CN-121977716-A - Ultrasonic temperature measurement system and method based on multi-feature fusion

CN121977716ACN 121977716 ACN121977716 ACN 121977716ACN-121977716-A

Abstract

The invention discloses an ultrasonic temperature measurement method and system based on multi-feature fusion, and belongs to the technical field of medical ultrasonic monitoring. The system comprises a signal acquisition module, a data processing module and an output module. The method comprises the steps of collecting an original radio frequency signal acquired by a monitoring probe spatially associated with a focused ultrasonic treatment probe, extracting three temperature rise characteristics based on thermal strain, nonlinear acoustic attenuation and Nakagami distribution statistical principles in parallel, inputting multi-parameter characteristics into a pre-trained knowledge distillation physical information neural network model, directly outputting a predicted temperature value, and calculating accumulated equivalent thermal dose according to the predicted temperature to judge damage. According to the invention, through multi-parameter complementary fusion and a lightweight intelligent model, accurate and real-time non-invasive monitoring of temperature change in a high-temperature treatment interval of more than 50 ℃ is realized, the root mean square error of temperature measurement is less than or equal to 0.8 ℃, the single-frame processing time is less than or equal to 5ms, and meanwhile, standardized thermal damage assessment is provided, and the controllability and safety of focused ultrasound treatment are improved.

Inventors

LI CHENGHAI
WU JIWEI
LI FAQI
WU XINYAO
SHU JIAFU

Assignees

重庆医科大学

Dates

Publication Date: 20260505
Application Date: 20260212

Claims (10)

