AI-based generator for surgical instruments

Abstract

This invention provides a generator that enables instrument manipulation to achieve a desired penetration depth into tissue, largely independent of the operator's visual observation or individual skill. [Solution] By using the generator 10 according to the present invention, the treatment of biological tissue with electrosurgical instruments, particularly argon plasma probes, can be reliably performed without depending on the individual skill of the operator. For example, multiple test treatments of tissue samples are performed with the help of image-assisted measurement generation using a camera or medical imaging device such as a CT scanner, and a training dataset is created based on these. From this training dataset, a control dataset is created based on machine learning, and the control dataset controls the device 10 located in the operating room without the help of camera observation of the surgical field during subsequent use. Only typical patterns of sensor data assigned to specific tissue effects during the training period under camera surveillance are evaluated.

Description

Technical Field

[0001] The present invention relates to an AI generator for operating a surgical instrument based on artificial intelligence or including a control module including artificial intelligence.

Background Art

[0002] An electrosurgical generator functions to supply an electric current and, if necessary, a fluid such as an additional operating medium such as argon, CO2, NaCl solution, etc. to an electrosurgical instrument. Thereby, since the introduction of energy into living tissue and thus the resulting surgical effect can be directly controlled by the control of the electric generator, this control is particularly important. However, the relationship between variables characterizing the current such as frequency, voltage, current intensity, power, modulation format, duty ratio, etc., and the electrical variables of the load caused by living tissue such as impedance, and the surgical effect that can be obtained, is complex. For this reason, various methods have been tried so far to use the concept of artificial intelligence for the control of the generator. In this regard, Patent Document 1 describes a computer-implemented method using electrical variables, switching conditions of the instrument, and image data obtained from a camera for training purposes. Even while each trained module is used later, this module uses the currently detected electrical variables and further image data to estimate whether the procedure will be successful with the current settings.

[0003] From Patent Document 2, an algorithm for estimating tissue parameters based on machine learning is known. The tissue parameters used to control the energy output are determined based on electrical parameters.

[0004] Patent Document 3 discloses an apparatus for performing endoscopic surgery, in which an optical recording device is part of the apparatus, and the field of view of the optical recording device is directed to the tissue being treated or the tissue being treated by the RF electrode. Before or during tissue treatment, the tissue type of the tissue within the area of ​​the RF electrode is classified based on the optical measurement signal of the recording device, and an RF mode that matches the determined tissue type is set based on the result of the optical classification.

[0005] Further prior art consists of Patent Documents 4, 5, 6, and 7. [Prior art documents] [Patent Documents]

[0006] [Patent Document 1] U.S. Patent Application Publication No. 2023 / 0071343 [Patent Document 2] U.S. Patent Application Publication No. 2020 / 0265309 [Patent Document 3] German Patent Application Publication No. 102020105835 Specification [Patent Document 4] German Patent Application Publication No. 102021101410 Specification [Patent Document 5] U.S. Patent Application Publication No. 2023 / 0420032 [Patent Document 6] European Patent Application Publication No. 4134029 [Patent Document 7] European Patent Application Publication No. 3541313 [Overview of the Initiative] [Problems that the invention aims to solve]

[0007] For electrosurgical procedures involving tissue, particularly argon plasma coagulation, practitioners require considerable experience to be able to assess the later-visible results—that is, the effects on the tissue—while the procedure is being performed. The plasma acting on the tissue emits light itself, which can cause blinding effects. Furthermore, the spectral composition of the emitted light can complicate the evaluation of the results by visual inspection alone. Additionally, the generation of smoke or vapor, or partial covering of the treatment site by instruments or tissue, can further complicate the evaluation of results during the procedure.

[0008] Based on these considerations, the objective of the present invention is to provide a generator that enables instrument manipulation to achieve a desired effect on tissue, such as a desired penetration depth of the effect into the tissue, independently of visual observation by the practitioner and largely independent of the practitioner's individual abilities. [Means for solving the problem]

[0009] This objective is achieved by the generator described in claim 1.

[0010] The generator according to the present invention comprises a control module connected to sensors for electrical variables. The electrical variables are derived primarily, and preferably, only from the current and voltage supplied to the electrical appliance connected to the generator. The electrical variables can be supplied by sensors, such as current sensors, voltage sensors, sensors for determining the power factor, sensors for determining the nonlinearity of the connected load, etc. However, the control module for operating the generator receives only electrical variables and does not receive image data that can be obtained from imaging devices such as cameras or CT scanners. Rather, the control module may be "blind" during the evaluation of the effects of surgical procedures. Thereafter, the control module operates based on an AI module that creates a control dataset during training execution. The electrical variables already described above are included in the training data during training and are later monitored during procedures on patients as well. During training, in addition to electrical variables, a set of image data including information on the penetration depth of tissue effects resulting from procedures such as coagulation or ablation can be used. For example, this penetration depth can be provided in numerical form as an effect label or profile, or in a color-coded manner, and can be associated with the detected electrical variables as additional training information. In this application, the term "image data" includes measurement data from a camera and figures such as images obtained from measurements of non-imaging methods. For example, point-like or two-dimensional spectral measurement data from a diffuse reflectance spectrometer can be processed into a parametric two-dimensional map, which forms image data in the sense of this specification.

