A microwave detection system for cancer cells

Abstract

This invention relates to a microwave detection system for cancer cells, belonging to the field of biomedical sensing technology. The system includes a metal substrate; a microwave sensor located on the upper surface of the metal substrate and attached to it; a microfluidic chip located on the upper surface of the microwave sensor and attached to it; and a cancer cell concentration detection module connected to the microwave sensor, used to collect capacitive sensing data and microwave sensing data related to the microwave sensor when a cell solution flows on the microfluidic chip. This invention offers advantages such as high sensitivity, strong anti-interference capability, and real-time performance in cancer cell detection.

Description

Technical Field

[0001] This invention relates to the field of biomedical sensing technology, and in particular to a microwave detection system for cancer cells. Background Technology

[0002] The detection of circulating tumor cells (CTCs) is of paramount clinical significance in the early diagnosis, prognostic monitoring, and treatment assessment of cancer. CTCs are tumor cells that detach from the primary tumor and enter the peripheral blood, and are considered the "seeds" of cancer metastasis. However, CTCs are extremely rare in blood (only a few to a few dozen per milliliter), and the background contains a large number of red blood cells and white blood cells (WBCs), making the efficient isolation and accurate detection of CTCs from complex samples a significant challenge.

[0003] Existing CTCs detection technologies mainly suffer from the following bottlenecks:

[0004] Traditional microfluidic chips are typically used only as physical sorting tools. Sorted cells need to be extracted from the chip for offline detection using flow cytometry or fluorescence microscopy. This process increases the risk of sample contamination and cannot achieve real-time quantification. Secondly, existing integrated microwave biosensors mostly employ traditional interdigital electrode structures. The electric field distribution of these structures is relatively diffuse between the interdigital electrodes, leading to energy leakage due to edge effects. Furthermore, their effective inductance density per unit area is low, making it difficult to achieve high-quality factor resonance in a compact size, thus limiting the sensitivity for detecting dielectric differences in single cells. For single microwave resonance detection, it is susceptible to environmental temperature, substrate loss, and cell position fluctuations; single low-frequency capacitance detection struggles to distinguish internal physiological states of cells. There is a lack of complementary multi-physics information. There is a lack of the ability to locally "capture" or respond with high sensitivity to tiny single cells. In addition, to achieve a specific resonant frequency within a limited flow channel coverage area, traditional structures are often large in size, making it difficult to perfectly match the micrometer-scale dimensions of microfluidic channels. Therefore, there is an urgent need for a novel sensing structure that can provide high local field strength and a compact size.

[0005] Current microwave sensing data processing often employs linear fitting or simple regression analysis, which struggles to capture the complex nonlinear characteristics inherent in broadband S-parameters, particularly neglecting the sequence correlation of frequency-scanning signals in the frequency domain. Microwave biosensors, as an emerging label-free detection technology, offer advantages such as non-invasiveness, fast response, and no need for fluorescent labeling. They reflect cell concentration and physiological state by detecting changes in the dielectric properties of biological samples (the real and imaginary parts of the dielectric constant). However, existing microwave sensors mostly operate in a single mode (resonance only or capacitance only). When dealing with complex biological samples, this single parameter is easily affected by environmental noise, leading to insufficient detection accuracy and robustness.

[0006] Current microfluidic sorting chips typically exist as independent separation modules, requiring manual transfer of sorted samples to external instruments for analysis. This "offline" operation not only increases the risk of sample contamination and loss but also fails to achieve integrated automated detection from "sample in" to "result out." Existing microwave sensors mostly employ linear interdigitated electrodes.

[0007] In summary, existing CTC detection technologies require offline detection, which increases the risk of sample contamination. Furthermore, the microfluidic chip structures used are relatively large and lack high sensitivity to small single cells. The sizes of existing microfluidic chips and microwave sensors cannot be effectively matched. Summary of the Invention

[0008] Therefore, the technical problem to be solved by the present invention is to overcome the fact that the existing CTCs detection technology requires offline detection, which increases the risk of sample contamination, and the microfluidic chip structure used is large in size and lacks high sensitivity response capability to tiny single cells, and the size of the existing microfluidic chip and microwave sensor cannot be effectively adapted.

[0009] To address the aforementioned technical problems, the present invention provides a cell detection system, comprising:

[0010] Metal base;

[0011] A microwave sensor is located on the upper surface of a metal base and is fitted to the metal base.

[0012] A microfluidic chip, wherein the microfluidic chip is located on the upper surface of a microwave sensor and is attached to the microwave sensor;

[0013] A cancer cell concentration detection module is connected to a microwave sensor. When the cell solution flows on the microfluidic chip, it collects capacitive sensing data and microwave sensing data from the microwave sensor, and predicts the concentration of cancer cells in the cell solution based on the capacitive sensing data and microwave sensing data.

[0014] In one embodiment of the present invention, the microfluidic chip includes a main detection area, which includes a first detection channel and a second detection channel;

[0015] One end of the first detection channel is provided with a first large-diameter cell outlet, and one end of the second detection channel is provided with a second large-diameter cell outlet. The first and second large-diameter cell outlets are used to collect enriched circulating tumor cells.

[0016] In one embodiment of the present invention, the microfluidic chip includes a first helical channel and a second helical channel, wherein the first helical channel and the second helical channel adopt an Archimedean spiral structure and the number of turns is set to 4-6 turns.

[0017] The first spiral channel is provided with a first injection port, and the second spiral channel is provided with a second injection port. Both the first injection port and the second injection port are used to inject cell solution.

[0018] The first spiral flow channel is connected to the other end of the first detection channel via a first buffer channel, and the first spiral flow channel is also provided with a first small particle size outlet.

[0019] The second spiral flow channel is connected to the other end of the second detection channel via the second buffer channel, and the second spiral flow channel is also provided with a second small particle size outlet;

[0020] The first and second small particle size outlets are used to discharge leukocyte waste fluid.

