Contact type ultrasonic deep tissue pressure measuring instrument and deep tissue pressure monitoring method

Abstract

The invention discloses a contact type ultrasonic deep tissue pressure measuring instrument and a deep tissue pressure monitoring method, and relates to the technical field of medical instruments. The pressure measuring instrument comprises an ultrasonic probe body, a grip strength detection module, a data processing module and a wireless transmission module. A deep learning model is arranged in the data processing module, and non-invasive, accurate and real-time measurement of deep tissue pressure is achieved by synchronously fusing grip signals of an operator and tissue deformation data in an ultrasonic image. The method comprises the steps of synchronous acquisition of grip strength and ultrasonic images, signal preprocessing, multi-modal deep learning fusion calculation and result wireless transmission and display. The problem of accurate monitoring of deep tissue pressure is solved, and the method is suitable for pressure treatment scenes such as lower limb deep vein thrombosis prevention and the like.

Description

Technical Field

[0001] This invention belongs to the field of medical device technology, specifically relating to a contact-type ultrasonic deep tissue pressure measuring instrument and a method for monitoring deep tissue pressure. Background Technology

[0002] Pressure therapy is an important and effective physical therapy method in clinical medicine. It primarily involves applying controlled external pressure to prevent deep vein thrombosis in the lower extremities, inhibit hypertrophic scarring after burns, promote venous return in the lower extremities, and assist in hemostasis. The key to the effectiveness of this treatment lies in precisely controlling the applied pressure within a specific "therapeutic window." Insufficient pressure will fail to achieve the desired therapeutic effect, while excessive pressure may compress nerves and blood vessels, leading to tissue ischemia, necrosis, or even new pressure injuries.

[0003] However, current pressure therapy techniques have a fundamental limitation: they cannot achieve non-invasive, precise, and real-time monitoring of pressure on deep tissues. Current mainstream clinical practice still relies on the operator's experience-based estimation and manual palpation. Healthcare professionals manually apply pressure based on personal experience and roughly judge its appropriateness by touch. This method is highly subjective, prone to significant errors, and cannot provide any objective quantitative indicators, making it difficult to guarantee treatment effectiveness and posing significant safety risks.

[0004] To overcome the limitations of subjectivity, some devices capable of measuring surface pressure have emerged in existing technologies, such as pressure measurement systems using airbag or membrane sensors. While these technologies provide quantifiable pressure data, their measurements are limited to the body surface. Due to significant individual and locational variations in the thickness, hardness, and elasticity of human tissues (including skin, subcutaneous fat, and muscle), unpredictable attenuation and changes occur during the transmission of surface pressure to deeper tissues (such as intermuscular veins). Therefore, the correlation between surface pressure values ​​and the actual stress on deeper tissues is weak, leaving treatment largely in a state of imprecise "blind pressure."

[0005] Ultrasound imaging technology, particularly ultrasound elastography, has been used to assess the relative mechanical properties of tissues, providing auxiliary information for identifying tissue properties. However, this technology typically provides qualitative or semi-quantitative relative values ​​(such as strain ratios), rather than the absolute pressure physical quantities required for therapeutic monitoring. The equipment is usually complex and expensive, and operation and analysis require a high degree of expertise. It is primarily used for diagnostic screening of diseases such as tumors and is difficult to integrate into routine pressure therapy procedures for low-cost, convenient, real-time monitoring.

[0006] In summary, current pressure therapy techniques severely limit their efficacy and safety due to the inability to overcome the technical bottleneck of non-invasive and precise monitoring of deep tissue pressure. Clinical practice urgently needs innovative equipment and methods that can directly link surface manipulation with deep biomechanical effects, enabling real-time, non-invasive, precise, and quantitative measurement of absolute pressure in deep tissues. Summary of the Invention

[0007] To address the aforementioned technical problems, this invention provides a contact-type ultrasonic deep tissue pressure measuring instrument and a method for monitoring deep tissue pressure. Through multimodal information fusion and deep learning technology, accurate, non-invasive, and real-time monitoring of deep tissue pressure is achieved.

[0008] To solve the above-mentioned technical problems, the present invention adopts the following technical solution:

[0009] A contact-type ultrasonic deep tissue pressure measuring device includes an ultrasonic probe body, a pressure monitoring component, and a wireless transmission module. The pressure monitoring component includes a grip force detection module for detecting the operator's grip force and a data processing module for identifying tissue deformation.

