Steam ablation multi-mode curative effect evaluation device

Abstract

The invention discloses a steam ablation multi-mode curative effect evaluation device, and belongs to the field of biomedical engineering. The near-infrared spectrum detection probe integrated with the steam ablation needle and the fluorescent probe are utilized, near-infrared optical parameters representing the thermal damage degree of the tissue and fluorescent signals representing the ablation range can be obtained in real time, and the thermal damage degree of the tissue can be determined through the synchronously collected signals, the specific result of the biological effect after ablation and the machine learning algorithm. And carrying out joint training on the multimode data, analyzing the relationship between the change of each parameter of the tissue in the thermal ablation process and the biological effect, and carrying out real-time monitoring and curative effect evaluation on the ablation range of steam thermal ablation. According to the invention, the near infrared spectrum and fluorescence imaging can be used for effective evaluation of the steam ablation curative effect, the treatment plan adjustment of the whole treatment process is facilitated, and the reliability and convenience of the treatment process are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of biomedical engineering, and particularly relates to a steam ablation multi-modal curative effect evaluation device. BACKGROUND

[0002] Steam thermal ablation technology is a new technology that achieves curative effect by applying high-temperature water vapor to a target region of a patient. Compared with other ablation technologies, steam ablation has the advantages of low cost, small side effects, and compatibility with drugs.

[0003] Because the diffusion range and action region of steam in tissue cannot be clearly displayed, the current steam thermal ablation equipment cannot monitor the ablation process in real time, lacks an accurate and reliable ablation effect evaluation method, and is difficult to accurately judge the necrosis degree and boundary of the tissue after ablation. The commonly used thermal ablation curative effect evaluation method is mainly image examination (such as CT, MRI, ultrasound, etc.), which can directly display the changes in tissue morphology and size, but cannot reflect the changes in cell activity and function inside the tissue in real time. Because the state of the tissue changes constantly during steam ablation, real-time evaluation is a key link to ensure the curative effect and safety of steam ablation. SUMMARY

[0004] The present application provides a steam ablation multi-modal curative effect evaluation device to realize the effective evaluation of steam ablation curative effect by near-infrared spectroscopy and fluorescence imaging, facilitate the treatment plan adjustment of the entire treatment process, and improve the reliability and convenience of the treatment process.

[0005] The present application provides a steam ablation multi-modal curative effect evaluation device, comprising:

[0006] a near-infrared spectroscopy acquisition system for acquiring near-infrared spectroscopy data during steam ablation;

[0007] a fluorescence imaging system for acquiring fluorescence signals during steam ablation;

[0008] a processing module for pre-processing the near-infrared spectroscopy data and the fluorescence signals, performing cross-modal feature fusion on the pre-processed near-infrared spectroscopy data and the fluorescence signals to obtain fusion features, training a multi-task learning network by using the fusion features to obtain an evaluation model, and evaluating the effect of steam ablation by using the trained evaluation model.

[0009] Optionally, in an embodiment of the present application, the near-infrared spectroscopy acquisition system comprises a light source, a fiber probe, and a spectrometer, the fiber probe and a temperature measurement probe are integrated on one ablation needle to form a comprehensive functional needle combining temperature measurement and near-infrared functions.

[0010] The fluorescence imaging system comprises an EMCCD fluorescence camera, a laser, and a laser light source probe.

[0011] Optionally, in one embodiment of the present invention, the fiber optic probe includes a spectral fiber and a light-guiding fiber, which are independently packaged and integrated into the same probe. The light generated by the halogen light source is transmitted to the tissue under test through the light-guiding fiber of the fiber optic probe, and after being scattered by the tissue under test, it is transmitted to the fiber optic spectrometer by the spectral fiber.

[0012] Optionally, in one embodiment of the present invention, the near-infrared spectral acquisition system is specifically used for:

[0013] The ablation site is illuminated by a light source, and a fiber optic probe is inserted into the tissue for testing. One fiber in the probe transmits the light from the light source to the tissue to be measured, while the spectrometer receives the light reflected and scattered by the tissue from the other fiber. The computer performs spectral acquisition and obtains near-infrared spectral data after processing the spectral data.