[0021] Both the first and second buffer channels adopt a serpentine structure;

[0022] The cross-sections of the first and second spiral channels are both rectangular, with a width ranging from 190 to 210 μm and a height ranging from 90 to 110 μm.

[0023] In one embodiment of the present invention, both the first detection channel and the second detection channel adopt a serpentine structure;

[0024] The number of serpentine loops in the first detection channel ranges from 3 to 5.

[0025] The number of turns in the serpentine structure of the second detection channel ranges from 4 to 6.

[0026] The length of a single loop in the serpentine structure corresponding to the first detection channel is 32.81±10mm, and the width is 3±1mm;

[0027] The length of a single loop in the serpentine structure corresponding to the second detection channel is 32.81±10mm, and the width is 3±1mm.

[0028] In one embodiment of the present invention, the microwave sensor includes a first signal port, a second signal port, and a sensing area, wherein the sensing area is connected to the first signal port and the second signal port, respectively.

[0029] The sensing area includes a first metal finger strip group and a second metal finger strip group. The first metal finger strip group includes a plurality of serrated and interconnected first fingers, which are connected to a first signal port after being connected together. The second metal finger strip group includes a plurality of serrated and interconnected second fingers, which are connected to a second signal port after being connected together. The first and second signal ports are used to connect a vector network analyzer and an LCR meter.

[0030] The width of the first and second finger strips ranges from 0.22 to 0.26 mm.

[0031] In one embodiment of the present invention, each of the first and second finger strips includes multiple V-bending cycles, and the angle range of each V-bending cycle is 120 degrees to 150 degrees.

[0032] The first finger bar of the first metal finger bar group and the second finger bar of the second metal finger bar group are arranged alternately;

[0033] The gap between adjacent first and second finger strips ranges from 0.18 to 0.20 mm.

[0034] In one embodiment of the present invention, when the operating frequency of the sensing region of the microwave sensor is in a preset microwave band, the sensing region is equivalent to a resonator, and the equivalent circuit of the resonator includes:

[0035] The first equivalent resistance R1 is connected in sequence to one end of the first coupling capacitor C1, the first inductor L1, the second inductor L1, the second coupling capacitor C2, and the second equivalent resistance R2; the other end of the first equivalent resistance R1 is connected to the first ground parasitic capacitance C. S1 The connection between the first inductor L1 and the second inductor L1 is connected to the second ground parasitic capacitance C. S2 The other end of the second equivalent resistance R2 is connected to the third ground parasitic capacitance C. S3 The connection is made such that the first equivalent resistance R1 and the second equivalent resistance R2 are of equal magnitude.

[0036] In one embodiment of the present invention, when the concentration of the solution containing circulating tumor cells changes, the flow through the first and second detection channels will cause a change in the complex permittivity of the gap between the first and second finger strips in the sensing area, as expressed by the formula:

[0037]

[0038] in, The complex permittivity of the medium in the gap between the sensing regions is given by [insert value here]. is the real part of the complex permittivity of the dielectric, which characterizes the ability of a material to store electrical energy through displacement polarization in an electric field; The imaginary unit, This represents the imaginary part of the complex permittivity of the dielectric.

[0039] In one embodiment of the present invention, a concentration detection module is further included. This concentration detection module is used to extract data from both the microwave sensor and the microfluidic chip, and to predict cell concentration based on the extracted data. Specifically:

[0040] The vector network analyzer is connected to the microwave sensor, and the microwave sensing data collected by the microwave sensor is input into the first LSTM network. The microwave feature vector V is output through the first LSTM network. micro ;

[0041] Connect the LCR meter to the microwave sensor to collect capacitance sensing data from the microwave sensor and input it into the second LSTM network. The second LSTM network then outputs the capacitance feature vector V. cap Wherein, the microwave feature vector V micro The dimension and capacitance feature vector V cap The dimensions are the same;

[0042] Based on the attention mechanism, the microwave feature vector V micro Capacitance eigenvector V cap The microwave feature vector V is obtained by passing it through three independent linear fully connected layers. micro The corresponding first query vector Q1, first key vector K1, and first value vector V1, as well as the capacitance feature vector V cap The corresponding mappings are the second query vector Q2, the second key vector K2, and the second value vector V2;

[0043] The microwave feature vector V micro The dot product of the attention vectors is performed on the first query vector Q1 and the first key vector K1 to obtain Q1*K1. T ; the capacitance feature vector V cap The corresponding mapping is Q2*K2 obtained by performing a dot product operation on the attention vectors of the second query vector Q2 and the second key vector K2. T ;

[0044] Based on the dot product result Q1*K1 T and Q2*K2 T The attention score matrix is ​​constructed by performing dot product operations on the results Q1*K1. T and Q2*K2 T Divide by their respective scaling factors After scaling, we obtain two scaled feature matrices, where dk The dimension of the corresponding key vector is defined; the two scaled feature matrices are concatenated and fused to obtain a joint feature matrix; then the joint feature matrix is ​​input into a preset weight generation network for feature mapping, and the output is an attention score matrix containing dual-modal weight information;

[0045] After normalizing the attention score matrix using the Sigmoid function, the confidence scores α for the microwave mode and β for the capacitor mode are obtained.