[0011] The sensors considered in the training dataset do not need to be identical to the sensors connected to the control module during surgery. Preferably, the measurements for the training dataset are provided by other sensors of a similar type that detect the same type of electrical values. This makes it possible to create training datasets independently of the specific configuration of the sensor technology used during surgery.

[0012] Image data can be recorded and stored as individual pictures when a desired effect level is reached. Alternatively, the image data may include multiple images or even a video sequence obtained at different time intervals. For training, image data from an external camera recording the procedure and its progress can be used. Alternatively or additionally, the image data may be from a medical imaging device, such as a computed tomography (CT) scanner, magnetic resonance imaging (MRT) scanner, ultrasound device, optical coherence tomography (OCT) scanner, or diffuse reflectance (DRS) scanner, rather than from a camera. The image data used in this embodiment may further include information regarding the penetration depth.

[0013] Furthermore, the tissue effects obtained on living tissue during training are classified and assigned effect labels. In this application, an effect label refers to each target variable related to the success of the procedure, including qualitative parameters such as the degree of coagulation or ablation, and quantitative parameters such as the penetration depth to which the tissue effect was obtained. This can be done manually or automatically.

[0014] In certain embodiments, image data may also serve as a source of information regarding the penetration depth of tissue effects induced by procedures such as coagulation or ablation. Depth effects can be provided in the form of numerical values, depth profiles, or color-coded diagrams within two-dimensional or three-dimensional images. Penetration depth measurements may be based on changes in tissue properties related to the coagulation or ablation effect. Such changes may be optical, auditory, electrical, or chemical in nature and can be detected by known medical measurement methods. The thus determined penetration depth is considered when assigning labels and is part of each effect label associated with the detected electrical variables.

[0015] The effect on tissue can be achieved by argon plasma coagulation, in which energy is transmitted by ionized argon gas without direct contact with the electrodes. Alternatively or additionally, contact coagulation can be used, in which the electrodes are in direct contact with the tissue surface during energy output. Both methods are suitable for providing a penetration depth of tissue effect, which can be detected as part of the effect label and combined with the control according to the present invention.

[0016] The AI ​​module uses machine learning to determine a control dataset from a training dataset, which represents a desired effect label, i.e., an electrical variable or pattern of electrical variables that must be present to obtain a tissue effect corresponding to a desired penetration depth of tissue effect. Thus, in the first embodiment, the pattern of electrical variables includes a predetermined value or range of values ​​for each monitored electrical variable. In the extended embodiment, the pattern of electrical variables includes individual values ​​of the electrical variables obtained at time intervals or the temporal progression of each monitored electrical variable. In another, more complex embodiment, the pattern includes at least the monitored electrical variables under two different power output conditions of the generator. Different power outputs can be achieved by different voltages, currents, current limits, modulation modes, etc. Different modulation modes can be, for example, continuous wave (CW), amplitude modulation, pulse pause sampling with a constant or variable pulse pause ratio. These different power outputs can define different modes that produce different surgical effects. Furthermore, the pattern may include the temporal progression of one or more electrical variables before and after switching the generator to a different power output or mode. By using these different techniques, different requirements regarding the accuracy of the treatment outcome can be met.

[0017] The above two modes are preferably distinguished at least with respect to the power output to the instrument. In the first mode, a high power can be output, and the second mode is characterized by a lower power. The power in the first mode is preferably set to a magnitude such that inactivation and coagulation of the biological tissue occur, and in relation thereto, a desired tissue effect having a predetermined penetration depth is rapidly obtained. The power is preferably higher than 10 W. The power in the second mode is preferably set to a magnitude that does not bring a surgical effect, particularly an inactivation effect, to the tissue. The power can be limited to a value of 10 W or less.

[0018] When a control module having an IA module determines a pattern of electrical variables to which a desired effect label is assigned according to any of the above embodiments, it is preferable to switch from the first mode to the second mode. Optionally, the control module can further monitor the pattern of electrical variables even after the switch to the second mode thereby. Thereby, in particular, the non-linear electrical characteristics and the time-dependent electrical characteristics in the biological tissue can be considered. For example, among other things, the tissue impedance can even change during the second mode of low power after the action by high power. For example, the tissue impedance can decrease due to the rewetting of a tissue section that was previously dry.