[0010] The data processing module is connected to the grip strength detection module and the ultrasonic probe body signal respectively;

[0011] The wireless transmission module is electrically connected to the data processing module and is used to transmit the pressure calculation results to the outside.

[0012] The data processing module has a built-in multimodal deep learning model that can simultaneously fuse grip force data and tissue deformation data and output deep tissue pressure values.

[0013] Furthermore, the grip force detection module is a flexible thin-film pressure sensor with a thickness ≤0.2mm, an area ≤1mm, and a sensitivity ≥90%, and is attached to the grip area of ​​the handheld part shell of the ultrasonic probe body.

[0014] Furthermore, the data processing module includes a main control chip and an image processing unit. The main control chip is an STM32 series microcontroller, and the image processing unit is connected to the signal output terminal of the ultrasound probe to acquire and analyze tissue deformation data in the ultrasound image.

[0015] Furthermore, the main control chip is electrically connected to the grip force detection module, and can synchronously receive grip force data output by the grip force detection module and tissue deformation data output by the image processing unit, and perform fusion calculation through the multimodal deep learning model.

[0016] Furthermore, the wireless transmission module is a Bluetooth module, used to send pressure data to an external display device, which is a smartphone or tablet running a visualization application that can receive and display pressure data through the wireless transmission module.

[0017] A contact ultrasound deep tissue pressure monitoring method, using the contact ultrasound deep tissue pressure measuring instrument as described above, specifically includes the following steps:

[0018] Step 1: The grip force detection module collects the grip force signal applied by the operator to the ultrasound probe in real time, and the ultrasound probe obtains real-time ultrasound images of deep tissues.

[0019] Step 2: Preprocess the grip strength signal and ultrasound image;

[0020] Step 3: Input the preprocessed grip strength data and ultrasound images into the trained multimodal deep learning model to process the grip strength data and ultrasound images simultaneously.

[0021] Step 4: Output deep tissue pressure estimates using a multimodal deep learning model;

[0022] Step 5: The pressure estimate and ultrasound image are sent to the display device via the wireless transmission module.

[0023] Further, step 2 involves preprocessing the grip strength signal and ultrasound image; specifically, it includes the following steps:

[0024] Step 2.1: Perform low-pass filtering on the grip strength signal using a fourth-order Butterworth filter with a cutoff frequency of 10Hz to remove high-frequency noise and motion artifacts.

[0025] Step 2.2: Normalize the ultrasound image, reducing the pixel values ​​from [0,255] to the range [0,1]. The calculation formula used is as follows: ;in, These are the original image pixel values. These are the normalized pixel values;

[0026] Step 2.3, Tissue Deformation Rate Calculation: The tissue deformation rate between consecutive ultrasound image frames is calculated using the time-domain difference method. The calculation formula is as follows:

[0027]

[0028] in express Organizational deformation at time, The sampling time interval, for The rate of tissue deformation at any given time.

[0029] Furthermore, the multimodal deep learning model adopts a multimodal fusion architecture, specifically including the following processing steps:

[0030] Step 3.1, Image Feature Extraction: The ultrasound image sequence is processed using a convolutional neural network based on the EfficientNet-B0 architecture, outputting a 256-dimensional image feature vector. ;

[0031] Step 3.2, Mechanical Feature Extraction: A network with gated recurrent units (GRUs) is used to process the grip force time-series data, outputting a 32-dimensional mechanical feature vector. ;

[0032] Step 3.3, Feature Fusion: A cross-modal attention mechanism is used to fuse image features and mechanical features. The specific calculation formula is as follows:

[0033]

[0034]

[0035] in, For the sigmoid function, and For learnable parameters, [;] indicates feature concatenation, and ⊙ indicates element-wise multiplication;

[0036] Step 3.4, stress regression prediction, will fuse features The input consists of a regressor composed of two fully connected layers, and the final output is an estimate of deep tissue pressure.

[0037] Furthermore, the multimodal deep learning model in step 4 employs a multi-task learning framework to simultaneously optimize deep tissue pressure prediction and tissue stiffness classification, specifically including:

[0038] Step 4.1: Main task loss function, using the Huber loss function for deep tissue pressure regression prediction, the calculation formula is as follows:

[0039]

[0040] in, This represents the actual stress value. To predict pressure values, This is a hyperparameter, with a value of 5 mmHg;

[0041] Step 4.2: The auxiliary task loss function adopts the cross-entropy loss function for tissue hardness classification. The calculation formula is as follows: ;in, This is a true label for tissue stiffness. The predicted tissue stiffness probability distribution; Step 4.3: The total loss function is the weighted sum of the two tasks, calculated as follows:

[0042] Where α and β are weighting coefficients, with values ​​of 0.8 and 0.2 respectively.