[0046] The confidence level α of the microwave mode is multiplied element-wise with the first value vector V1 to obtain the microwave attention feature. The confidence level β of the capacitive mode is element-wise multiplied with the second value vector V2 to obtain the capacitive attention feature. ;

[0047] Subsequently, residual connections are introduced to connect the first value vector V1 with the microwave attention feature. The summation yields the microwave attention output vector V. micro_out The second value vector V2 is combined with the capacitive attention feature. The summation yields the capacitive attention output vector V. cap_out Finally, the microwave attention output vector V is... micro_out and the capacitance attention output vector V cap_out We perform weighted fusion of feature vectors to obtain the final fused feature F. fusion , is represented as:

[0048] ;

[0049] Wherein, W1 and W2 are the learnable fusion weight matrices corresponding to microwave mode and capacitor mode, respectively;

[0050] The fusion feature F fusion The input is a regression prediction module, which predicts the concentration of circulating tumor cells in the solution. The regression prediction module includes multiple fully connected layers connected in sequence, and the layers between adjacent fully connected layers include a ReLU activation function and a Dropout layer.

[0051] Compared with the prior art, the above-described technical solution of the present invention has the following advantages:

[0052] The present invention achieves efficient enrichment of CTCs through a unique helical microfluidic design, realizes dual-mode complementary detection of microwave and low-frequency capacitance through an innovative microwave resonator structure, and achieves high-precision data fusion processing through the MCAF-Net architecture. The system has significant advantages such as high sensitivity, strong anti-interference ability, label-free operation, and high degree of automation, and can effectively assist medical staff in the early screening and prognosis assessment of cancer.

[0053] The microwave sensor of this invention is used to achieve efficient label-free sorting of circulating tumor cells (CTCs) and white blood cells (WBCs). This invention can be connected to a vector network analyzer (VNA) or an LCR meter simultaneously, and can operate in both high-frequency resonant mode (approximately 7.3 GHz) and low-frequency capacitive mode (1 MHz) to detect changes in dielectric loss caused by circulating tumor cells (microwave sensing data) and membrane capacitance polarization effect (capacitive sensing data), respectively. Based on the collected microwave sensing data and capacitive sensing data, the concentration of cancer cells can be predicted. Attached Figure Description

[0054] To make the content of this invention easier to understand, the invention will be further described in detail below with reference to specific embodiments and accompanying drawings.

[0055] Figure 1 This is a schematic diagram of the microfluidic chip structure in an embodiment of the present invention;

[0056] Figure 2 This is a schematic diagram of the microwave sensor structure in an embodiment of the present invention;

[0057] Figure 3 This is a schematic diagram of the equivalent circuit of the microwave sensor in an embodiment of the present invention;

[0058] Figure 4 This is a schematic diagram of the MCAF-Net network in an embodiment of the present invention. Detailed Implementation

[0059] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, so that those skilled in the art can better understand and implement the present invention. However, the embodiments described are not intended to limit the present invention.

[0060] Example 1

[0061] Reference Figure 1 and 2 As shown, this invention relates to a microwave detection system for cancer cells, comprising:

[0062] Metal base;

[0063] A microwave sensor (made on glass, which acts as a substrate) is located on the upper surface of a metal base and is attached to the metal base.

[0064] A microfluidic chip (made of silicone) is located on the upper surface of a microwave sensor and is attached to the microwave sensor.

[0065] A cancer cell concentration detection module is connected to a microwave sensor. When the cell solution flows on the microfluidic chip, it collects capacitive sensing data and microwave sensing data from the microwave sensor, and predicts the concentration of cancer cells in the cell solution based on the capacitive sensing data and microwave sensing data.

[0066] The following is a detailed description of this embodiment:

[0067] like Figure 1 and Figure 2 As shown, the microwave sensor and microfluidic chip are mounted on a metal (aluminum) base. The aluminum base provides not only mechanical support but also serves as RF grounding and electromagnetic shielding. The substrate material for the microwave sensor is borosilicate glass (i.e., the microwave sensor is mounted on glass). Borosilicate glass features low dielectric loss (tanδ≈0.0059), high surface flatness, and good optical transparency, with dimensions of 51mm × 25mm and a thickness of 1mm. The borosilicate glass is crucial for microscope alignment of the microfluidic chip. The first signal port 2-1 and the second signal port 2-2 at both ends of the microwave sensor are led out via microstrip lines through SMA RF connectors and connected to an external vector network analyzer (VNA) and LCR meter via 50Ω impedance matching. The inlet and outlet of the microfluidic chip are connected via tubing to a microinjection pump, a leukocyte waste collection bottle, and a tumor cell collection bottle. The microinjection pump is used to store solutions containing circulating tumor cells and leukocytes, and it is connected to the first injection port 1-1 and the second injection port 1-2 of the microfluidic chip. The leukocyte waste collection bottle is connected to the first small-diameter outlet 5-1 and the second small-diameter outlet 5-2 of the microfluidic chip. The tumor cell collection bottle is connected to the first large-diameter cell outlet 3-1 and the second large-diameter cell outlet 3-2 of the microfluidic chip.

[0068] Furthermore, the microfluidic chip includes a main detection area 7, which includes a first detection channel 7-1 and a second detection channel 7-2. One end of the first detection channel 7-1 is provided with a first large-diameter cell outlet 3-1, and one end of the second detection channel 7-2 is provided with a second large-diameter cell outlet 3-2. The first large-diameter cell outlet 3-1 and the second large-diameter cell outlet 3-2 are used to collect enriched circulating tumor cells (CTCs).

[0069] Furthermore, the microfluidic chip includes a first helical channel 4-1 and a second helical channel 4-2. Both the first helical channel 4-1 and the second helical channel 4-2 adopt an Archimedean spiral structure, with each channel having 4-6 turns. The first helical channel 4-1 is provided with a first injection port 1-1, and the second helical channel 4-2 is provided with a second injection port 1-2. Both the first injection port 1-1 and the second injection port 1-2 are used to inject a solution containing circulating tumor cells and leukocytes. The other end of the first helical channel 4-1 is connected to the first detection channel 7-1 via a first buffer channel 6-1. 4-1 is also provided with a first small particle size outlet 5-1; the other end of the second spiral channel 4-2 and the second detection channel 7-2 are connected via a second buffer channel 6-2, and the second spiral channel 4-2 is also provided with a second small particle size outlet 5-2; the first small particle size outlet 5-1 and the second small particle size outlet 5-2 are used to discharge leukocyte (WBC) waste liquid; the first buffer channel 6-1 and the second buffer channel 6-2 both adopt a serpentine structure; the cross-section of the first spiral channel 4-1 and the second spiral channel 4-2 are both rectangular, with a width ranging from 190 to 210 μm and a height of 90 to 110 μm.