[0019] Furthermore, the control module can switch back from mode 2 to mode 1 (strong) if the pattern of electrical variables detected in mode 2 (weak) corresponds to an effect label that is too low, such as a penetration depth that is too shallow and / or a degree of tanning that is too low for the patient's tissue effect. The reverse is also true; the control module can switch from mode 2 to mode 3 if the pattern detected in mode 2 (weak) corresponds to an effect label that is too high, such as a penetration depth that is too deep and / or a degree of tanning that is too high for the patient's tissue effect. Since tissue changes due to electrical action are almost irreversible, this situation is in itself an unexpected and unintended case. Mode 3 can be a switch-off mode that prevents further access to the desired tissue effect. Tissue effects are changes in tissue, such as coagulation.

[0020] The control module may have a distance measurement function. The control module may act to inhibit the generation of data that are electrical variables of the sensor that are difficult or no longer interpretable. When the distance between the instrument and the tissue is too long, signal distortion may occur due to the large (too large) effect of the electrical nonlinearity of the arc or plasma provided between the instrument (especially its electrodes) and the tissue, which affects or hinders reliable signal interpretation. For example, distance determination can be performed by pattern recognition, in which case a specific pattern of electrical variables based on excessively long discharge distances (plasma) is recognized, for example, to prompt a surgeon to change the position of the instrument.

[0021] The present invention also relates to a method for training a control module of an electric generator and its AI module, and to the subsequent operation of the generator as described above. This method includes a training step in which a training data set is generated. The training data set is preferably obtained with invariant settings, which are predefined for the generator. With this setting, the instrument provided by the generator acts on the tissue, whereby electrical variables obtained from this action, in particular their time course and values, and the resulting tissue changes are detected, and these tissue changes can be recorded by a camera in the form of pictures, picture sequences or video sequences. Additionally or alternatively, the depth of the tissue change can also be detected in the form of measurement data, which can include, for example, image sequences, video sequences, spectral transitions, etc. by CT, MRT, ultrasound, OCT, or DRS imaging devices. The treated tissue samples are examined automatically or manually and classified in terms of the resulting tissue effects and / or the depth of penetration of the resulting tissue effects. Accordingly, an effect label is assigned to the obtained patterns of images, image sequences and / or video sequences, and electrical variables. From the overall obtained patterns of electrical variables, images, image sequences and / or video sequences, and the assigned effect labels (training data set), a control data set is generated that represents only the relationship between the electrical variables and the effect labels. Thus, the electrical action on the living tissue is controlled by the control data set without using camera images or other images generated by imaging, so that the treatment result corresponds to the desired effect label.

[0022] This provides a generator that, like conventional generators, allows the user to select the desired intensity of the effect in a specific mode. However, the intensity of the effect is no longer an abstract scale value assigned to the output power, e.g., 1-10, but rather a preset treatment result. For example, the intensity of the effect can be preset to multiple levels, e.g., four levels (low coagulation, weak coagulation, strong coagulation, very strong coagulation). If the practitioner selects weak coagulation, the generator operates in a first mode based on a control dataset until weak coagulation is achieved, and then switches to a second mode where no further coagulation is achieved. If the practitioner selects strong coagulation, the generator similarly operates in the first mode until strong coagulation is actually achieved, and then switches to the second mode once strong coagulation is achieved. In this way, the treatment result is almost independent of the practitioner's individual skill. The practitioner can always obtain the desired surgical effect on the tissue in front of them, essentially independent of their subjective optical assessment of the changes occurring in the tissue during treatment.

[0023] In certain embodiments, the present invention includes a method for coagulation and / or ablation of patient tissue, particularly mucosal tissue. This method is particularly useful for surface treatment of large areas, where a tissue effect over the largest possible area is desired, such as in the treatment of obesity, Barrett's esophagus, endometriosis, or similar diseases. For example, the method can be used to perform ablation and / or coagulation of the gastric mucosa. For example, this method is: - The step of inserting an electrosurgical instrument, which is guided endoscopically, into a tissue area of ​​the patient's body to be treated, for example, the patient's stomach, - A step of selecting a desired tissue effect that includes at least a desired penetration depth of the tissue effect and / or a desired degree of tissue inactivation for treating tissue sections, - A step of detecting electrical variables during treatment using an instrument, and determining the current tissue effect, particularly the current penetration depth and / or the current degree of tissue inactivation, based on these electrical variables and a control dataset pre-generated by machine learning, wherein the control dataset is obtained from a training dataset containing electrical variables and target variables assigned to those electrical variables, - A step of controlling the power output of the device according to a specific tissue effect so that the desired tissue effect is achieved and the power output of the device is automatically reduced or terminated so as not to become excessive thereafter. Includes.

[0024] The method steps shown can also be performed in a different order, or at least partially in parallel, as long as it is technically useful and convenient.