[0014] Optionally, in one embodiment of the present invention, the fluorescence imaging system is specifically used for:

[0015] Tissue ablation is performed using a vaporized aqueous solution of fluorescent dye as a heat source. A laser emits light of a specific wavelength to excite the fluorescent sample. After the fluorescent molecules absorb the excitation light, they emit a fluorescent signal. An EMCCD fluorescence camera detects the fluorescent signal, converts the light signal into an electrical signal, processes it to generate a fluorescence image, displays the fluorescence intensity distribution, and transmits it to a host computer.

[0016] Optionally, in one embodiment of the present invention, the fluorescence imaging system employs a fluorescence labeling strategy, dissolving a high-temperature stable fluorescent dye in vapor to form an aerosol, which diffuses with the vapor to label the tissue area. An EMCCD fluorescence camera captures the fluorescence signal, dynamically tracks the vapor diffusion boundary, and displays the vapor penetration range in real time.

[0017] Optionally, in one embodiment of the present invention, the vapor flow rate is quantified by fluorescence intensity gradient to guide dose control, the fluorescence diffusion radius is used to assess the vapor diffusion range and determine whether the target area has been reached, and the fluorescence intensity reflects the vapor concentration and tissue absorption, indirectly assessing the thermal field distribution.

[0018] Optionally, in one embodiment of the present invention, the near-infrared spectral data includes temperature parameters and fNIRs parameters. The processing module preprocesses the near-infrared spectral data and fluorescence signals, including: in time series processing, extracting temperature / near-infrared data at a preset step size, extracting the mean, standard deviation, trend gradient, and the first 5 FFT main frequency components, and eliminating inter-device differences through Z-score standardization; in fluorescence signal processing, extracting ablation zone morphological features through U-Net pre-segmentation, and using cubic spline fitting to describe the ablation boundary.

[0019] Optionally, in one embodiment of the present invention, cross-modal feature fusion is performed on the preprocessed near-infrared spectral data and fluorescence signal to obtain fused features, including: extracting temporal features using 1D CNN and Transformer encoder to capture temperature change patterns; using the ViT model for image feature extraction; extracting tissue heterogeneity features through multi-layer Transformer encoding; and using a two-layer cascaded cross-attention mechanism for cross-modal fusion with dynamically adjusted gating weights, wherein the temporal proportion is 60% to 80%.

[0020] Optionally, in one embodiment of the present invention, training a multi-task learning network using fused features to obtain an evaluation model includes:

[0021] The main task of the multi-task learning network is cell death rate, and the auxiliary task is self-supervised learning. In the loss function design, the main task adopts MSE+L1 regularization, and the auxiliary task uses improved Dice Loss. In the adaptive weighting, the gradient is automatically balanced based on Kendall uncertainty theory. In the prediction head structure, the main task has a 128-dimensional bottleneck layer + Swish activation, and the auxiliary task has 3 layers of transposed convolution to restore voxel spatial resolution.

[0022] This invention relates to a multimodal efficacy evaluation device for steam ablation. In steam ablation treatment, temperature directly reflects the degree of tissue damage, while fluorescence imaging clearly observes the boundary between tumor tissue and surrounding normal tissue, allowing for timely assessment of the treatment range and effect. This invention utilizes a near-infrared spectroscopy probe integrated with the steam ablation needle and a fluorescence probe to acquire near-infrared optical parameters characterizing the degree of tissue thermal damage and fluorescence signals characterizing the ablation range in real time. By simultaneously acquiring near-infrared signal data, temperature data, and fluorescence signals, combined with the specific results of the biological effects after ablation, and employing machine learning algorithms to jointly train the multimodal data, the relationship between changes in various parameters and biological effects during the thermal ablation process of tumor tissue is analyzed. This enables real-time monitoring of the ablation range and efficacy evaluation of steam thermal ablation. This invention can be used for efficacy evaluation of steam ablation technology, achieving effective assessment of steam ablation efficacy using near-infrared spectroscopy and fluorescence imaging, facilitating adjustments to the treatment plan throughout the treatment process, and improving the reliability and convenience of the treatment process.