[0070] Furthermore, both the first detection channel 7-1 and the second detection channel 7-2 adopt a serpentine structure; the number of serpentine coils in the first detection channel 7-1 ranges from 3 to 5; the number of serpentine coils in the second detection channel 7-2 ranges from 4 to 6; the first detection channel 7-1 and the second detection channel 7-2 are located in the same plane on the vertical line, and their lengths are aligned on the horizontal plane, together forming a planar detection area. The length of a single coil of the serpentine structure corresponding to the first detection channel 7-1 is 32.81±10mm, and the width is 3±1mm; the length of a single coil of the serpentine structure corresponding to the second detection channel 7-2 is 32.81±10mm, and the width is 3±1mm. The purpose of constructing two detection channels in this embodiment is to enable simultaneous injection of cell solution into the detection channels through the first injection port 1-1 and the second injection port 1-2, ensuring uniform flow of the cell solution and making the subsequent prediction results more accurate.

[0071] In summary, the microfluidic chip of this embodiment can meet the requirements of high-throughput and high-purity sorting. The microfluidic chip, covering the microwave sensor, is made of polydimethylsiloxane (PDMS). The structure includes two sets of helical structures: a first helical channel 4-1 and a second helical channel 4-2. Cell solution is injected into the first injection port 1-1 and the second injection port 1-2. The first small-diameter outlet 5-1 and the second small-diameter outlet 5-2 are connected to the outer wall of the helical channels, used to discharge leukocyte (WBC) waste fluid. Unremoved circulating tumor cells (CTCs) solution flows through the inner wall of the helical channels, via the first buffer channel 6-1 into the first detection channel 7-1, and via the second buffer channel 6-2 into the second detection channel 7-2. Finally, the enriched circulating tumor cells (CTCs) are collected at the first large-diameter cell outlet 3-1 and the second large-diameter cell outlet 3-2, with the pathway directly aligned with the interdigitated sensitive area (i.e., sensing area 8) of the microwave sensor. The main detection zone 7 is a high aspect ratio serpentine channel. Its structure further eliminates minor flow disturbances that may arise from the Archimedean spiral flow channel, ensuring the stability of the fluid state during microwave sensor measurement and thus obtaining a high signal-to-noise ratio detection signal. The main detection zone 7 ensures that the fluid output from the cell separation zone is effectively collected and stabilized, allowing the lower microwave sensor to perform reliable terminal detection. The first buffer channel 6-1 and the second buffer channel 6-2 are buffer channels distributed at the inlet of the main detection zone 7. Through their long, meandering paths, they dampen and buffer fluid pressure fluctuations, making the fluid flow rate into the subsequent key functional area (main detection zone 7) more stable and uniform. More importantly, it ensures that the fluid flowing out from the Archimedean spiral (i.e., the first spiral flow channel 4-1 and the second spiral flow channel 4-2) flows into the main detection zone 7 in a stable and predictable state, laying the foundation for efficient and controllable diffusion and detection within the channel network. In other embodiments, it is necessary to precisely control the time from sample injection to entry into the main detection zone 7. The length and volume of the first buffer channel 6-1 and the second buffer channel 6-2 can serve as a precise timing module, achieving time delays from milliseconds to minutes.

[0072] Furthermore, the microwave sensor includes a first signal port 2-1, a second signal port 2-2, and a sensing region 8. The sensing region 8 is connected to the first signal port 2-1 and the second signal port 2-2, respectively. The sensing region 8 includes a first metal finger strip group and a second metal finger strip group. The first metal finger strip group includes several serrated and interconnected first fingers 8-1, which are connected together and then connected to the first signal port 2-1 via a microstrip line. The second metal finger strip group includes several serrated and interconnected second fingers 8-2, which are connected together and then connected to the second signal port 2-2 via a microstrip line. The first signal port 2-1 and the second signal port 2-2 are used to connect a vector network analyzer and an LCR meter. The width of the first finger strip 8-1 and the second finger strip 8-2 ranges from 0.22 to 0.26 mm.

[0073] Furthermore, each first finger bar 8-1 and second finger bar 8-2 includes multiple V-bending cycles, with the angle range of each V-bending cycle being 120 degrees to 150 degrees; the first finger bar 8-1 of the first metal finger bar group and the second finger bar 8-2 of the second metal finger bar group are staggered; the gap between adjacent first finger bars 8-1 and second finger bars 8-2 is 0.18 to 0.20 mm.

[0074] In this embodiment, the main detection area 7 (corresponding to the first detection channel 7-1 and the second detection channel 7-2) and the sensing area 8 (the first finger strip 8-1 and the second finger strip 8-2) are in direct contact.