[0025] By using the method described above, damage to the deeper layers of the treated tissue can be reduced to a very small degree, so that, for example, only the surface tissue layer, such as the surface mucosa, is treated. Depending on the treatment, the surface tissue layer may contain ghrelin cells, which are important for treating obesity, or endometriotic lesions.

[0026] The desired penetration depth of the tissue effect can be set before tissue ablation and / or coagulation. In particular, this method does not require further surgical procedures such as introducing a substance between or into the mucosa to protect the submucosa.

[0027] In a preferred embodiment, the penetration depth is approximately two-thirds, preferably one-half or one-quarter, of the thickness of the submucosa. Additionally or alternatively, it is possible to generate gastric wall scarring that reduces the elasticity of the patient's stomach wall and, consequently, the stomach's storage capacity, thereby achieving a desired degree of scarring. By using a control dataset trained on mucosal tissue, the energy output can be precisely set for specific characteristics of the gastric mucosa.

[0028] Further details and advantageous embodiments of the present invention are derived from the following description and drawings. [Brief explanation of the drawing]

[0029] [Figure 1] Figure 1 is a schematic diagram of the generator according to the present invention, the instrument connected to the generator, and the living subject during a surgical procedure. [Figure 2] Figure 2 is a schematic diagram of an example of a generator for generating training datasets using a camera during application to a living organism, along with the equipment connected to the generator. [Figure 2a] Figure 2a is a schematic diagram of another example of a generator for generating training datasets with a camera during operation on a living subject, along with the equipment connected to the generator. [Figure 2b] Figure 2b is a schematic diagram of another example of a generator for generating training datasets using a CT scanner during application to a living subject, along with the equipment connected to the generator. [Figure 3] Figure 3 is a block diagram illustrating how to obtain a training dataset from a generator. [Figure 3a] Figure 3a is a block diagram illustrating how to obtain a training dataset from a generator using a CT scanner instead of a camera. [Figure 4] Figure 4 shows different patterns of electrical variables obtained during the treatment of biological tissue. [Figure 5] Figure 5 shows the generator being switched to different modes. [Modes for carrying out the invention]

[0030] Figure 1 shows a generator 10 according to the present invention to which an instrument 11, in particular an argon plasma instrument, is connected. The instrument 11 is connected to a generator output unit 13 via a cable 12, and a neutral electrode 14 is also connected to the generator output unit 13 to guide the output current back to the instrument 11. This structure is applied to monopolar instruments. In the case of bipolar instruments, both poles of the output of the generator output unit 13 are connected to the instrument 11. For example, a control module 19 can control the degree of effect that can be obtained on tissue, i.e., the degree of tanning or deactivation of the treated tissue and / or a preset penetration depth of the tissue effect. Thus, the control is based on a control dataset created using training data. For example, a training dataset can be generated for a defined tissue type, e.g., mucosal tissue.

[0031] Although not shown in Figure 1 for clarity, if the apparatus 11 is specifically an argon plasma apparatus, then the apparatus 11 is further supplied with gas, particularly argon, from the generator 10 or another supply device. As a result, the training dataset, and consequently the control dataset, may also include gas flow parameters.

[0032] The apparatus 11 includes an electrode 15, shown by a dashed line in Figure 1, which can be placed within a gas transport channel, particularly an argon transport channel. A plasma discharge 16 supplied by the generator 10 is generated from this electrode 15, and within it, an electric current flows from the electrode 15 to the biological tissue 17 to which the neutral electrode 14 is connected. This current produces a thermal effect on the biological tissue.

[0033] To provide the device 11, that is, to supply power to the output unit 13, the generator 10 includes a power supply 18, in particular an RF power supply for a high-frequency voltage controllable by a control module 19. This allows the control module 19 to control selected electrical variables, such as modulation, voltage magnitude (amplitude), current intensity, etc.

[0034] For example, the power supply 18 vibrates at a high frequency and has an output section 13 of several thousand volts. peak The sensor is formed by an oscillator that outputs a voltage (peak voltage) and a power of several watts, preferably greater than 10W, for example, 100W. To detect electrical characteristic variables such as voltage or current, the phase angle between voltage and current, and the non-harmonic distortion of current, etc., are useful to the sensor block 20, which provides the measured variables to the control module 19 as indicated by arrow 21. Arrow 21 indicates the direction of the flow of information, and therefore the tip of the arrow is only in one direction. However, it is also possible to configure the embodiment so that the control module 19 specifically requests sensor data and sends a data query to the sensor block 20 for this purpose.

[0035] The control module 19 controls the oscillator 18, which is represented by the arrow 22. Furthermore, the control module 19 can receive information from the oscillator 18 that is independent of the surgical effect obtained on the tissue 17. For example, such information may be about the vibration frequency of the oscillator 18 or its modulation format (e.g., continuous wave (CW), or pulsed, e.g., on / off sampled).

