable lithium-ion battery has a capacity of 10000mAh and is used for energy storage and long-lasting power supply. The multi-channel voltage regulator output module is used to output stable DC voltage at multiple levels to adapt to the power supply requirements of different modules in the device. The power monitoring module is used to display the remaining power and estimated working time in real time, and trigger a prompt when the power is low; The overload / short circuit has a response time of less than 50ms and quickly cuts off the power supply in case of abnormal current or short circuit.

6. A non-contact in vitro screening method for tumor cells, characterized in that, Based on the screening device according to any one of claims 1-5, the process includes the following steps: Step 1: Use a wireless radio frequency probe to emit electromagnetic waves in the 1MHz-100GHz frequency band into biological tissue, and use a wideband radio frequency transceiver antenna unit to simultaneously receive the echo signal and the near-field scattered signal of the detection area; use a multi-level adjustable impedance matching network to adjust the impedance parameters in real time to ensure the stability of electromagnetic wave transmission and reception. Step 2: Image enhancement and feature extraction of the near-field scattering signal are performed using a general imaging unit, and the echo signal is filtered, analyzed, and key features are selected using a data recognition and analysis unit; Step 3: Input the preprocessed echo signal and near-field scattering signal into the AI ​​fusion system. After feature extraction, cross-modal fusion, and weighted voting, the system outputs the probability determination result and confidence level of the benign or malignant region to be tested. Step 4: The visual interface displays the judgment results in multiple dimensions, marks abnormal areas, verifies them by combining the status data of the wireless radio frequency probe, and automatically generates and exports a standardized screening report.

7. The non-contact in vitro tumor cell screening method according to claim 6, characterized in that, In step 2, the general imaging unit performs noise suppression, edge enhancement and structural feature extraction on the near-field scattering signal through the electromagnetic imaging module, and extracts the structural and texture features of the detection area. The data recognition and analysis unit performs preprocessing, phase analysis, amplitude analysis, type-specific noise filtering, feature screening, and standardization on the echo signal through the signal amplification and filtering module, extracting key features for subsequent diagnostic analysis.

8. The non-contact in vitro tumor cell screening method according to claim 6, characterized in that, Step 3 specifically includes: Step 3.1: The near-field scattering signal is processed by the IC-CF-DMAS image reconstruction module. The data preprocessing is optimized by scattering parameter grouping and time-domain transformation. The coherence factor weighting and iterative correction mechanism is embedded in the DMAS basic operation to extract the scattering feature map and location information of the detection area. Step 3.2: Process the echo signal using a one-dimensional convolutional neural network module. Utilize convolutional kernels of different scales to process the radio frequency echo signal sequence in parallel, capture its dynamic characteristics in the time and frequency domain, and output a high-dimensional feature vector of the echo signal. Step 3.3: Classify the scattering feature map output from Step 3.1 using the first classification sub-network to generate a first preliminary classification result; classify the high-dimensional feature vector of the echo signal output from Step 3.2 using the second classification sub-network to generate a second preliminary classification result. Step 3.4: Construct the mutual attention relationship between the scattering image features and the radio frequency echo signal features through the cross-attention fusion module, output the fused feature vector, and obtain the third preliminary classification result after processing by the built-in classifier; Step 3.5: The weighted voting decision module dynamically assigns weights to the first, second, and third preliminary classification results based on the confidence level of each channel on the validation set, and generates the final comprehensive judgment result through weighted voting. Step 3.6: Using a fully connected layer and the Softmax function in the classification head, output the probability judgment result and confidence level of the benign or malignant region to be tested. If the confidence level is lower than the preset threshold, trigger a suspected warning.

Images