[0075] In summary, referring to Figure 1 As shown, the cell sorting function of this invention is based on the principle of inertial microfluidics. The first helical channel 4-1 and the second helical channel 4-2 are specifically designed with an Archimedean spiral structure, consisting of several turns (5 turns). This multi-turn design increases the channel length, ensuring that particles have sufficient residence time in the flow field to reach their equilibrium position. The channel cross-section is rectangular, with a width w = 200 μm and a height h = 100 μm. The radius of curvature of the Archimedean spiral structure starts from R at the first turn at the inlet. min =3500μm gradually increases to the fifth ring R at the exit. max =7100μm. When fluid flows through a curved channel at a specific velocity (optimized to 400-1000μL / min in this embodiment), two main hydrodynamic effects are generated within the flow field, acting together on the suspended particles. Due to the centrifugal force of the fluid in the curved pipe, the fluid with a faster velocity at the center of the channel is squeezed against the outer walls of the first spiral channel 4-1 and the second spiral channel 4-2, while the fluid with a slower velocity at the upper and lower walls flows back towards the inner wall circumferentially. This forms a pair of symmetrical vortices rotating in opposite directions on the cross-sections of the first spiral channel 4-1 and the second spiral channel 4-2, called Dean vortices. Dean vortices exert a lateral drag force on the particles, which can be derived from Stokes' law:

[0076]

[0077] in, and These represent the maximum flow velocity and particle diameter, respectively. It is a dimensionless Dean number. Indicates the radius of curvature of the channel. Let F be the fluid density. When a particle moves in a Poiseuille flow field, it is subjected to shear lift due to the velocity shear gradient and wall-induced lift due to wall repulsion. The two are collectively referred to as inertial lift F. L The formula is:

[0078]

[0079] in, The lift coefficient, For maximum flow rate, This is the hydraulic diameter. Because F... L The fourth power of the particle diameter ( ) is directly proportional to FD and the particle diameter ( The effect of centrifugal force is directly proportional to the size of the particles. Large particles (such as 20 μm CTCs) are dominated by inertial lift, which allows them to migrate and focus more quickly on the inner walls of the first helical channel 4-1 and the second helical channel 4-2. This means that the solution containing circulating tumor cells will enter the main detection area 7 through the first buffer channel 6-1 and the second buffer channel 6-2. Small particles (such as 10-15 μm WBCs) are more affected by Dean's drag force, are drawn into the Dean's vortex, and are pushed towards the outer walls of the first helical channel 4-1 and the second helical channel 4-2. This means that leukocyte waste will be discharged from the first small particle outlet 5-1 and the second small particle outlet 5-2. In short, due to the different centrifugal forces, the two different types of particles will exit from different outlets in the first helical channel 4-1 and the second helical channel 4-2.

[0080] like Figure 2As shown, the core sensing element in this embodiment is a microwave sensor (i.e., a sawtooth microwave interdigital resonator). Unlike traditional comb-shaped interdigital electrodes, the microwave sensor is fabricated on a borosilicate glass substrate. The first finger strip 8-1 and the second finger strip 8-2 are composed of a 500nm thick Ti / Cu metal layer. It includes two signal ports (first signal port 2-1 and second signal port 2-2) and a sensing region 8 in the middle. The sensing region 8 is composed of two sets of metal fingers (first finger strip 8-1 and second finger strip 8-2). In this embodiment, the first finger strip 8-1 and the second finger strip 8-2 are not designed as traditional straight lines, but are designed as continuous sawtooth shapes. The first finger strip 8-1 and the second finger strip 8-2 extend along the direction of fluid flow, and each finger strip contains multiple V-shaped bending cycles. A uniform gap of millimeters is always maintained between the two sets of fingers. At each sharp corner of the sawtooth, the charge density increases sharply due to the extremely small radius of curvature. This forms a series of high-intensity "electric field hot spots" on the flow channel cross-section. When cells flow through these hotspots, even minute differences in dielectric properties can induce significant signal disturbances. Current flows along a serrated path, with a physical path length much greater than a straight distance, significantly increasing the equivalent inductance. This allows the microwave sensor to operate at lower microwave frequencies (e.g., 7.3 GHz) while maintaining a small size, avoiding excessive attenuation of high-frequency signals in aquatic environments. Simultaneously, the serrated structure maximizes the effective sensing area, increasing the total length of the interdigital slits, thereby improving the fundamental capacitance in low-frequency modes and enhancing the signal-to-noise ratio.

[0081] Furthermore, when the operating frequency of the sensing region 8 of the microwave sensor in this embodiment is within a preset microwave band (approximately 7.3 GHz), the sensing region 8 is equivalent to a resonator. The equivalent circuit of the resonator includes: a first equivalent resistor R1, one end of which is sequentially connected to one end of a first coupling capacitor C1, a first inductor L1, a second inductor L1, a second coupling capacitor C2, and a second equivalent resistor R2; the other end of the first equivalent resistor R1 is connected to a first ground parasitic capacitor C... S1 The connection between the first inductor L1 and the second inductor L1 is connected to the second ground parasitic capacitance C. S2 The other end of the second equivalent resistor R2 is connected to the third ground parasitic capacitance C. S3 The connection is made such that the first equivalent resistance R1 and the second equivalent resistance R2 are of equal magnitude.

[0082] The formula for the resonant frequency of the equivalent circuit is: The entire circuit forms an LC oscillation, and the resonant frequency is:

[0083]

[0084] Based on formula (3) and Figure 3In the microwave band, the slender first finger 8-1 and second finger 8-2 exhibit inductive characteristics. Resistances R1 and R2 remain constant and are the equivalent resistances of the two microstrip lines, primarily due to the skin effect loss of the metallic conductor at high frequencies. (The enhanced equivalent inductance generated by the metal wires of the first finger bar 8-1 and the second finger bar 8-2 remains unchanged, and the capacitance remains unchanged.) Equivalent to Figure 3 Equivalent capacitances C1 and C2, C S1 / C S2 / C S3 The electric field coupling between the metal structure and the grounding layer of the metal base is negligible. The change in the cell dielectric constant (corresponding to the following formula (4)) can change the values ​​of capacitors C1 (corresponding to the sensing area 8 part contacted by the first detection channel 7-1, equivalent to the left part of the sensing area 8) and C2 (corresponding to the sensing area 8 part contacted by the second detection channel 7-2, equivalent to the right part of the sensing area 8). The change in capacitors C1 and C2 thus changes the resonant frequency. Size, get It can also construct the concentration relationship of cancer cells (a linear equation in one variable).