[0043] Furthermore, the deep learning model also integrates an online adaptive learning mechanism, which can continuously fine-tune and optimize the model parameters based on new data from clinical use.

[0044] Compared with the prior art, the present invention has the following advantages:

[0045] 1. Compared with traditional surface pressure measurement (mean absolute error ±10.5 mmHg, correlation coefficient 0.65) and ultrasound elastography (correlation coefficient 0.52), this invention fuses grip force and tissue deformation data through a multimodal deep learning model. Experiments show that the mean absolute error is only ±2.1 mmHg, the correlation coefficient is as high as 0.96, and the accuracy is improved by about 80%. This solves the problem of "blind pressure" in pressure therapy, elevating the treatment from an experience-based operation to a precise quantitative level.

[0046] 2. The deep learning model used in this invention does not simply stitch together data. Instead, it adaptively balances and fuses image features and mechanical temporal features through a cross-modal attention mechanism, overcoming the limitations of a single data source. Experiments show that in repeatability tests, the standard deviation of the pressure measurement in this invention is ±1.5 mmHg, and the coefficient of variation (CV) is as low as 3.2%, significantly better than traditional surface pressure measurement (CV=12.5%) and ultrasound elastography (CV=15.8%), resulting in more accurate and reliable final pressure estimates with stronger anti-interference capabilities.

[0047] 3. This invention not only completes the main task of pressure prediction, but also performs joint optimization through an auxiliary tissue stiffness classification task, making the learned features more discriminative. Furthermore, the system integrates an online adaptive learning mechanism, enabling continuous fine-tuning using new data in clinical practice. Experiments show that after one week of online learning, the model's average error on new patient data can be reduced from the initial ±3.5 mmHg to ±2.3 mmHg, thus adapting to different patients' tissue characteristics (such as obesity and edema) and different operators' usage habits. This effectively solves the measurement bias problem caused by individual differences, improving the system's robustness and clinical universality.

[0048] 4. This invention embeds advanced algorithms into a portable ultrasound probe system, which connects wirelessly to smart mobile terminals (phones, tablets) for display. The overall system latency is less than 50ms, meeting real-time monitoring requirements. Clinical trials show that in scenarios involving the prevention of deep vein thrombosis in the lower extremities, the early warning accuracy reaches 98%, with a false alarm rate of less than 2%. This design retains the portability and accessibility of traditional ultrasound while avoiding the use of expensive and bulky dedicated elastography equipment. The entire system is low-cost, easy to operate, and readily integrated into existing clinical stress treatment procedures, possessing extremely high commercialization potential and market value. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 This is a schematic diagram of the contact ultrasonic deep tissue pressure measuring instrument of the present invention;

[0051] Figure 2 This is a structural block diagram of each module of the contact ultrasonic deep tissue manometer of the present invention;

[0052] Figure 3 This is a schematic diagram of the multimodal fusion architecture of the present invention;

[0053] In the diagram: 1. Grip strength detection module; 2. Data processing module; 201. Main control chip; 202. Image processing unit; 3. Ultrasonic probe body. Detailed Implementation

[0054] To gain a deeper understanding of this invention, we will provide a comprehensive and detailed description. However, this invention has various implementations and is not limited to the specific examples listed herein. These examples are presented to enhance a full understanding of the disclosure of this invention.

[0055] like Figures 1 to 3As shown, this invention specifically provides a contact-type ultrasound deep tissue pressure measuring instrument and a corresponding pressure monitoring method. The pressure measuring instrument mainly includes an ultrasound probe body 3, a grip force detection module 1 integrated into the probe handle, a data processing module 2, and a wireless transmission module. The grip force detection module 1 uses a flexible thin-film pressure sensor with a thickness not exceeding 0.2 mm, an area not exceeding 1 cm², and a sensitivity of over 90%. It is firmly attached to the outer shell surface of the probe gripping area using medical-grade adhesive. The data processing module includes a main control chip and an image processing unit. The main control chip is an STM32F series microcontroller, responsible for system control and data acquisition scheduling. The image processing unit uses an embedded GPU module, directly receiving the image output signal from the ultrasound device, and is responsible for real-time image processing algorithms and deep learning model inference. The wireless transmission module uses a low-power Bluetooth chip CC2640R2F to achieve data transmission with a smart mobile terminal.