1. An ultrasonic thermometry system based on multi-feature fusion, comprising: The signal acquisition module is used for receiving an original radio frequency signal from a monitoring ultrasonic transducer which is coaxially arranged with the focused ultrasonic treatment probe and is arranged in the center of the transducer so as to acquire the radio frequency signal of the energy action area of the focused ultrasonic treatment probe; the data processing module is electrically connected with the signal acquisition module, and the data processing module comprises: The characteristic extraction submodule is used for extracting a first temperature rise characteristic based on a thermal strain principle, a second temperature rise characteristic based on a nonlinear sound attenuation principle and a third temperature rise characteristic based on a Nakagami distribution statistical principle from the original radio frequency signal in parallel; A temperature prediction sub-module, in which a knowledge distillation physical information neural network model is pre-stored for receiving the first, second and third temperature rise characteristics and outputting corresponding predicted temperature values, and And the output module is used for outputting the predicted temperature value.
2. The multi-feature fusion-based ultrasound thermometry system of claim 1, wherein the signal acquisition module includes a high-speed analog-to-digital converter and a graphics processor for supporting real-time parallel computation of the feature extraction sub-module and the temperature prediction sub-module.
3. The multi-feature fusion based ultrasound thermometry system of claim 1, wherein the feature extraction sub-module is configured to extract the first temperature rise feature by: according to the first radio frequency signal before heating and the second radio frequency signal after heating, determining axial time shift delta t (z) through cross-correlation calculation; Performing differential operation on the axial time shift delta t (z) to obtain thermal strain epsilon (z); Calculating the first temperature rise characteristic Δt strain (z) based on the formula Δt strain (z) = (kc 0/2) ∈ (z); wherein c0 is the tissue reference sound velocity, and k is the calibration coefficient.
4. The multi-feature fusion based ultrasound thermometry system of claim 1, wherein the feature extraction sub-module is configured to extract the second temperature rise feature by: Calculating integral powers P pre (z) and P post (z) of the first radio frequency signal before heating and the second radio frequency signal after heating in the sliding window respectively; Calculating an attenuation change amount Δatt (z) based on the formula Δatt (z) = -Re [ ln (P post (z)/P pre (z)) ]/2z, where Re is a real part and z is depth; The second temperature rise characteristic Δt att (z) is calculated based on the formula Δt att (z) =a (Δatt (z)) b+c, where a, b, c are fitting coefficients.
5. The multi-feature fusion based ultrasound thermometry system of claim 1, wherein the feature extraction sub-module is configured to extract the third temperature rise feature by: Performing Hilbert transform on the original radio frequency signal to extract a signal envelope A (z); Within the statistical window, the second moment E [ A 2 (z) ] and the fourth moment E [ A 4 (z) ] of the signal envelope A (z) are calculated; Wherein the second moment E [ A 2 (z) ] represents the average intensity of the echo signal, and the fourth moment E [ A 4 (z) ] is used to characterize the degree of dispersion of the intensity; Calculating the third temperature rise characteristic Δt naka (z) based on the formula Δt naka (z) =dΔm (z) +e; Wherein Deltam (z) is the variation of the shape parameters before and after heating, and d and e are calibration coefficients.
6. The multi-feature fusion-based ultrasound thermometry system of claim 1, wherein the knowledge-distilled physical information neural network model is trained by a method comprising the steps of: constructing a training data set, wherein the data set comprises living in-vitro tissue experiment radio frequency signal data and a corresponding real temperature label; Constructing a teacher network and a student network; Training the teacher network by using the training data set, wherein a training loss function comprises a physical equation residual loss term constructed based on a Pennes biological heat transfer equation; And after the teacher network training is completed, carrying out knowledge distillation training on the student network by utilizing the teacher network so as to obtain the knowledge distillation physical information neural network model.
7. The multi-feature fusion based ultrasound thermometry system of claim 1, wherein the output module is further configured to: Calculating an accumulated equivalent thermal dose CEM43 ℃ from the predicted temperature value; The tolerance of different tissues to thermal injury is significantly different, the irreversible injury threshold can be quantified through CEM43 ℃, the value represents equivalent conversion of the exposure time of the tissues at different temperatures, and when the accumulated thermal dose reaches the CEM43 ℃ threshold of the corresponding tissues, the irreversible thermal injury can be judged.
8. The multi-feature fusion based ultrasound thermometry system of claim 1, wherein the knowledge-distilled physical information neural network model in the temperature prediction sub-module is configured to perform the following optimized prediction procedure: (a) The input and fusion are carried out, namely, the first, the second and the third temperature rise characteristics which are output by the characteristic extraction submodule in parallel are received, and are fused with the historical temperature sequence and the preoperative treatment energy parameters to form a multidimensional fusion characteristic vector; (b) The double-network collaborative reasoning is that the fusion feature vector is simultaneously input into a pre-trained teacher network and a light student network; the teacher network is ResNet-18 variant with four hidden layers, the number of neurons is set, and a high-fidelity temperature predicted value is output; the student network is a fully-connected network with four hidden layers, the number of neurons and the total parameters are set, and a real-time temperature predicted value is output; (c) Dynamic fusion and physical constraints: In the training stage, data fitting loss is calculated through a dynamic weight strategy, wherein weight coefficients corresponding to the first, second and third temperature rise characteristics are dynamically adjusted according to the real-time predicted temperature rise, and an adjusting formula is as follows: w strain =max(0.1,0.8-0.014・ΔTpred); w att =min(0.7,0.014・ΔTpred); w naka = 0.2 (fixed); construction of the Total loss function L total =1.0×L PDE +λ data ×L data +0.5×L KD ; Wherein, the Is a physical constraint residual loss based on a discretized Pennes biological heat transfer equation; To use distillation temperature After softening teacher network output, the distillation loss calculated with student network output is=1.25; (d) And when the CEM43 ℃ value exceeds a preset damage threshold corresponding to the type of the target tissue, generating a control instruction to dynamically adjust the output power or the acting time of the focused ultrasound treatment probe to form a treatment closed loop.
9. An ultrasound thermometry method based on multi-feature fusion, characterized in that it is performed by a system according to any one of claims 1 to 8, the method comprising: Acquiring an original radio frequency signal from a monitoring ultrasonic transducer through the signal acquisition module; The feature extraction step is that the first temperature rise feature, the second temperature rise feature and the third temperature rise feature are extracted in parallel from the original radio frequency signal through the feature extraction submodule; And a temperature prediction step, namely inputting the first temperature rise characteristic, the second temperature rise characteristic and the third temperature rise characteristic into the knowledge distillation physical information neural network model through the temperature prediction submodule to obtain and output a predicted temperature value.
10. The multi-feature fusion-based ultrasound thermometry method of claim 9, further comprising, after the temperature prediction step: And a thermal dose judging step of calculating an accumulated equivalent thermal dose CEM43 ℃ according to the predicted temperature value and comparing the accumulated equivalent thermal dose CEM43 ℃ with a preset threshold range to generate a thermal damage judging result.

