Method for programming an ultrasonic system for ultrasonic surgery
Patent Information
- Authority / Receiving Office
- WO · WO
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2026-02-03
- Publication Date
- 2026-08-13
Smart Images
Figure IMGF000008_0001 
Figure IMGF000012_0001 
Figure 00000015_0000
Abstract
Description
[0001] Method for programming an ultrasound system for ultrasound surgery
[0002] The invention relates to a method for programming an ultrasound system for ultrasound surgery comprising an ultrasound instrument having a sonotrode and a piezoelectric transducer, an ultrasound generator driving the transducer and a control system influencing operating parameters of the ultrasound generator.
[0003] In ultrasonic surgery, particularly in liver surgery, neurosurgery, and spinal surgery, ultrasonic systems are used for the coarse removal and fine preparation of tissue. Known ultrasonic systems designed for ultrasonic surgery, such as the ultrasonic system described above in EP 4389024 Al, consist primarily of an ultrasonic generator that transmits alternating current at a very high frequency to an ultrasonic instrument, which, as a so-called ultrasonic aspirator, is additionally equipped with a device for rinsing and / or suction. The ultrasonic instrument converts the electrical energy into mechanical motion using piezoceramics, which is amplified by a sonotrode. Upon contact between the vibrating sonotrode and tissue, a tissue effect occurs, depending on the technical design, in the form of fragmentation, ablation, coagulation, or dissection.The extent of these effects depends on the energy input of the sonotrode and on the acoustic and biological properties of the tissue, which ultimately manifest themselves in its mechanical properties. Soft tissue (e.g., liver parenchyma) requires less energy input than highly elastic tissue (e.g., blood vessels) or hard tissue (e.g., bone).
[0004] Medical professionals, especially in neurosurgery, face the challenge of precisely identifying different tissue types during surgery. This includes determining where healthy tissue ends and tumor tissue begins, as well as adjusting device settings for optimal tumor resection.
[0005] Surgeons must strike a balance between completely removing the tumor tissue and minimizing damage to healthy tissue, as damage to healthy tissue can have serious consequences and remaining tumor tissue can cause the tumor to recur. In making these decisions, surgeons often rely on tactile sensations gained by palpating the tissue to determine its type based on its mechanical properties. However, this method is highly subjective and can be affected by the surgeon's experience, stress, and fatigue.
[0006] In addition to these fundamental problems, there is also a practical problem regarding the handling of existing ultrasound systems. These systems essentially combine three functions: tissue fragmentation through the vibration of the instrument tip (sonotrode), aspiration of the fragmented tissue, and irrigation. These three parameters must be repeatedly adjusted manually to ensure an optimal configuration for each tissue type and surgical situation, thus guaranteeing efficient and safe surgical procedures.Because the parameters cannot usually be adjusted by the surgeon themselves due to the device's design, and an optimal combination of settings is often not directly selectable, there is a significant investment of personnel and time, which also involves iterative trial and error to determine suitable parameter combinations. This increases the complexity of the operation and can slow down the surgical process.
[0007] Although EP 4389024 Al already proposes to provide predetermined combinations of operating parameters with which the ultrasound generator and thus the sonotrode can be operated depending on the tissue type, uncertainties remain regarding the determination of the tissue type.
[0008] The object of the invention is therefore to create an ultrasound system for ultrasound surgery of the type mentioned above, with the help of which an operative procedure can be carried out in a targeted, efficient and more successful manner compared to known ultrasound systems.
[0009] This problem is solved according to the invention by the method with the features of claim 1 and the features of claim 11. The dependent claims describe advantageous embodiments of the invention. The invention thus proposes a method for programming an ultrasound system for ultrasound surgery, comprising an ultrasound instrument with a sonotrode and a piezoelectric transducer, an ultrasound generator driving the transducer, and a control system influencing the operating parameters of the ultrasound generator, with the following steps:
[0010] a) Operating the sonotrode at a predetermined operating frequency,
[0011] b) Establishing contact between the sonotrode operated at the predetermined operating frequency and a test object having a predetermined modulus of elasticity,
[0012] c) Acquiring at least one electrical measurement quantity applied to the sonotrode before and during step b, wherein the electrical measurement quantity is acquired as an independent measurement value at a predetermined sampling rate,
[0013] d) Creating a dataset comprising those measurements where each immediately following a previous measurement differs from the previous measurement by a predetermined threshold, e) Identifying characteristics of the test object by machine learning based on the dataset and
[0014] f) Storing the characteristic features identified by machine learning as an object type classifying an object with a predetermined elastic modulus in the ultrasound system.
[0015] The object type to which a predetermined modulus of elasticity or range of elastic modulus is assigned may, for example, correspond to a specific hardness type, such as "soft" or "hard", with these object types being further subdivided, for example, into "very soft", "medium soft", "medium hard", and "very hard".