[0036] The control module 19 includes a control dataset created by machine learning from a training dataset. The control dataset establishes a relationship between the strength or effect level of the effect, including the desired penetration depth of the tissue effect, and electrical variables provided by the oscillator 18 and / or sensor block 20. The strength or tissue effect of each effect, including the penetration depth, is preset using an input device 24, which is part of the generator 10 or configured separately from the generator 10 by a portable device such as a tablet or mobile phone. In particular, the input device 24 can also be part of the instrument 11.

[0037] As is clear, the control of the generator 10 is based solely on the effect intensity and / or desired penetration depth of the tissue effect, which are preset by the input device 24, and the electrical variables from the transducer 18 and / or sensor block 20. A camera for inspecting the effect obtained on the biological tissue 17 is not provided, nor is it necessary. Unlike conventional devices, the effect actually obtained on the biological tissue is characterized by the effect intensity or desired penetration depth preset by the input device, and this effect is characterized, for example, by the degree of tissue tanning or a specific penetration depth into the tissue. As soon as the desired effect is obtained, the generator switches to a second mode, and no further action is exerted on the tissue. This is done without requiring a camera image of the treated tissue. For this reason, the desired effect is obtained, but even if the plasma jet is directed to the tissue area for an unnecessarily long time, the desired effect is not exceeded.

[0038] Figure 2 shows that a training dataset 27 is generated, and based on this, a control dataset is then created by the AI ​​module 23. For this purpose, a separate training module 19' is provided. In addition to the variables generated by the sensor block 20, image data v from the camera 25 is supplied to the training module 19', and the camera field of view includes, in particular, the location of the tissue 17 that is affected by the discharge 16. The training module 19' helps in the generation of the training dataset 27. Furthermore, a training input means 24' is provided within the training module 19', through which the person responsible for generating the training dataset 27 can input the obtained effects. For example, if multiple tanning stages of tissue are distinguished, e.g., 10 tanning stages, they can range from no tanning to early carbonization. During training, this degree of tanning is assigned to a pattern of electrical variables generated using the sensor block 20 or received directly from the oscillator 18 as effect labels in the training dataset 27. Furthermore, different tissue types may be set using the training input means 24'. The training module 19' further includes a block 26' that generates and / or stores control signals for the oscillator 18 and the power output of the generator 10 to the instrument 11 associated with the generator 10. The control signals can be assigned to different modes used during the generation of the training dataset 27. In the first mode, the oscillator 18 outputs high power to the instrument 11, resulting in a surgical effect on the tissue 17, such as visible coagulation. In the second, weaker mode, the power output from the oscillator 18 to the instrument 11 is reduced to such an extent that the discharge 16 no longer produces a visible effect on the surface of the tissue 17.

[0039] In Figure 2, the generator 10 comprises an input device 24 and a block 26. Through the input device 24, the person responsible for generating the training dataset 27 can input the effects obtained when the discharge 16 acts on the tissue 17 for testing purposes. The block 26 controls the power output of the generator 10 to the oscillator 18 and, consequently, to the instrument 11.

[0040] Figure 2a shows another example of the generator 10' for generating the training dataset. The explanation given for Figure 2, with reference to the reference numerals, applies mutatis mutandis to the example shown in Figure 2a. The example shown in Figure 2a is substantially different from the example shown in Figure 2 in that the training module 19' is not provided separately from the generator 10'.

[0041] Figure 2b shows another example in which a computed tomography (CT) scanner 25' is used as the imaging device instead of the camera 25. The descriptions relating to Figures 2 and 2a, with reference numerals, apply mutatis mutandis to the example shown in Figure 2b. The CT scanner 25' determines image data v' as measurement data during or after its action on the tissue, from which, among other things, the penetration depth is derived. The penetration depth can have a continuous value or different discrete penetration levels, for example, 10 levels. The determined depth values ​​are associated with electrical variables detected by sensors S1 to S3 in the form of effect labels and incorporated into the training dataset 27. In the example shown, the training module 19' is integrated into the generator 10 as in Figure 2a, but it can also be configured as a separate unit as in Figure 2. Based on this training data, the control module 19 can then be controlled to obtain a predetermined effect depth d accurately but not exceed it.