[0056] Furthermore, the deep tissue pressure monitoring method provided by this invention mainly includes the following steps:

[0057] Step 1: System initialization, including main control chip clock configuration, ADC module initialization, image processing unit startup, Bluetooth protocol stack loading, and ultrasound image acquisition parameter settings; specifically as follows: The main control chip first configures the system clock to 168MHz, initializes the ADC module to acquire grip force sensor signals at a sampling rate of 1kHz, sets the image receiving resolution of the image processing unit to 1920×1080 pixels and the frame rate to 30fps; the Bluetooth module is configured to use the GATT protocol to establish a data transmission channel with the mobile terminal.

[0058] Step 2: The operator holds the probe to contact the tissue being tested. The system simultaneously acquires the analog voltage signal output by the grip force sensor and the B-mode image sequence generated by the ultrasound equipment. The main control chip ensures precise synchronization between the mechanical signal and the image frame through hardware timestamps.

[0059] Step 3: Preprocess the collected raw data.

[0060] Step 3.1: Perform low-pass filtering on the grip strength signal using a fourth-order Butterworth filter with a cutoff frequency of 10Hz to remove high-frequency noise and motion artifacts.

[0061] Step 3.2: Normalize the ultrasound image, reducing the pixel values ​​from [0,255] to the range [0,1]. The calculation formula used is as follows: ;in, These are the original image pixel values. These are the normalized pixel values;

[0062] Step 3.3, Tissue deformation rate calculation: The tissue deformation rate between consecutive ultrasound image frames is calculated using the temporal difference method (tissue deformation rate is estimated by pixel displacement between consecutive frames). The sampling time interval is 33.3 ms. The calculation formula is as follows:

[0063]

[0064] in express Organizational deformation at time, The sampling time interval, for The rate of tissue deformation at any given time.

[0065] Step 4: Input the preprocessed grip strength time series data and ultrasound image sequence into the trained deep learning model. The model adopts a multimodal fusion architecture and optimizes the pressure regression prediction and tissue stiffness classification tasks.

[0066] Step 4.1, Image Feature Extraction: The ultrasound image sequence is processed using a convolutional neural network based on the EfficientNet-B0 architecture, outputting a 256-dimensional image feature vector. ;

[0067] Step 4.2, mechanical feature extraction: The grip force time series data is processed using a GRU network with gated recurrent units, and a 32-dimensional mechanical feature vector is output. ;

[0068] Step 4.3, Feature Fusion: A cross-modal attention mechanism is used to fuse image features and mechanical features. The specific calculation formula is as follows:

[0069]

[0070]

[0071] in, For the sigmoid function, and For learnable parameters, [;] indicates feature concatenation, and ⊙ indicates element-wise multiplication;

[0072] Step 4.4, stress regression prediction, will fuse features The input consists of a regressor composed of two fully connected layers, and the final output is an estimate of deep tissue pressure.

[0073] Step 5: The fused feature vectors are used to calculate the deep tissue pressure estimate through a fully connected layer regressor. The results are then transmitted in real time via Bluetooth to a dedicated application on a smartphone or tablet for visualization.

[0074] Step 5.1: Main task loss function, using the Huber loss function for deep tissue pressure regression prediction, the calculation formula is as follows:

[0075]

[0076] in, This represents the actual stress value. To predict pressure values, For hyperparameters;

[0077] Step 5.2: The auxiliary task loss function adopts the cross-entropy loss function for tissue hardness classification. The calculation formula is as follows: ;in, This is a true label for tissue stiffness. The predicted tissue stiffness probability distribution; Step 5.3: The total loss function is the weighted sum of the two tasks, calculated as follows:

[0078] Where α and β are weighting coefficients, with weighting coefficients of 0.8 and 0.2, respectively.

[0079] Model training used a paired dataset acquired from ex vivo experiments, which included grip force sequences, ultrasound images, and ground truth values ​​of invasive pressure measured simultaneously, and was performed end-to-end using the Adam optimizer.