[0016] Preferably, steps b) to e) are repeated with at least one further test object that differs from the test object having a predetermined elastic modulus with respect to its elastic modulus, wherein step f) comprises storing the characteristic features identified by machine learning for classifying the objects with a different predetermined elastic modulus. Furthermore, a subsequent step g) is provided in which at least one predetermined operating parameter adapted to the respective object type is stored in the ultrasound system.
[0017] The procedure is carried out in such a way that the operating frequency of the sonotrode has a frequency in the range of 20 kHz to 40 kHz.
[0018] The test object is preferably designed as a tissue phantom, wherein the tissue phantom is particularly preferably made of (partially) cross-linked polyvinyl alcohol.
[0019] The electrical measurement quantity is further preferably selected from the group of electrical measurement quantities consisting of RMS voltage, RMS current, frequency and phase.
[0020] The sampling rate specifically has a frequency in the range of 0.3 kHz to 1 kHz.
[0021] In particular, the data set includes measurements that meet the following condition:
[0022] ti
[0023] ? previous = 1 y
[0024] w -i
[0025] i=twl
[0026] fv -üP reV ' l 2
[0027] value t = f -prev - sign(V t — V prev ),
[0028] where
[0029] w is the window size, which indicates how many of the last w data points are used to calculate the mean measurement V. prev to be taken into account at time t;
[0030] V t the currently measured value; and
[0031] value t a value that exceeds the predetermined threshold.
[0032] The surgical ultrasound instrument is preferably an ultrasound aspirator, wherein it is particularly preferred that the sonotrode of the ultrasound aspirator is operated during simultaneous irrigation and aspiration of the test object.
[0033] Finally, a method for operating an ultrasound system programmed according to one of the preceding claims for ultrasound surgery is also proposed, comprising an ultrasound instrument having a sonotrode and a piezoelectric transducer, an ultrasound generator driving the transducer and a control system influencing operating parameters of the ultrasound generator, comprising the steps:
[0034] a) Operating the sonotrode at a predetermined operating frequency,
[0035] b) Establishing contact between the sonotrode operated at the predetermined working frequency and a biological tissue,
[0036] c) Acquiring at least one electrical measurement quantity applied to the sonotrode before and during step b, wherein the electrical measurement quantity is acquired as an independent measurement value at a predetermined sampling rate,
[0037] d) Creating a dataset comprising those measurements where each immediately following a previous measurement differs from the previous measurement by a predetermined threshold; e) Identifying features characteristic of the biological tissue by machine learning based on the dataset.
[0038] f) Assigning the characteristic features of the biological tissue to an object type created using machine learning and
[0039] g) Displaying predetermined operating parameters adapted to the assigned object type.
[0040] The invention is further explained below:
[0041] The sonotrode of the ultrasound system operates at a frequency of 20,000 to 40,000 Hz. Additionally, if the ultrasound instrument is configured as an ultrasound aspirator, aspiration and irrigation can be activated. The ultrasound generator, or the control unit that controls the ultrasound generator, drives the sonotrode and continuously acquires data during operation, which is used for contact and tissue classification. The values that can be acquired and used for contact and tissue classification are the effective voltage, effective current, frequency, and phase.
[0042] These values are continuously output at a sampling rate of 300 to 1000 Hz and forwarded via a CAN protocol to a connected computing unit, which is part of the ultrasound system, for further processing.
[0043] The processing unit receives, stores, and processes the data supplied by the ultrasound generator. Machine learning models are used to analyze the acquired data, which places high demands on computing power. Furthermore, the results must be provided at near real-time speed to avoid disrupting the clinical workflow. All calculations are preferably performed locally to completely avoid data transmission over networks during clinical use and thus also comply with data protection requirements.
[0044] The method presented here is preferably performed on test objects designed as tissue phantoms. These tissue phantoms can be produced reproducibly using standardized procedures, ensuring consistent tissue properties and creating a reliable database for training algorithms.
[0045] The fabrication of the tissue phantoms is based primarily on a polyvinyl alcohol (PVA) mixture, whereby the mechanical properties of the tissue phantom can be specifically adjusted by the amount of PVA added: a higher concentration results in a firmer model, whereas a lower concentration produces a softer material. The PVA mixture is preferably poured into silicone molds and then subjected to a cooling process. The tissue phantoms produced in this way are stored in double-distilled water to prevent drying out and to preserve their mechanical properties.
[0046] The tissue phantoms produced in this way have an E-modulus of up to 100 kPa, whereas tumor tissue is typically in a range of approximately 100 Pa to 7,000 Pa.