[0085] When the concentration of the solution containing circulating tumor cells changes, the flow through the first detection channel 7-1 and the second detection channel 7-2 will cause a change in the complex permittivity of the gap between the first finger strip 8-1 and the second finger strip 8-2 in the sensing region 8, as shown in the formula:

[0086]

[0087] in, Let be the complex permittivity of the medium in the 8-gap sensing region. is the real part of the complex permittivity of the dielectric, characterizing the material's ability to store electrical energy through displacement polarization in an electric field. The imaginary unit, This represents the imaginary part of the complex permittivity of the dielectric material, specifically the dielectric loss. It arises from polarization loss and reflects the energy dissipation characteristics of the material.

[0088] for (Imaginary part / dielectric loss): At high frequencies (7.3 GHz), data is collected using a vector network analyzer. Electromagnetic waves can penetrate the cell membrane and enter the cytoplasm. The cytoplasm contains a large number of ions, which generate conduction current under the action of an alternating electric field, leading to energy dissipation. The higher the cell concentration, the greater the loss, and the increase in the resistive component in the equivalent circuit, resulting in a decrease in the quality factor. When the operating frequency is low (1 MHz), data is collected using an LCR meter. The inductive effect and skin effect are negligible, and the microwave sensor degenerates into a pure capacitor. At low frequencies, opposite charges accumulate on adjacent interdigitates, and the potential difference causes the electric field lines to arc across the gap, penetrating the liquid in the flow channel and forming an edge electric field. Sensing mechanism (interfacial polarization detection): At a frequency of 1 MHz, it is at the low-frequency end of the β-dispersion region. At this time, the intact cell membrane acts as an insulating shell. According to the Maxwell-Wagner interfacial polarization theory, the insulating cell membrane hinders the movement of ions, causing charges to accumulate at the cell membrane interface, producing strong interfacial polarization. This polarization effect causes the real part of the effective dielectric constant of the cell suspension to be reduced. It increases significantly with increasing cell volume fraction.

[0089] This embodiment also includes a cancer cell concentration detection module. This module collects capacitive sensing data and microwave sensing data from a microwave sensor when the cell solution flows on the microfluidic chip, and predicts the concentration of cancer cells in the cell solution based on the capacitive and microwave sensing data. To fully utilize the complementarity of the dual-modal data, please refer to [link to relevant documentation]. Figure 4 This invention proposes a customized deep learning network architecture—MCAF-Net. The overall architecture of MCAF-Net adopts a typical fusion topology structure, mainly consisting of three core networks: a two-stream feature extraction sub-model, a multimodal attention transformation model, and a regression variable model.

[0090] The dual-stream feature extraction sub-model is as follows:

[0091] Connect the vector network analyzer to the microwave sensor (first signal port 2-1, second signal port 2-2), collect microwave sensing data about the microwave sensor, and input the data into the first LSTM network. The first LSTM network then outputs the microwave feature vector V. micro Connect the LCR meter to the microwave sensor (first signal port 2-1, second signal port 2-2), collect capacitance sensing data about the microwave sensor, and input the data into the second LSTM network. The second LSTM network then outputs the capacitance feature vector V. cap Among them, the microwave feature vector V micro The dimension and capacitance feature vector V capThe dimensions are the same. It should be noted that the first signal port 2-1 and the second signal port 2-2 can be connected to both the vector network analyzer and the LCR meter simultaneously by leading out multiple connecting lines.

[0092] The multimodal attention transfer model is as follows:

[0093] This embodiment is based on the Query-Key-Value (QKV) attention mechanism, which integrates the microwave feature vector V. micro Capacitance eigenvector V cap The microwave feature vector V is obtained by passing it through three independent linear fully connected layers. micro The corresponding first query vector Q1, first key vector K1, and first value vector V1, as well as the capacitance feature vector V cap The corresponding second query vector Q2, second key vector K2, and second value vector V2.

[0094] microwave feature vector V micro The dot product of the attention vectors is performed on the first query vector Q1 and the first key vector K1 to obtain Q1*K1. T ; the capacitance feature vector V cap The dot product of the attention vectors is performed on the corresponding second query vector Q2 and second key vector K2 to obtain Q2*K2. T .

[0095] Based on the dot product result Q1*K1 T and Q2*K2 T The attention score matrix is ​​constructed by performing dot product operations on the results Q1*K1. T and Q2*K2 T Divide by their respective scaling factors (Actually includes) and ) is scaled to obtain two scaled feature matrices, where d k The dimension of the corresponding key vector is used; the two scaled feature matrices are concatenated and fused to obtain a joint feature matrix; then, the joint feature matrix is ​​input into a preset weight generation network for feature mapping, and the output is an attention score matrix containing bimodal weight information. The weight generation network in this embodiment adopts a multilayer perceptron structure, including an input layer, a hidden layer, and an output layer connected in sequence. The input layer receives input data, the hidden layer uses the ReLU nonlinear activation function to extract the nonlinear coupling relationship between microwave and capacitive bimodal features, and the output layer outputs an attention score matrix containing bimodal weight information. After normalizing the attention score matrix using the Sigmoid function, the confidence α of the microwave mode and the confidence β of the capacitive mode are obtained.

[0096] The microwave attention features are obtained by element-wise multiplication of the confidence α of the microwave mode with the first value vector V1. Similarly, by element-wise multiplying the confidence β of the capacitive mode with the second value vector V2, the capacitive attention feature is obtained. Subsequently, residual connections are introduced to connect the first value vector V1 with the microwave attention feature. The summation yields the microwave attention output vector V. micro_out The second value vector V2 is combined with the capacitance attention feature. The summation yields the capacitive attention output vector V. cap_out , is represented as:

[0097]

[0098]

[0099] Residual connections ensure smooth backpropagation of gradients, preventing network degradation. Physically, this means the network is corrected based on the original measurements, rather than completely reconstructed, guaranteeing the physical interpretability of the predictions. Finally, the microwave attention output vector V is... micro_out and the capacitance attention output vector V cap_out We perform weighted fusion of feature vectors to obtain the final fused feature F. fusion , is represented as:

[0100]

[0101] Where W1 and W2 are the learnable fusion weight matrices corresponding to the microwave mode and the capacitor mode, respectively.