[0023] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0024] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0025] Figure 1 This is a system diagram of the steam ablation device of the present invention;

[0026] Figure 2 This is a structural framework diagram of the steam instrument design of the present invention;

[0027] Figure 3 This is the overall design drawing of the instrument of the present invention;

[0028] Figure 4 This is a conceptual diagram of the steam ablation experiment for rabbit in situ tumors according to the present invention;

[0029] Figure 5 This invention provides a framework for a real-time efficacy evaluation model of steam ablation based on multimodal data.

[0030] Figure 6 This invention provides a research scheme for multimodal real-time efficacy evaluation in steam ablation.

[0031] Figure 7 The images are the actual coagulated area images, fused images, color-gradient images, and binarized images of the ex vivo liver of the present invention. Detailed Implementation

[0032] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0033] The steam ablation multimodal efficacy assessment device includes:

[0034] The near-infrared spectroscopy acquisition system is used to collect near-infrared spectral data during steam ablation. The near-infrared spectral data includes temperature parameters and fNIRs parameters.

[0035] A fluorescence imaging system was used to acquire fluorescence signals during vapor ablation. The acquired near-infrared spectral data and fluorescence signals were used as multimodal data.

[0036] The processing module is used to preprocess near-infrared spectral data and fluorescence signals, and to perform cross-modal feature fusion on the preprocessed near-infrared spectral data and fluorescence signals to obtain fused features. The fused features are used to train a multi-task learning network to obtain an evaluation model, and the trained evaluation model is used to evaluate the effect of vapor ablation.

[0037] In embodiments of the present invention, the acquired near-infrared spectral data and fluorescence signals are preprocessed, including: in time series processing, temperature / near-infrared data are truncated at a certain step size, and the mean, standard deviation, trend gradient, and the first 5 FFT main frequency components are extracted; Z-score standardization is used to eliminate differences between devices. In fluorescence image processing, the morphological features (area, roundness, edge curvature) of the ablation zone are extracted through U-Net pre-segmentation, and cubic spline fitting is used to describe the ablation boundary. In data augmentation, time series data are superimposed with Gaussian noise of different degrees, and fluorescence images are randomly rotated by a certain angle and horizontally flipped; to prevent overfitting, the label smoothing coefficient is set to 0.1.

[0038] In one embodiment of the present invention, cross-modal feature fusion is performed on preprocessed near-infrared spectral data and fluorescence signals to obtain fused features, including: extracting temporal features using a 1D CNN and a Transformer encoder to capture temperature change patterns. Image feature extraction uses a ViT model, and tissue heterogeneity features are extracted through multi-layer Transformer encoding. Cross-modal fusion will employ a two-layer cascaded cross-attention mechanism with dynamically adjusted gating weights, where temporal features account for 60%–80%.

[0039] In one embodiment of the present invention, a multi-task learning network is constructed using the obtained fusion features, an interpretable model is designed, and the model is validated. The multi-task learning network is trained using the fusion features to obtain an evaluation model, including: the main task of the multi-task learning network is cell mortality, and the auxiliary task is self-supervised learning. In the loss function design, the main task uses MSE+L1 regularization (λ=0.01), and the auxiliary task uses an improved Dice Loss (weights in edge regions are increased by 50%). In adaptive weighting, gradient automatic balancing is achieved based on Kendall's uncertainty theory. In the prediction head structure, the main task has a 128-dimensional bottleneck layer + Swish activation, and the auxiliary task has 3 layers of transposed convolutions to restore voxel spatial resolution.

[0040] In one embodiment of the present invention, the near-infrared spectral acquisition system includes a light source, an optical fiber probe, and a spectrometer. The optical fiber probe and the temperature probe are integrated on a single ablation needle to form a comprehensive functional needle that combines temperature measurement and near-infrared functions.

[0041] The fluorescence imaging system includes an EMCCD (Electronic Microcontroller Center) fluorescence camera, a laser, and a laser source probe. The laser includes a 488nm laser and a 650nm laser.

[0042] In one embodiment of the present invention, the fiber optic probe includes a spectral fiber and a light-guiding fiber, which are independently packaged and integrated into the same probe. The light generated by the halogen light source is transmitted to the tissue under test through the light-guiding fiber of the fiber optic probe, and after being scattered by the tissue under test, it is transmitted to the fiber optic spectrometer by the spectral fiber.