[0080] Step 6: The system supports online learning. When clinical validation data is obtained, the model parameters can be fine-tuned to improve the accuracy of individualized measurements. When the system detects a small deviation between the measurement results and the clinical reference values, new data pairs can be added to the training sample library. The system periodically starts an incremental learning process during idle periods, using a small learning rate (1e-5) to fine-tune the parameters of the last three layers of the model, while retaining existing knowledge to avoid catastrophic forgetting.

[0081] The grip strength detection module's specific circuitry connects to the main control chip via an I2C interface. The sensor is powered by a stable 3.3V voltage provided by an LDO regulator chip, and an RC low-pass filter circuit is configured at the signal output terminal to further improve signal quality.

[0082] like Figure 3 As shown, in the multimodal fusion architecture of the deep learning model, the image feature extraction branch uses an EfficientNet-B0 network pre-trained on ImageNet, outputting a 256-dimensional feature vector; the mechanics feature extraction branch uses a two-layer bidirectional GRU network with 64 hidden layer units, outputting a 32-dimensional feature vector. In the feature fusion stage, attention weight matrices are used to adaptively weight the bimodal features, and finally, stress regression prediction is completed through two fully connected layers (64 and 32 neurons).

[0083] The wireless transmission solution employed in this invention offers significant advantages in clinical environments. Bluetooth 5.0 technology supports a communication distance of up to 100 meters and features low power consumption, allowing for 8 hours of continuous operation on a single charge. Compared to traditional wired connections, wireless transmission avoids the limitations of cables on operational flexibility and simplifies equipment sterilization procedures. In multi-device collaborative scenarios such as operating rooms and rehabilitation departments, the communication solution employed in this invention does not cause electromagnetic interference to other medical devices, and the data transmission latency is less than 50ms, meeting the clinical needs for real-time monitoring.

[0084] To verify the effectiveness, accuracy, and practicality of this invention, systematic experimental tests were conducted. The technical effects of this invention are illustrated below using specific experimental data:

[0085] The testing equipment used was the pressure measuring instrument prototype described in this embodiment of the invention. The control devices were a traditional airbag-type surface pressure measuring instrument and a clinical ultrasound elastography system. Twenty healthy volunteers, aged 25-60 years and weighing 50-85 kg, were recruited. With ethical approval, invasive fiber optic pressure sensors were pre-implanted in the target muscle groups of the volunteers' calves as the gold standard reference.

[0086] The device simultaneously recorded the estimated deep tissue pressure output by the invention, the readings of the surface pressure measuring instrument, and the strain ratio from ultrasound elastography, and compared them with the gold standard readings of invasive pressure sensors. The results showed that the mean absolute error between the measured values ​​of the invention and the gold standard values ​​was ±2.1 mmHg, the root mean square error was ±2.8 mmHg, and the correlation coefficient reached 0.96. Under the same conditions, 10 repeated measurements were performed on the same site, and the standard deviation of the pressure measurements was ±1.5 mmHg, with a coefficient of variation of 3.2%.

[0087] Compared with traditional body surface pressure measuring instruments (mean absolute error ±10.5 mmHg, coefficient of variation 12.5%) and ultrasonic elastography technology (coefficient of variation 15.8%), the measurement accuracy and repeatability of the present invention are significantly improved.

[0088] Contents not described in detail in this specification are prior art known to those skilled in the art. Although illustrative specific embodiments of the invention have been described above to facilitate understanding by those skilled in the art, it should be understood that the invention is not limited to the scope of the specific embodiments. Various modifications are readily apparent to those skilled in the art as long as they fall within the spirit and scope of the invention as defined and determined by the appended claims, and all inventions utilizing the concept of this invention are protected.

Claims

1. A contact type ultrasonic deep tissue manometer comprising an ultrasonic probe body, a pressure monitoring assembly, and a wireless transmission module, characterized in that, The pressure monitoring assembly comprises a grip detection module (1) for detecting the operator's grip and a data processing module (2) for identifying tissue deformation; The data processing module (2) is respectively connected with the grip detection module (1) and the ultrasonic probe body (3) in signal connection; The wireless transmission module is electrically connected with the data processing module (2) and is used for transmitting the pressure calculation result to the outside; The data processing module (2) is built-in with a multi-modal deep learning model, which can synchronously fuse the grip data and the tissue deformation data and output the deep tissue pressure value.

2. The contact ultrasound deep tissue manometer according to claim 1, characterized in that, The grip detection module (1) is a flexible thin film pressure sensor, which has a thickness of ≤0.2mm, an area of ≤1㎠, and a sensitivity of ≥90%, and is attached to the grip area of the handheld part shell of the ultrasonic probe body (3).