Description

Ultrasonic temperature measurement system and method based on multi-feature fusion Technical Field The invention relates to the technical field of medical ultrasonic monitoring, in particular to an ultrasonic temperature measurement system and method based on multi-feature fusion. Background In Focused Ultrasound (HIFU) therapy, intraoperative accurate monitoring of target temperature is critical to ensure ablation efficacy and to prevent damage to surrounding normal tissue. At present, the ultrasonic-based temperature measurement technology mainly depends on a single physical principle, and has inherent limitations such as that after tissue is denatured (about 50 ℃ or more) by a thermal strain-based method, the physical basis is changed, the temperature measurement accuracy is reduced, an acoustic attenuation-based method is mainly sensitive to a high-temperature stage (> 50 ℃), and a Nakagami statistical distribution-based method lacks clear physical mechanism support. The prior art fails to efficiently integrate these complementary multi-parameter information. In addition, conventional thermometry models often rely on simplified physical assumptions, which make it difficult to accommodate dynamic changes in tissue characteristics during treatment, resulting in thermometry bias. At the same time, there is a lack of machine learning models that can be efficiently mapped directly from the original ultrasonic Radio Frequency (RF) signal to the temperature profile. In a specific implementation level, the existing scheme generally does not disclose a specific judgment threshold value of irreversible thermal injury, specific algorithm and calibration details of multi-parameter extraction and a complete strategy of model training and optimization, so that the practicability and repeatability of the method are insufficient, and the requirements of clinic on real-time and accurate temperature prediction and evaluation of a critical treatment interval of more than 50 ℃ cannot be met. Disclosure of Invention In view of the above, the invention aims to provide an ultrasonic temperature measurement method and system based on multi-feature fusion, which solve the problems of inaccurate single ultrasonic temperature measurement, poor model real-time performance and nonstandard damage evaluation in HIFU treatment. Three complementary temperature characteristics of thermal strain, acoustic attenuation and Nakagami are extracted in parallel, fusion prediction is carried out by utilizing a light knowledge distillation physical information neural network (KD-PINN), and standardized damage judgment is realized by combining a CEM43 ℃ threshold. In order to achieve the above purpose, the present invention provides the following technical solutions: the invention provides an ultrasonic temperature measurement system based on multi-feature fusion, which comprises the following steps: The signal acquisition module is used for receiving an original radio frequency signal from a monitoring ultrasonic transducer, and the monitoring ultrasonic transducer and the focused ultrasonic treatment probe are coaxially arranged in space and are arranged in the center of the transducer so as to acquire the radio frequency signal of an energy acting area of the focused ultrasonic treatment probe; the data processing module is electrically connected with the signal acquisition module, and the data processing module comprises: The characteristic extraction submodule is used for extracting a first temperature rise characteristic based on a thermal strain principle, a second temperature rise characteristic based on a nonlinear sound attenuation principle and a third temperature rise characteristic based on a Nakagami distribution statistical principle from the original radio frequency signal in parallel; A temperature prediction sub-module, in which a knowledge distillation physical information neural network model is pre-stored for receiving the first, second and third temperature rise characteristics and outputting corresponding predicted temperature values, and And the output module is used for outputting the predicted temperature value. Further, the signal acquisition module comprises a high-speed analog-to-digital converter and a graphics processor, and is used for supporting real-time parallel computation of the feature extraction sub-module and the temperature prediction sub-module. Further, the feature extraction sub-module is configured to extract the first temperature rise feature by: according to the first radio frequency signal before heating and the second radio frequency signal after heating, determining axial time shift delta t (z) through cross-correlation calculation; Performing differential operation on the axial time shift delta t (z) to obtain thermal strain epsilon (z); Calculating the first temperature rise characteristic Δt strain (z) based on the formula Δt strain(z)=(kc0/2) epsilon (z); Where c 0 is the tissue reference soun