[0043] In one embodiment of the present invention, the near-infrared spectral acquisition system is specifically used for:

[0044] The ablation site is illuminated by a light source, and a fiber optic probe is inserted into the tissue for testing. One fiber in the probe transmits the light from the light source to the tissue being measured, while a spectrometer receives the light reflected and scattered by the tissue from the other fiber. A computer performs spectral acquisition, and after processing the spectral data, near-infrared spectral data is obtained, such as for biological tissue. s μ a Parameters such as SO2 and Δ[Hb].

[0045] In one embodiment of the present invention, the fluorescence imaging system is specifically used for:

[0046] Tissue ablation is performed using the vaporization of an aqueous solution of fluorescent dye as a heat source. A laser emits light of a specific wavelength to excite the fluorescent sample. After absorbing the excitation light, the fluorescent molecules emit a fluorescent signal with a longer wavelength. An EMCCD fluorescence camera detects the fluorescence signal, converts the light signal into an electrical signal, processes it to generate a fluorescence image, displays the fluorescence intensity distribution, and transmits it to a host computer.

[0047] In one embodiment of the present invention, the fluorescence imaging system employs a fluorescence labeling strategy, dissolving high-temperature stable fluorescent dyes (such as Cy5, ICG) in steam to form an aerosol, which diffuses with the steam to label the tissue area. An EMCCD fluorescence camera captures the fluorescence signal, dynamically tracks the steam diffusion boundary, and displays the steam penetration range in real time.

[0048] In one embodiment of the present invention, the vapor flow rate is quantified by fluorescence intensity gradient to guide dose control, the fluorescence diffusion radius is used to assess the vapor diffusion range and determine whether the target area has been reached, and the fluorescence intensity reflects the vapor concentration and tissue absorption, indirectly assessing the thermal field distribution.

[0049] This invention relates to a multimodal efficacy assessment device for steam ablation, addressing the current inability to achieve real-time efficacy assessment during steam ablation procedures. This project integrates imaging (fluorescence imaging), multi-parameter monitoring (temperature, near-infrared tissue parameters), and steam thermal ablation treatment information to develop a multimodal data real-time monitoring and assessment system. By fusing temperature, optical parameters, and fluorescence data, a real-time thermal map of the ablation zone is generated, establishing a model linking thermal field distribution to biological effects. A "three-state cell damage model" (normal, transitional, and necrotic states) is proposed to accurately assess the boundaries of tumor ablation. Furthermore, multi-parameter feedback is used to adjust the ablation dose, ensuring controllability and high precision of intraoperative efficacy.

[0050] Ex vivo liver ablation was performed using vaporized aqueous solutions of fluorescent dyes (such as ICG) as a heat source. An EMCCD fluorescence camera was used to monitor the fluorescence distribution in the ablation area in real time, obtaining visible light and fluorescence gradation images. Artificial intelligence algorithms were used for image fusion and edge recognition to determine the ablation boundary: 1. Gaussian or median filtering was used to remove noise, and image pixel values ​​were normalized to the range [0,1] or [-1,1]. Data was enhanced through rotation, flipping, and scaling. 2. Convolutional neural networks (CNNs) were used to process the images, and U-Net was used for image segmentation, edge detection, and delineation.

[0051] This project addresses the issue of dimensional heterogeneity in multimodal data monitored during ablation. It will achieve this by deeply coupling spatiotemporally heterogeneous data and employing a cross-modal alignment strategy at the input layer. A spatiotemporal encoder will map 1D temperature / near-infrared parameters and other signals with 2D fluorescence images to a unified 3D latent space. At the model architecture level, a multi-task learning network will be designed, using cell mortality rate as the core supervisory signal and introducing self-supervised auxiliary tasks to enhance feature representation capabilities. To address the limited data volume, a transfer learning strategy will be combined to reuse the pre-trained visual backbone network. StyleGAN will be used for 2D fluorescence image data augmentation, and robust spatiotemporal features will be extracted through contrastive learning pre-training. To address model interpretability, a Transformer attention mechanism will be integrated to dynamically capture temperature-sensitive regions. Furthermore, the proposed heat conduction theory equation will be embedded into the network as a regularization constraint, improving prediction accuracy while also better aligning with pathological patterns.