3. The contact ultrasound deep tissue manometer according to claim 1, characterized in that, The data processing module (2) comprises a main control chip (201) and an image processing unit (202), the main control chip (201) adopts an STM32 series microcontroller, and the image processing unit (202) is connected with the signal output end of the ultrasonic probe and is used for acquiring and analyzing the tissue deformation data in the ultrasonic image.

4. The contact ultrasound deep tissue manometer according to claim 1, characterized in that, The main control chip (201) is electrically connected with the grip detection module (1), can synchronously receive the grip data output by the grip detection module (1) and the tissue deformation data output by the image processing unit (202), and performs fusion calculation through the multi-modal deep learning model.

5. The contact ultrasound deep tissue manometer according to claim 1, characterized in that, The wireless transmission module is a Bluetooth module, which is used for sending the pressure data to an external display device.

6. A method of contact ultrasound deep tissue pressure monitoring, characterized by, The contact type ultrasonic deep tissue pressure gauge according to any one of claims 1-5 is used for deep tissue pressure monitoring, and specifically comprises the following steps: Step 1: Real-time acquisition of the grip signal of the operator applied to the ultrasonic probe through the grip detection module, and acquisition of the real-time ultrasonic image of the deep tissue through the ultrasonic probe; Step 2: Preprocessing of the grip signal and the ultrasonic image; Step 3: Input of the preprocessed grip data and ultrasonic image into the trained multi-modal deep learning model, and synchronous processing of the grip data and the ultrasonic image; Step 4: Output of the deep tissue pressure estimation value through the multi-modal deep learning model; Step 5: Sending of the pressure estimation value and the ultrasonic image to a display device through the wireless transmission module.

7. The method of claim 6, wherein the method further comprises: In step 2, the grip signal and the ultrasonic image are preprocessed, specifically comprising the following steps: Step 2.1: Low-pass filtering of the grip signal, using a fourth-order Butterworth filter with a cutoff frequency of 10Hz to remove high-frequency noise and motion artifacts; Step 2.2, the ultrasound image is normalized, the image pixel value is normalized from [0, 255] to [0, 1] range, and the calculation formula adopted is: ; wherein, is the original image pixel value, is the normalized pixel value; Step 2.3: Calculation of the tissue deformation rate using the time domain difference method to calculate the tissue deformation rate between consecutive ultrasonic image frames, and the calculation formula is: ； wherein denotes the tissue deformation variable at the time instant is the sampling time interval, is the tissue deformation rate at the time instant 8. The method of claim 6, wherein the method further comprises: The multi-modal deep learning model adopts a multi-modal fusion architecture, specifically comprising the following processing steps: Step 3.1, image feature extraction, process the ultrasound image sequence using a convolutional neural network based on EfficientNet-B0 architecture, output a 256-dimensional image feature vector ; Step 3.2, mechanical feature extraction, the network with gated recurrent unit processes the grip timing data, and outputs a 32-dimensional mechanical feature vector ; Step 3.3: Feature fusion, using a cross-modal attention mechanism to fuse image features and mechanical features, and the specific calculation formula is: ； ； where, is a sigmoid function, and are learnable parameters, [;] denotes feature concatenation, and ⊙ denotes element-wise multiplication. Step 3.4, pressure regression prediction, fuse features The regressor consisting of two fully connected layers takes the input and outputs the final estimate of deep tissue pressure.

9. The method of claim 6, wherein the method further comprises: In step 4, the multi-modal deep learning model adopts a multi-task learning framework to simultaneously optimize deep tissue pressure prediction and tissue hardness classification, specifically comprising: Step 4.1: Main task loss function, Huber loss function is used for deep tissue pressure regression prediction, the formula is: ； wherein, is a true pressure value, is a predicted pressure value, is a hyperparameter; Step 4.2: The auxiliary task loss function adopts a cross-entropy loss function for the hardness classification, and the calculation formula is: wherein, is the true tissue hardness class label, is the predicted tissue hardness probability distribution; Step 4.3: The total loss function is the weighted sum of the two tasks, and the calculation formula is: ; where a and β are weighting factors.

10. The method of claim 6, wherein the method further comprises: The deep learning model also integrates an online adaptive learning mechanism, which can continuously fine-tune and optimize the model parameters based on new data in clinical use.

Images