[0042] Figure 3 illustrates the generation of a training dataset, which includes video sequences or individual pictures from camera 25 and data from sensor block 20. This data can be the outputs g1, g2, g3 of one or more sensors, e.g., sensors S1, S2, S3, characterized by, for example, the magnitude of the current, the magnitude of the phase angle between the current and the voltage, the crest factor of the current, and / or the nonlinearity of the current. Further variables can also be determined; the variables mentioned above are merely examples. In the first embodiment, the variables detected by sensors S1-S3 can only be detected at the end of their operation. Therefore, the variables form a static pattern without a time parameter. However, it is also possible to detect temporal transitions. This is applied to camera 25 and sensor block 20, so that the temporal transitions of each variable can be detected. Thus, the top of Figure 3 shows that data from sensor block 20, camera 25, and input means 24' are combined to generate a training dataset 27. The current state of optical and electrical variables is characterized by data from the camera 25 and sensor block 20, while the obtained results are characterized by effect labels L1...L10 input via the input means 24'.

[0043] To generate the training dataset 27, multiple sample treatment steps are performed in which appropriate specimens are electrically treated by the apparatus shown in Figure 2. From this training dataset 27, the control dataset 26 shown at the bottom of Figure 3 is determined using machine learning. The control dataset 26 includes the desired effect as an input variable (e.g., moderate tanning) via the input device 24. The control dataset 26 easily obtains the respective values ​​that sensors S1 to S3 need to have to obtain the desired effect from the information in the training dataset. The training dataset 27 may include effect labels that characterize the visible tissue change and quantitative parameters such as the penetration depth of the obtained tissue effect. The control dataset 26 here monitors sensors S1 to S4 for the achievement of predetermined values. As a result, the control module 19 operates as follows.

[0044] The oscillator 18 can operate in at least two modes: a first mode with high power to obtain a surgical effect on the tissue 17, and a second mode with lower power to obtain no surgical effect. While generating a training dataset with the apparatus shown in Figure 2, the oscillator 18 first operates in the first mode, then switches off or transitions to the second mode, after which an effect label is defined according to the obtained treatment result. The effect label may include qualitative features such as the degree of visible tissue change (e.g., tanning) and / or quantitative features such as the penetration depth to which the tissue effect was obtained. While actually treating a patient using the apparatus shown in Figure 1, the control module 19 operates, here, in the first mode at the start of the treatment, with the control dataset. The actual tissue alteration (e.g., tanning) obtained, which is not easily recognizable to the surgeon, is shown by the curve 29 at the top of Figure 4. As is clear, tanning increases with time until it reaches a maximum value. The variables of sensors S1, S2, S3 and additional sensors may have different temporal trends. For example, the value g1 of sensor S1 may represent a current that may have a decreasing trend. The value g2 of sensor S2 can be, for example, a humidity value, and tissue humidity may decrease over time as the treatment progresses. The third value g3 of sensor S3 can be any other electrical variable, or a variable characterizing the tissue 17 calculated from the electrical variables.

[0045] Figure 3a shows another training scenario in which image recording is performed using a CT device 25'. The CT device 25' provides image data, from which the penetration depth d of the obtained tissue effect is determined. The depth indication as a component of the effect label is assigned to electrical variables g1 to g3 simultaneously detected by sensors S1 to S3 and input into the training dataset 27. This embodiment can be combined with a training module 19' incorporated into the generator 10, or with a training module embodiment separate from the generator 10.

[0046] In the embodiment of Figure 3, the practitioner pre-set the effect, i.e., the degree of tanning shown as the acceptable range of box 30 in Figure 4. In the embodiment of Figure 3a, the practitioner pre-set each effect depth, i.e., the (maximum) penetration depth of the tissue effect. In the control dataset 26, the values ​​characterized by box 30 are assigned to the transition of characteristics, and consequently to the pattern of electrical variables monitored by sensors S1, S2, and S3 shown in Figure 4. The pattern is determined over time t m The pattern can be a combination of the variables g1, g2, and g3 of sensors S1, S2, and S3 at a single point (measurement time). However, the pattern can also include part or the entire temporal progression of the variables of the three sensors S1, S2, and S3. As soon as this pattern is recognized, the control module 19 reduces the energy output to the device 11 from a high value HIGH to a low value LOW, as shown in Figure 5.

[0047] By using the generator 10 according to the present invention, the treatment of biological tissue with electrosurgical instruments, particularly argon plasma probes, can be reliably performed without depending on the individual skill of the operator. Multiple test treatments of tissue samples are performed using camera-assisted measurement generation, and a training dataset is created based on these. From this training dataset, a control dataset is generated using machine learning, and the control dataset controls the operating room device 10 during subsequent use without the assistance of camera observation of the surgical field. During the training period under camera surveillance, only typical patterns of sensor data assigned to a specific tissue effect are evaluated. This allows the achievement of a preset value to be determined based on the degree of effect and / or penetration depth d.