[0047] Using tissue phantoms, various data sets have been collected as part of investigations into the present invention, which are used for different applications: first, a data set was created for the development and validation of the contact classification, then a data set for tissue recognition was generated.
[0048] The tissue models serve to collect training data for tissue classification. Various tissue models with different elastic moduli were used for this purpose. The data acquisition process followed a defined sequence: 1. Activation of the instrument, 2. Establishment of tissue contact for approximately 1 second, and 3. Dissolution of the contact. Accordingly, Fig. 1 shows an exemplary complete signal waveform (A) with the contact initiation (B) and contact termination (C) shown over time.
[0049] Further preferred parameters for carrying out the method according to the invention relate to the applied contact force, which is less than 1 N, and a working speed adapted to typical surgical procedures. It is particularly preferred that the instrument angle, applied force, and working speed be varied within realistic ranges to cover the broadest possible range of applications. During data collection, it is therefore essential to ensure that all tissue strengths of the tissue phantoms are recorded with similar frequency, resulting in a balanced distribution of classes for training the tissue classification system. The elastic modulus of the respective tissue phantom serves as the "ground truth."
[0050] To develop the classification system according to the invention, which distinguishes between contact and non-contact, a data set is created in each case, which records during electrical ablation processes whether the instrument is in contact with the test object or not.
[0051] Tissue strength is determined by the interaction between the sonotrode and the test object. The system analyzes the data acquired by the sensors to characterize the object's mechanical properties. Due to the system's complexity, arising from the instrument's control and the subtle variations in the specified tissue strengths reflected in the test objects, the processing is divided into several steps and supported by machine learning (ML). The first processing step involves contact detection. Subsequent processing steps are only triggered if contacts are found. Since a human-operated instrument can only be consciously moved within a timeframe of approximately 0.1–3 seconds, only temporal signal changes within this range are considered.Signal changes over a longer time frame, on the other hand, can be attributed to temperature gradients. The signal profiles before and after the detected contact are compared. To efficiently train the machine learning models based on the contacts found, the calculation of suitable features is crucial. These features serve as fundamental information that enables the models to recognize relevant patterns and relationships in the data.
[0052] The data is treated as a data stream, and not as a static database, whereby each new time step must be classified solely based on knowledge of the previous data, without having any information about future values.
[0053] Several methods are conceivable for detecting the feedback of tissue upon contact on its vibrational behavior. This includes, among other things, modeling the signal attenuation associated with mass coupling. Statistical methods such as Change Point Detection (CPD) algorithms or machine learning techniques like Random Forests and neural networks (LSTM, RNN, etc.) are also suitable. The present method proposes a simple, threshold-based approach that can react dynamically based on past values. The basic approach involves evaluating a specific past time window relative to the current values. The evaluation determines whether the current values differ significantly from previous data, indicating a change in the contact state. The equations used are as follows:
[0054]
[0055] Here, w is the window size, indicating how many of the last w data points are considered for calculating the mean voltage level Vtprev at time t. A deviation is detected if the calculated value exceeds a predetermined threshold. This threshold is determined during the training phase by iteratively finding the optimal value that best distinguishes the predefined "ground truth" labels (contact / non-contact) for the pre-recorded database.
[0056] The advantage of this approach lies in its simplicity and short computation time, which is particularly important for real-time applications. By continuously considering previous data, sudden changes, especially in voltage values – as shown in Fig. 1 – can be reliably detected without the need for time-consuming training phases or complex models.
[0057] In the feature generation phase, the dataset is used according to the invention for tissue classification to extract meaningful features that are relevant for tissue recognition. Contact classification is used to identify individual contacts. For each identified contact, a selection window of 500 milliseconds before and after the contact is defined, representing the transition phase between the tissues. This time span makes it possible to capture relevant information about the dynamic properties of the contacts.
[0058] Within this contact area, various features are calculated to quantify the characteristic features of the data. These include, in particular:
[0059] The “absolute increase”, which represents the absolute change in the measured quantity within the contact window and provides information about the intensity of the transition.
[0060] The “relative increase”, which considers a change in relation to the previous state in order to assess the proportion of the increase in the context of the entire measurement series.
[0061] The "slope" indicates how quickly the transition occurs over time, which is important for the analysis of material behavior.
[0062] The calculations are preferably performed for the three relevant signal parameters: RMS voltage, frequency, and phase – see Fig. 2, which shows the signal waveform (signal over time) with detected contact and generated features for voltage (A), phase (B), and frequency (C). Each of these features contributes to the comprehensive characterization of the tissue transitions and improves the performance of subsequent machine learning models.
[0063] The result of this phase is a comprehensive database containing all extracted features, along with their corresponding elastic moduli. This structured database forms the basis for creating and training machine learning models for tissue classification and enables a precise analysis of material properties.