[0052] The core issue in evaluating the efficacy of tumor thermal ablation is the real-time assessment of tumor cell inactivation and the definition of the ablation boundary during surgery. This invention discovers that the reduced scattering coefficient can characterize the degree of tissue protein coagulation and correlates the intraoperative changes in the reduced scattering coefficient, the reduced scattering coefficient in different regions, and the results of biological effect studies (such as cell inactivation rate), thus achieving microwave ablation efficacy evaluation and boundary definition based on the reduced scattering coefficient. Therefore, this project uses real-time intraoperative data (tissue reduced scattering coefficient and tissue temperature) and postoperative biological effect data (quantitative indicators: cell apoptosis rate, mechanical mortality rate) to establish a mathematical model for real-time efficacy evaluation and boundary definition standards for steam thermal ablation. By correlating the real-time intraoperative data and postoperative biological effect data, a cell inactivation evaluation standard and evaluation model based on the reduced scattering coefficient are obtained. Based on the results of the biological effects, the reduced scattering coefficient threshold of the ablation boundary is obtained, thereby obtaining the real-time tumor ablation boundary.

[0053] like Figure 1 The diagram shown is a system equipment diagram of the steam ablation device of the present invention. Figure 1 (a) is a conceptual design diagram of the instrument, which is a mobile device. The instrument includes a frame (main unit), an external display screen, and external accessories (vapor ablation needle, fluorescence camera, and temperature and near-infrared combined efficacy assessment probe). Figure 1 (b) is a schematic diagram of the internal unit layout of the host, which integrates all hardware components, including the host power supply, main control board, near-infrared spectrometer, broadband light source, steam generator and control module, industrial computer, etc. Figure 1 (c) is a schematic diagram of the system panel, which includes important operation buttons.

[0054] like Figure 2The diagram shows the structural framework of the steam instrument of this invention, including: a steam generator control unit, a host computer, an industrial control computer, a fluorescence imaging unit, and a temperature and near-infrared detection monitoring unit. Fluorescent dye is added before ablation. The liquid used for ablation is pumped into the steam generator via a gear pump. A pressure sensor measures the inlet pressure, and a pressure / flow regulating valve controls the inlet volume. The steam generator vaporizes the liquid into gas, which is then introduced into the steam ablation needle via a heating system. An outlet temperature sensor precisely controls the temperature, and a steam outlet pressure sensor precisely controls the pressure. After the ablation needle is inserted into the appropriate location in the tissue, the steam carrying the fluorescent dye enters the tissue through the ablation needle to begin ablation. The fluorescence camera laser emits light of a specific frequency to excite the fluorescent sample. After absorbing the excitation light, the fluorescent molecules emit fluorescence with a longer wavelength. The EMCCD fluorescence camera quickly captures the fluorescence signal, processes it, and transmits the fluorescence image to the industrial control computer. The optical fiber in the fiber optic probe of the ablation needle transmits light to the tissue being measured. A spectrometer receives the light reflected and scattered by the tissue from another optical fiber, transmitting the near-infrared spectrum and temperature data to the industrial control computer. The industrial control computer has multiple functions, including: receiving feedback on steam temperature and steam pressure from the steam parameter monitoring unit and setting the steam temperature and steam pressure; controlling the image capture time of the EMCCD fluorescence camera; controlling the light intensity and temperature measurement frequency of the fiber optic light source; and feeding the parameters back to the host computer. The host computer uses the parameters fed back from the industrial control computer to generate a fused image of fluorescence and near-infrared spectroscopy through various algorithms, and evaluates the therapeutic effect based on the temperature distribution field, ablation boundary, and steam velocity field.

[0055] like Figure 3 As shown, the overall system of this invention comprises four units: a steam generation and control unit, a steam parameter monitoring unit, a fluorescence imaging unit, and a temperature and near-infrared evaluation unit. The system design utilizes independent unit control boards. The steam generation and control unit includes: one steam generator, one external steam ablation needle, and a gear pump. The temperature and near-infrared evaluation unit includes: one fiber optic spectrometer, one near-infrared light source, one dual-fiber optic probe (with three temperature measurement points), a temperature acquisition board, one temperature measurement needle (with three temperature measurement points), and one combined temperature and near-infrared probe. The fluorescence imaging unit uses an EMCCD camera. Overall system control commands are issued from the computer to the main control board, and then from the main control board to each unit.