[0048] The present invention also makes it possible to deliver a thermal effect to tissue very quickly (as quickly as possible) and in a controlled manner (without over-application). For this purpose, the control module 19 operates in treatment phases having two operating conditions (modes): a high-power condition and a low-power condition. The high-power condition (first mode) is used to obtain the desired thermal effect on the tissue. However, the low-power condition (second mode) is not for obtaining a thermal effect. When the plasma that helps to non-contactively transfer power to the tissue is ignited in the first mode and current flows, electrical data is detected that becomes input data for the control module 19 (here, peak voltage Up, peak current I p Effective (root mean square) voltage U rms Effective (root mean square) current I rms (Power factor, frequency, resistance, spark generation). Based on the predictions of the control module 19, the oscillator automatically switches between a first mode and a second mode to immediately limit tissue damage and obtain the desired degree of deactivation. If the tissue damage predicted by the control module 19 is lower than a preset threshold (desired effect), the system 18 remains in the first mode. If the AI ​​prediction exceeds a predetermined effect, the system 18 switches to the second mode to prevent further thermal damage. The second mode is characterized by an output power that is low enough (10W in this case) to prevent further tissue damage, but high enough to enable effective AI prediction of the electrical data obtained in this mode. Thus, the output power is high enough in the second mode to maintain a stable plasma in order to obtain electrical data that forms the basis for determining whether it is necessary to switch back to the first mode.

[0049] In the second mode, the voltage can be pulsed to further reduce energy input and the resulting tissue damage. Pulsing in this case means that in the second mode, the voltage (and current) alternates between on and off periods. During the on period, a specified low power is applied. During the off period, no power is output to the tissue. In this way, the total energy input is significantly reduced during the second mode, thereby further reducing tissue deactivation under these conditions. The sum of the off and on periods can be set to 10 ms. The on period can be set from 2.4 ms to 10 ms. The remaining time is the off period, during which no current is output. If 10 ms is selected as the on period, the off period is 0 ms, and no power reduction occurs, so the low power condition is always the on period. In controlling between high-power and low-power conditions (first and second modes), over-application (of current) must be avoided, and a reproducible, and therefore homogeneous, tissue effect must be obtained. As soon as the preset organizational effect is achieved, the control module 19 switches to a condition with less available power. At this low power level, no significant organizational effect is produced.

[0050] The pure delay time from the point when the measured electrical data exceeds the control module 19's prediction until the condition switch becomes possible is approximately 15 ms. For example, the transient response of the RF power supply 18 adds a further delay time in the millisecond range. In controlling between high-power and low-power conditions, over-application (of current) must be avoided, and a reproducible, and therefore homogeneous, organizational effect must be obtained. As soon as the preset organizational effect is obtained, the control module 19 switches to a condition with less available power. At this low power, no significant organizational effect occurs.

[0051] To train the AI, electrical data is characterized by the degree of deactivation of the tissue surface. For this purpose, images of the treated area are recorded and grouped (labeled) based on the color of the surface. Each electrical data (i.e., the pattern formed by the electrical data) is used to train different groups or labels (which can then be selected as effect levels in the user interface). The color of the treated tissue is selected because it is information available to the physician during the surgical procedure. Depending on the equipment used (a system consisting of an endoscope, image processor, monitor, etc.), the color of the tissue during a particular treatment period may vary from system to system and cannot be compared. However, to ensure that the image color is not tampered with by the equipment used (e.g., camera) and that the results are comparable, specified settings are used during image recording.

[0052] The control module 19 does not require images; only electrical data is used to predict tissue inactivation.

[0053] AI-controlled argon plasma coagulation is merely an example of implementing AI-controlled functions in a module where power output is controlled by a previously trained (immutable) AI algorithm. Training of the control module 19, based on labels where visual data of real tissue is compared with corresponding electrical data, can be used for various modes / displays, e.g., shrinkage (visual) for the quality of thermal melting, degree of deactivation during electrosurgical cutting of tissue (visual coagulation zone, incidental damage), penetration depth (e.g., during ESD), expansion of the RF ablation zone (visual ablation zone), and electrical data during electrosurgical cutting.

[0054] Furthermore, this method is not limited to electrical data and can be extended based on additional data, for example, to control cushion formation in hydroelectric power generation technology, temperature control during RFA, or ice ball formation in refrigeration technology.

[0055] After switching to the low-power second mode, the electrical variables of sensors S1, S2, and S3 may have different values ​​and may change again over time, as shown by the branching of the dotted curve in Figure 4. For example, the current of sensor S1 may initially decrease due to the decrease in power, but may increase again slightly as the tissue becomes wet again over time. Similarly, the variables monitored by the other sensors S2 and S3 may also change again over time. Furthermore, the patterns created in this way can be included in the training dataset, so these patterns can help the control module 19 verify the obtained effects against desired effect labels.