[0064] The machine learning model used for tissue classification is preferably based on a hierarchical structure. Instead of directly detecting the exact value of the Young's modulus using a single model, classification is preferably performed by a sequence of several hierarchically organized models. This approach improves the precision in distinguishing tissues with similar Young's modulus values and also increases the reliability of the predictions.
[0065] The hierarchical structure allows for the training of specific models at each level, which only need to distinguish between a few categories. For example, a model at a particular level only needs to differentiate between the category "very hard" and the category "very soft," instead of handling all possible tissue classes simultaneously, as would be the case with a flat classification approach.
[0066] If the model is uncertain about the final stiffness class assignment, for example, whether the tissue is "very soft" or "medium soft," the hierarchy allows a return to a higher level for a coarser but more reliable classification. A flat classification approach cannot be used in this case.
[0067] The present invention offers significant advantages. Firstly, it enables surgical procedures to be performed with increased accuracy. Due to the automatic recognition of the tissue type, an improved assessment of tissue properties can be made, leading to more accurate device settings and thus better surgical outcomes. Simultaneously, the surgeon's cognitive load is minimized, as the largely automated device settings relieve the surgeon of the burden of adjusting device properties, allowing them to focus on other aspects of the procedure. Finally, the data acquisition and processing enable adjustments to the device settings, which in turn increases the efficiency of the procedure.
Claims
REQUIREMENTS 1. Method for programming an ultrasound system for ultrasound surgery comprising an ultrasound instrument having a sonotrode and a piezoelectric transducer, an ultrasound generator driving the transducer and a control system influencing operating parameters of the ultrasound generator, characterized by the steps: a) Operating the sonotrode at a predetermined operating frequency, b) Establishing contact between the sonotrode operated at the predetermined operating frequency and a test object having a predetermined modulus of elasticity, c) Acquiring at least one electrical measurement quantity applied to the sonotrode before and during step b, wherein the electrical measurement quantity is acquired as an independent measurement value at a predetermined sampling rate, d) Creating a dataset comprising those measurements where each measurement immediately following a previous measurement differs from the previous measurement by a predetermined threshold, e) Identifying characteristics of the test object through machine learning using the data set and f) Storing the characteristic features identified by machine learning as an object type classifying an object with a predetermined elastic modulus in the ultrasound system.
2. Method according to claim 1, characterized by repeating steps b) to e) with at least one further test object which differs from the test object having a predetermined modulus of elasticity with respect to the modulus of elasticity, wherein step f) comprises storing the characteristic features identified by machine learning of the object types classifying the objects with a predetermined modulus of elasticity that differs from each other.
3. Method according to one of the preceding claims, characterized by the step following step f): g) Storing at least one predetermined operating parameter adapted to the respective object type in the ultrasound system.
4. Method according to one of the preceding claims, characterized in that the operating frequency of the sonotrode is a frequency in the range of 20 kHz to 40 kHz.
5. Method according to one of the preceding claims, characterized in that the test object is a tissue phantom.
6. Method according to one of the preceding claims, characterized in that the electrical measurement quantity is selected from the group of electrical measurement quantities consisting of RMS voltage, RMS current, frequency and phase.
7. Method according to one of the preceding claims, characterized in that the sampling rate is a frequency in a range of 0.3 kHz to 1 kHz.
8. Method according to one of the preceding claims, characterized in that the data set comprises measured values that satisfy the following condition: where w is the window size, which indicates how many of the last w data points are used to calculate the mean measurement V. t prev to be taken into account at time t; V t the currently measured value; and value ta value that exceeds the predetermined threshold.
9. Method according to one of the preceding claims, characterized in that the surgical ultrasound instrument is an ultrasound aspirator.
10. Method according to claim 9, characterized in that the sonotrode of the ultrasonic aspirator is operated during simultaneous irrigation and aspiration of the test object.
11. Method for operating an ultrasound system programmed according to one of the preceding claims for ultrasound surgery, comprising an ultrasound instrument having a sonotrode and a piezoelectric transducer, an ultrasound generator driving the transducer, and a control system influencing operating parameters of the ultrasound generator. characterized by the steps: a) Operating the sonotrode at a predetermined operating frequency, b) Establishing contact between the sonotrode operated at the predetermined operating frequency and a biological tissue, c) Acquiring at least one electrical measurement applied to the sonotrode before and during step b, wherein the electrical measurement is acquired as an independent measurement at a predetermined sampling rate, d) Creating a dataset comprising those measurements where each immediately following a previous measurement differs from the previous measurement by a predetermined threshold; e) Identifying features characteristic of the biological tissue by machine learning based on the dataset. f) Assigning the characteristic features of the biological tissue to an object type created using machine learning and g) Displaying predetermined values adapted to the assigned object type B etri eb sparam eter .