[0102] The regression prediction module is as follows:

[0103] F fusion feature F fusion Input the regression prediction module to obtain the concentration value of cells in the solution. The regression prediction module includes multiple fully connected layers, and the layers between adjacent fully connected layers include a ReLU activation function and a Dropout layer (to prevent overfitting).

[0104] In this embodiment, mean squared error (MSE) or mean absolute error (MAE) can be used as the training target to drive the update of the entire network parameters.

[0105] The MCAF-Net architecture proposed in this invention has significant academic and engineering value compared to traditional signal processing methods. By using LSTM instead of a one-dimensional convolutional neural network, it mathematically aligns with the physical properties of microwave S-parameters as a "frequency domain sequence," enabling the capture of the long-range dependence between the fine structure of resonance peaks and the broadband background. Simultaneously, the gating mechanism of LSTM inherently possesses filter functionality, effectively eliminating random electromagnetic interference in the experimental environment and improving the signal-to-noise ratio. Through the attention transformation module, the network can dynamically learn the nonlinear relationship between microwave modes (reflecting cytoplasmic loss) and capacitance modes (reflecting cell membrane polarization), achieving a detection effect of "1+1>2." Finally, the introduction of residual connections ensures that while increasing the number of network layers to improve feature extraction capabilities, the physical characteristics of the original signal are not lost, making model training more stable and efficient.

[0106] In summary, this invention achieves efficient enrichment of CTCs through a unique helical microfluidic design, realizes dual-mode complementary detection of microwave and low-frequency capacitance through an innovative microwave resonator structure, and achieves high-precision data fusion processing through the MCAF-Net architecture. This system has significant advantages such as high sensitivity, strong anti-interference capability, label-free operation, and high degree of automation, effectively assisting medical personnel in early cancer screening and prognostic assessment.

[0107] Obviously, the above embodiments are merely illustrative examples for clear explanation and are not intended to limit the implementation. Those skilled in the art will recognize that other variations or modifications can be made based on the above description. It is neither necessary nor possible to exhaustively list all possible implementations here. However, obvious variations or modifications derived therefrom are still within the scope of protection of this invention.

Claims

1. A microwave detection system for cancer cells, characterized in that, include: Metal base; A microwave sensor is located on the upper surface of a metal base and is fitted to the metal base. A microfluidic chip, wherein the microfluidic chip is located on the upper surface of a microwave sensor and is attached to the microwave sensor; A cancer cell concentration detection module is connected to a microwave sensor. When the cell solution flows on the microfluidic chip, it collects capacitive sensing data and microwave sensing data from the microwave sensor, and predicts the concentration of cancer cells in the cell solution based on the capacitive sensing data and microwave sensing data.

2. The microwave detection system for cancer cells according to claim 1, characterized in that: The microfluidic chip includes a main detection area (7), which includes a first detection channel (7-1) and a second detection channel (7-2). One end of the first detection channel (7-1) is provided with a first large-diameter cell outlet (3-1), and one end of the second detection channel (7-2) is provided with a second large-diameter cell outlet (3-2). The first large-diameter cell outlet (3-1) and the second large-diameter cell outlet (3-2) are used to collect enriched circulating tumor cells.

3. The microwave detection system for cancer cells according to claim 2, characterized in that: The microfluidic chip includes a first helical channel (4-1) and a second helical channel (4-2). The first helical channel (4-1) and the second helical channel (4-2) adopt an Archimedean spiral structure, and the number of turns is set to 4-6 turns. The first spiral channel (4-1) is provided with a first injection port (1-1), and the second spiral channel (4-2) is provided with a second injection port (1-2). Both the first injection port (1-1) and the second injection port (1-2) are used to inject a solution containing circulating tumor cells and white blood cells. The first spiral channel (4-1) is connected to the other end of the first detection channel (7-1) via the first buffer channel (6-1), and the first spiral channel (4-1) is also provided with a first small particle size outlet (5-1). The second spiral flow channel (4-2) is connected to the other end of the second detection channel (7-2) via the second buffer channel (6-2), and the second spiral flow channel (4-2) is also provided with a second small particle size outlet (5-2). The first small particle size outlet (5-1) and the second small particle size outlet (5-2) are used to discharge leukocyte waste fluid.

4. The microwave detection system for cancer cells according to claim 2, characterized in that: Both the first detection channel (7-1) and the second detection channel (7-2) adopt a serpentine structure; The number of serpentine loops in the first detection channel (7-1) ranges from 3 to 5. The number of turns in the serpentine structure of the second detection channel (7-2) ranges from 4 to 6. The first detection channel (7-1) and the second detection channel (7-2) are located in the same plane and are aligned in length on the horizontal plane, together forming a planar detection area; The length of a single loop of the serpentine structure corresponding to the first detection channel (7-1) is 32.81±10mm, and the width is 3±1mm; The length of a single loop of the serpentine structure corresponding to the second detection channel (7-2) is 32.81±10mm, and the width is 3±1mm.

5. The microwave detection system for cancer cells according to claim 3, characterized in that: Both the first buffer channel (6-1) and the second buffer channel (6-2) adopt a serpentine structure; The cross-sections of the first spiral channel (4-1) and the second spiral channel (4-2) are both rectangular, with a width ranging from 190 to 210 μm and a height ranging from 90 to 110 μm.