[0056] By using the generator 10 according to the present invention, the treatment of biological tissue with electrosurgical instruments, particularly argon plasma probes, can be reliably performed without depending on the individual skill of the operator. For example, multiple test treatments of tissue samples are performed with the help of image-assisted measurement generation using a camera or medical imaging device such as a CT scanner, and a training dataset is created based on these. From this training dataset, a control dataset is created based on machine learning, and the control dataset controls the device 10 located in the operating room during subsequent use without the help of camera observation of the surgical field. Only typical patterns of sensor data assigned to specific tissue effects during the training period under camera surveillance are evaluated. [Explanation of symbols]

[0057] 10, 10' Generator 11 Equipment 12 lines 13 Output section of generator 10 14 Neutral electrode 15 electrodes 16 discharge 17. Living tissue 18 Power supply 19 AI-equipped control modules 20 Sensor Blocks 21, 22 Arrows 23 AI Modules 24 Input devices 24' Input means 25 Cameras 25' Computed Tomography Scanning Device 26 blocks 27 Training datasets 29 Curve 30 Allowable area box in Figure 4 S1~S3 Sensors g1~g3 Electrical variables e1~e3 Generator input section L1…L10 Effect Labels v Image data v' Image data from CT scanner M Pattern of electrical variables g1~g3 t time t m Measurement point

Claims

1. A control module (19) having input sections (e1, e2, e3) to which only sensors (S1, S2, S3) for electrical variables (g1, g2, g3) are connected, A power supply (18) connected to the control module (19) and controllable by the control module (19), the power supply (18) being connected to the medical device (11) to supply power to the medical device (11), The control module (19) includes a control dataset (26) based on a training dataset (27), the training dataset (27) including image data (v) and electrical variables (g1, g2, g3) detected by the sensors (S1, S2, S3). generator.

2. The control module (19) operates the power supply (18) in either the first mode (HIGH) or the second mode (LOW). The power supply (18) supplies high power in the first mode (HIGH) and low power in the second mode (LOW). The generator according to claim 1.

3. In the first mode (HIGH), the power of the generator (10) is set to a level that achieves inactivation and coagulation of the biological tissue (17), and in the second mode (LOW), the power of the generator (10) is set to a level that avoids inactivation and coagulation of the biological tissue (17) but is sufficient to ensure stable plasma ignition and effective data determination. The generator according to claim 2.

4. The control dataset (26) is created by machine learning based on the training dataset (27) which has image data (v). The generator according to claim 1.

5. The image data (v) is an individual image, image sequence, or video data, and the electrical variables (g1, g2, g3) are a plurality of measurement points determined at time intervals as individual measurement values, or the temporal changes of the electrical variables (g1, g2, g3). The generator according to claim 1.

6. The control dataset (26) includes effect labels (L1...L10) obtained from manual evaluation of tissue test treatment results. The generator according to claim 4.

7. The control dataset (26) includes effect labels (L1...L10) that characterize different tissue inactivation levels and / or different penetration depths of tissue effects. The generator according to claim 5.

8. When the control module (19) recognizes a pattern of the electrical variables (g1, g2, g3) assigned to a desired effect label (L1...L10), it switches from the first mode (HIGH) to the second mode (LOW). The generator according to claim 7.

9. After switching to the second mode (LOW), the control module continues to monitor the pattern (M) of the electrical variables (g1, g2, g3). The generator according to claim 8.

10. The control module (19) switches from the second mode (LOW) to the first mode (HIGH) if the pattern (M) recorded in the second mode (LOW) corresponds to an effect label (L1...L10) that is too low. The generator according to claim 9.

11. The control module (19) switches from the second mode (LOW) to the third mode (OFF) if the pattern recorded in the second mode (LOW) corresponds to an effect label (L1...L10) that is too high. The generator according to claim 9.

12. The control module (19) includes a distance measurement function. The generator according to claim 1.

13. The control module (19) determines the distance between the instrument (11) and the living object (17) based on the electrical variables (g1, g2, g3) detected by the sensors (S1, S2, S3). The generator according to any one of claims 1 to 12.

14. The control module (19) determines the distance based on the nonlinearity of the load which is effective at the output section (13) of the generator (10). The generator according to claim 13.

15. A method for generating a control data set (26) for a control module (19) in an electrosurgical generator (10), and a method for subsequently operating the generator (10), During the training process, the provided or obtained electrical variables (g1, g2, g3) are determined by using the generator (10') with predetermined settings to act on the biological tissue (17), and the resulting tissue changes are recorded by the imaging device (25, 25') to generate a training dataset (27). Effect labels (L1...L10) are assigned to the aforementioned tissue changes. A control dataset (26) is determined from the training dataset (27) using machine learning, and the control dataset (26) represents the relationship between the electrical variables (g1, g2, g3) and the effect labels (L1...L10). During the treatment process, the generator is controlled based on the control dataset (26) in order to obtain the desired effect corresponding to the selected effect labels (L1...L10). method.

Images