[0056] like Figure 4The diagram shown is a conceptual illustration of a steam ablation experiment for an in situ tumor in rabbits. A fenestration was performed on the rabbit's abdomen to expose the liver and tumor. An EMCCD was placed above the tumor to ensure the camera could capture images of the tumor site. An ablation needle was inserted horizontally, and temperature and near-infrared probes were inserted at the tumor edge to acquire real-time fluorescence images of the ablation site and temperature and reduced scattering data at the edge. Combined with simulated temperature field data of in vivo liver tumors, the steam transport and thermal field distribution of in vivo in situ liver tumor steam ablation were studied. Tissue from the ablation site was removed, and the effective ablation area size was evaluated and measured based on reduced scattering data. A dose-response model was established based on an ex vivo liver steam ablation dose-response model.

[0057] like Figure 5 As shown, this is a real-time efficacy assessment model for steam ablation using multimodal data. Near-infrared spectroscopy is 1D time series data, with temperature and near-infrared parameters standardized using Z-score. Fluorescence imaging is 2D spatial image data, used to extract morphological features of the ablation area. Time series data preprocessing includes: market segmentation for statistical feature extraction, FFT for dominant frequency component extraction, and Gaussian noise enhancement. Spatial data preprocessing includes: U-Net pre-segmentation for morphological feature extraction, cubic spline fitting of ablation boundaries, and StyleGAN image enhancement. Temporal features are extracted using CNN and Transformer encoding, while spatial features are extracted using Vision Transformer for tissue heterogeneity extraction. The extracted temporal and spatial features are dynamically adjusted using a two-layer cascaded cross-attention mechanism and gated weights to unify the three-dimensional spatial representation. In the design of the multi-task learning network loss function, the main task uses MSE+L1 regularization (λ=0.01), and the auxiliary task uses an improved Dice Loss (with 50% weight increase in edge regions). In adaptive weighting, automatic gradient balancing is achieved based on Kendall's uncertainty theory. The primary task for predicting the head structure involves a 128-dimensional bottleneck layer plus Swish activation, while the secondary task involves three layers of transposed convolution to restore voxel spatial resolution. Finally, experimental validation will be conducted, using 80% of the animal ablation experimental data as training data and the remaining 20% ​​as validation data, employing five-fold cross-validation. Efficacy evaluation indicators will include cell mortality assessment and ablation area assessment.

[0058] like Figure 6 As shown, during the ablation process, imaging (fluorescence 2D images), multi-parameter monitoring (temperature, near-infrared tissue parameters), and microwave thermal ablation treatment information (distribution of SDH enzyme activity in tissue cells after ablation) are integrated to establish a multimodal intraoperative efficacy evaluation standard for steam ablation, achieving real-time definition of the ablation boundary. Fluorescence images are processed to calculate fluorescence intensity and density; temperature and near-infrared optical parameters of the fluorescence edge are obtained, and combined with the SDH enzyme activity at the corresponding sites, cell inactivation is determined, and a multimodal information fusion-based ablation boundary threshold standard is established.

[0059] like Figure 7 The image shown represents the actual coagulation area, fused image, colorimetric image, and binarized image of the ex vivo liver according to the present invention. After the ablation of the pig liver was completed by introducing ablation water containing a high-temperature resistant fluorescent dye, the actual coagulation area of ​​the pig liver was cut along the cross-section of the tissue. The colorimetric image and the binarized image showed a high degree of consistency, verifying the feasibility of using fluorescence 2D imaging technology as a steam ablation assessment technique.

[0060] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0061] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A steam ablation multimodal efficacy evaluation device, characterized in that, include: Near-infrared spectroscopy acquisition system is used to acquire near-infrared spectral data during vapor ablation; A fluorescence imaging system is used to acquire fluorescence signals during vapor ablation. The processing module is used to preprocess near-infrared spectral data and fluorescence signals, and to perform cross-modal feature fusion on the preprocessed near-infrared spectral data and fluorescence signals to obtain fused features. The fused features are used to train a multi-task learning network to obtain an evaluation model, and the trained evaluation model is used to evaluate the effect of vapor ablation.