6. The microwave detection system for cancer cells according to claim 1, characterized in that: The microwave sensor includes a first signal port (2-1), a second signal port (2-2), and a sensing area (8). The sensing area (8) is connected to the first signal port (2-1) and the second signal port (2-2), respectively. The sensing area (8) includes a first metal finger strip group and a second metal finger strip group. The first metal finger strip group includes several serrated and interconnected first fingers (8-1), which are connected to a first signal port (2-1). The second metal finger strip group includes several serrated and interconnected second fingers (8-2), which are connected to a second signal port (2-2). The first signal port (2-1) and the second signal port (2-2) are used to connect a vector network analyzer and an LCR meter. The width of the first finger strip (8-1) and the second finger strip (8-2) ranges from 0.22 to 0.26 mm.

7. The microwave detection system for cancer cells according to claim 6, characterized in that: Each of the first finger strip (8-1) and the second finger strip (8-2) includes multiple V-bending cycles, and the angle range of each V-bending cycle is 120 degrees to 150 degrees; The first finger bar (8-1) of the first metal finger bar group and the second finger bar (8-2) of the second metal finger bar group are staggered; The gap between adjacent first finger strips (8-1) and second finger strips (8-2) ranges from 0.18 to 0.20 mm.

8. The microwave detection system for cancer cells according to claim 1, characterized in that: When the operating frequency of the sensing region (8) of the microwave sensor is in a preset microwave band, the sensing region (8) is equivalent to a resonator, and the equivalent circuit of the resonator includes: The first equivalent resistance R1 is connected in sequence to one end of the first coupling capacitor C1, the first inductor L1, the second inductor L1, the second coupling capacitor C2, and the second equivalent resistance R2; the other end of the first equivalent resistance R1 is connected to the first ground parasitic capacitance C. S1 The connection between the first inductor L1 and the second inductor L1 is connected to the second ground parasitic capacitance C. S2 The other end of the second equivalent resistance R2 is connected to the third ground parasitic capacitance C. S3 The connection is made such that the first equivalent resistance R1 and the second equivalent resistance R2 are of equal magnitude.

9. The microwave detection system for cancer cells according to claim 1, characterized in that: When the concentration of the solution containing circulating tumor cells changes, the flow through the first detection channel (7-1) and the second detection channel (7-2) will cause a change in the complex permittivity of the gap between the first finger strip (8-1) and the second finger strip (8-2) of the sensing region (8), as shown in the formula: ； in, The complex permittivity of the medium in the gap between sensing regions (8) is given by: is the real part of the complex permittivity of the dielectric, which characterizes the ability of a material to store electrical energy through displacement polarization in an electric field; The imaginary unit, This represents the imaginary part of the complex permittivity of the dielectric.

10. The microwave detection system for cancer cells according to claim 1, characterized in that: The cancer cell concentration detection module is used to collect capacitive sensing data and microwave sensing data from a microwave sensor when the cell solution flows on the microfluidic chip, and to predict the concentration of cancer cells in the cell solution based on the capacitive sensing data and microwave sensing data, specifically: The vector network analyzer is connected to the microwave sensor, and the microwave sensing data collected by the microwave sensor is input into the first LSTM network. The microwave feature vector V is output through the first LSTM network. micro ; Connect the LCR meter to the microwave sensor to collect capacitance sensing data from the microwave sensor and input it into the second LSTM network. The second LSTM network then outputs the capacitance feature vector V. cap Wherein, the microwave feature vector V micro The dimension and capacitance feature vector V cap The dimensions are the same; Based on the attention mechanism, the microwave feature vector V micro Capacitance eigenvector V cap The microwave feature vector V is obtained by passing it through three independent linear fully connected layers. micro The corresponding first query vector Q1, first key vector K1, and first value vector V1, as well as the capacitance feature vector V cap The corresponding mappings are the second query vector Q2, the second key vector K2, and the second value vector V2; The microwave feature vector V micro The dot product of the attention vectors is performed on the first query vector Q1 and the first key vector K1 to obtain Q1*K1. T ; the capacitance feature vector V cap The corresponding mapping is Q2*K2 obtained by performing a dot product operation on the attention vectors of the second query vector Q2 and the second key vector K2. T ; Based on the dot product result Q1*K1 T and Q2*K2 T The attention score matrix is ​​constructed by performing dot product operations on the results Q1*K1. T and Q2*K2 T Divide by their respective scaling factors After scaling, we obtain two scaled feature matrices, where d k The dimension of the corresponding key vector is defined; the two scaled feature matrices are concatenated and fused to obtain a joint feature matrix; then the joint feature matrix is ​​input into a preset weight generation network for feature mapping, and the output is an attention score matrix containing dual-modal weight information; After normalizing the attention score matrix using the Sigmoid function, the confidence scores α for the microwave mode and β for the capacitor mode are obtained. The confidence level α of the microwave mode is multiplied element-wise with the first value vector V1 to obtain the microwave attention feature. The confidence level β of the capacitive mode is element-wise multiplied with the second value vector V2 to obtain the capacitive attention feature. ; Subsequently, residual connections are introduced to connect the first value vector V1 with the microwave attention feature. The summation yields the microwave attention output vector V. micro_out The second value vector V2 is combined with the capacitive attention feature. The summation yields the capacitive attention output vector V. cap_out Finally, the microwave attention output vector V is... micro_out and the capacitance attention output vector V cap_out We perform weighted fusion of feature vectors to obtain the final fused feature F. fusion , is represented as: ； Wherein, W1 and W2 are the learnable fusion weight matrices corresponding to microwave mode and capacitor mode, respectively; The fusion feature F fusion The input is a regression prediction module, which predicts the concentration of circulating tumor cells in the solution. The regression prediction module includes multiple fully connected layers connected in sequence, and the layers between adjacent fully connected layers include a ReLU activation function and a Dropout layer.

Images