2. The apparatus according to claim 1, characterized in that, The near-infrared spectroscopy acquisition system includes a light source, a fiber optic probe, and a spectrometer. The fiber optic probe and the temperature probe are integrated into a single ablation needle, forming a comprehensive functional needle that combines temperature measurement and near-infrared functions. The fluorescence imaging system includes an EMCCD fluorescence camera, a laser, and a laser source probe.

3. The apparatus according to claim 2, characterized in that, The fiber optic probe consists of a spectral fiber and a light-guiding fiber, which are separately packaged and integrated into the same probe. The light generated by the halogen source is transmitted to the tissue under test through the light-guiding fiber of the fiber optic probe, and after being scattered by the tissue under test, it is transmitted to the fiber optic spectrometer by the spectral fiber.

4. The apparatus according to claim 2, characterized in that, The near-infrared spectral acquisition system is specifically used for: The ablation site is illuminated by a light source, and a fiber optic probe is inserted into the tissue for testing. One fiber in the probe transmits the light from the light source to the tissue to be measured, while the spectrometer receives the light reflected and scattered by the tissue from the other fiber. The computer performs spectral acquisition and obtains near-infrared spectral data after processing the spectral data.

5. The apparatus according to claim 2, characterized in that, The fluorescence imaging system is specifically used for: Tissue ablation is performed using a vaporized aqueous solution of fluorescent dye as a heat source. A laser emits light of a specific wavelength to excite the fluorescent sample. After the fluorescent molecules absorb the excitation light, they emit a fluorescent signal. An EMCCD fluorescence camera detects the fluorescent signal, converts the light signal into an electrical signal, processes it to generate a fluorescence image, displays the fluorescence intensity distribution, and transmits it to a host computer.

6. The apparatus according to claim 1, characterized in that, The fluorescence imaging system employs a fluorescence labeling strategy, dissolving high-temperature stable fluorescent dyes in steam to form an aerosol, which diffuses with the steam to label the tissue area. An EMCCD fluorescence camera captures the fluorescence signal, dynamically tracks the steam diffusion boundary, and displays the steam penetration range in real time.

7. The apparatus according to claim 1, characterized in that, The vapor flow rate is quantified by fluorescence intensity gradient to guide dose control. The fluorescence diffusion radius is used to assess the vapor diffusion range and determine whether the target area has been reached. The fluorescence intensity reflects the vapor concentration and tissue absorption, and indirectly assesses the thermal field distribution.

8. The apparatus according to claim 1, characterized in that, The near-infrared spectral data includes temperature parameters and fNIRs parameters. The processing module preprocesses the near-infrared spectral data and fluorescence signals, including: in time series processing, temperature / near-infrared data is truncated at a preset step size, and the mean, standard deviation, trend gradient, and the first 5 FFT main frequency components are extracted; Z-score standardization is used to eliminate differences between devices; in fluorescence signal processing, the morphological features of the ablation zone are extracted through U-Net pre-segmentation, and cubic spline fitting is used to describe the ablation boundary.

9. The apparatus according to claim 1, characterized in that, The preprocessed near-infrared spectral data and fluorescence signals are fused across modal features to obtain fused features, including: extracting temporal features using 1D CNN and Transformer encoder to capture temperature change patterns; using the ViT model for image feature extraction; extracting tissue heterogeneity features through multi-layer Transformer encoding; and using a two-layer cascaded cross-attention mechanism for cross-modal fusion with dynamically adjusted gating weights, of which temporal features account for 60% to 80%.

10. The apparatus according to claim 1, characterized in that, The evaluation model is obtained by training a multi-task learning network using fused features, including: The main task of the multi-task learning network is cell death rate, and the auxiliary task is self-supervised learning. In the loss function design, the main task adopts MSE+L1 regularization, and the auxiliary task uses improved Dice Loss. In the adaptive weighting, the gradient is automatically balanced based on Kendall uncertainty theory. In the prediction head structure, the main task has a 128-dimensional bottleneck layer + Swish activation, and the auxiliary task has 3 layers of transposed convolution to restore voxel spatial resolution.

Images