Search

CN-121971162-A - Skin laser treatment auxiliary robot system based on artificial intelligence computer vision

CN121971162ACN 121971162 ACN121971162 ACN 121971162ACN-121971162-A

Abstract

The invention relates to the technical field of skin laser treatment auxiliary robot systems based on artificial intelligence computer vision, and particularly discloses a skin laser treatment auxiliary robot system based on artificial intelligence computer vision. The system comprises a hyperspectral vision acquisition device, a skin biochemical characteristic analysis unit, a physical information neural network prediction engine, a laser parameter optimization controller and a laser execution mechanism, wherein the hyperspectral vision acquisition device is used for acquiring multi-layer biochemical-optical characteristics of skin, combining with a generation type artificial intelligent model embedded with physical priori to predict three-dimensional thermal field distribution in real time, dynamically optimizing laser pulse width, light spot diameter and energy density, and realizing cooperative control of focus accurate treatment and epidermis safety protection. Through the technical scheme, the subcutaneous thermal injury process can be visually predicted, and the accuracy, safety and clinical adaptability of laser treatment are improved.

Inventors

  • Liang Diequn

Assignees

  • 梁轶群

Dates

Publication Date
20260505
Application Date
20260318

Claims (10)

  1. 1. An artificial intelligence computer vision-based skin laser therapy auxiliary robot system, comprising: the hyperspectral vision acquisition device is configured to perform non-contact scanning on a target skin area, acquire hyperspectral image data and generate a spatial distribution map reflecting biochemical components of skin tissues; The skin biochemical characteristic analysis unit is in communication connection with the hyperspectral vision acquisition device and is configured to perform spectrum unmixing treatment on the hyperspectral image data, extract tissue optical characteristic parameters and construct a biochemical-optical combined characterization model of the multilayer skin structure; The physical information neural network prediction engine is coupled with the skin biochemical characteristic analysis unit and is configured to call a generated artificial intelligent model fused with physical constraints based on the biochemical-optical combined characterization model to generate a three-dimensional thermal field distribution virtual image for characterizing energy deposition and thermal effect evolution in the tissue in real time; The laser parameter optimization controller is connected with the physical information neural network prediction engine, and is configured to dynamically calculate optimal laser emission parameters meeting a preset safety threshold and a treatment threshold according to the three-dimensional thermal field distribution virtual image and output a control instruction; And the laser executing mechanism responds to the control instruction of the laser parameter optimizing controller and is configured to emit laser beams according to the optimal laser emission parameters, irradiate the target focus area and carry out closed-loop parameter adjustment according to the real-time thermal response.
  2. 2. The artificial intelligence computer vision-based skin laser therapy assisted robotic system of claim 1, wherein the hyperspectral vision acquisition device comprises: The wide-spectrum illumination light source module consists of a plurality of light-emitting diode arrays with different center wavelengths, or is provided with a halogen tungsten lamp matched with a light filtering component with continuous spectrum output, and the emitted spectrum range of the light source module covers a wavelength range of 400-2500 nanometers; The light splitting component adopts a transmission type grating structure or an acousto-optic tunable filter structure and is configured to carry out dispersion treatment on skin reflected light, and the spectral resolution of the light splitting component is configured to be capable of distinguishing the characteristic absorption peak difference of oxyhemoglobin and deoxyhemoglobin at the vicinity of 540 nanometers and 576 nanometers; the high-sensitivity image sensing array adopts a charge coupled device or a complementary metal oxide semiconductor sensor with a refrigerating function and is used for capturing a spectral image after light splitting and converting the spectral image into a digital signal; The calibration and correction module is configured to carry out distortion restoration on the image through built-in geometric correction logic, and carry out normalization processing on the acquired original radiation brightness by utilizing preset standard whiteboard reflection data so as to obtain an absolute reflectivity image of the target skin area; the hyperspectral vision acquisition device adopts a push-broom type scanning mode, and a scanning head is driven by a stepping motor to perform uniform linear motion on the surface of the skin, so that a three-dimensional hyperspectral data cube with two-dimensional coordinate space dimensions and one-dimensional wavelength information dimensions is constructed.
  3. 3. The artificial intelligence computer vision-based skin laser therapy assisted robot system of claim 1, wherein the skin biochemical feature analysis unit comprises: The spectrum unmixing module is configured to decompose the reflection spectrum of each pixel point into the contribution sum of various biochemical components by adopting a nonlinear optimization algorithm based on a least square method based on a preset skin tissue optical model, and invert the concentration distribution of melanin, the total amount of hemoglobin and the blood oxygen saturation in skin tissues by calculating the matching degree of a reflectivity curve and the known pure substance absorption spectrum; the image segmentation module is configured to run a multi-scale convolutional neural network, identify skin texture, pore distribution and edges of a focus area, and divide the image into a epidermis area, a blood vessel area and an abnormal pigmentation area; The model construction module is configured to convert the extracted biochemical parameters into absorption coefficients, scattering coefficients and anisotropic factors for describing the photophysical characteristics of skin, and establish a digital skin model with a three-layer structure, wherein the first layer is an epidermis layer, the optical characteristics of which are determined by melanin content, the second layer is a dermis shallow layer, the characteristics of which are influenced by the hemoglobin distribution of a capillary network, and the third layer is a dermis deep layer, and larger blood vessels and collagen fibers are involved; the model construction module is further configured to automatically adjust inversion strategies according to different skin types, and ensure consistency of biochemical feature extraction in different populations by increasing compensation coefficients of melanin absorption for dark skin types.
  4. 4. The artificial intelligence computer vision based skin laser treatment assisted robotic system of claim 1, wherein the physical information neural network prediction engine has integrated therein a generated artificial intelligence architecture comprising: The generator part adopts a residual network structure and is configured to be input by taking the biochemical-optical joint characterization model as a condition, and generates three-dimensional voxel thermal field distribution data through multi-layer deconvolution operation, wherein the thermal field distribution data characterize the temperature value of each space coordinate point in the tissue at different time points after laser irradiation; A physical constraint operator is embedded in a loss function of the physical information neural network, and the evolution process of the generated temperature field along with time is required to meet a thermal diffusion rule, namely, the time derivative of the local temperature is required to have linear correlation with the second derivative of the temperature space distribution and the locally absorbed heat source intensity; A discriminator section configured to perform countermeasure training by a historical clinical thermal injury sample, perform true-false discrimination on the virtual image generated by the generator section, and cause the generated three-dimensional thermal field distribution to conform to a physical law and to be close to a real clinical observation; the physical information neural network prediction engine is configured to predict transient temperature fluctuation generated by a photo-thermal effect inside the skin under the action of laser pulses in inference time by introducing a physical loss term based on a partial differential equation in a training stage.
  5. 5. The artificial intelligence computer vision based skin laser treatment assist robot system of claim 4 wherein the physical information neural network prediction engine further comprises: A light transmission simulation subroutine configured to synchronously perform a light transmission simulation based on a monte carlo method, discretize a skin model into fine volume units, and simulate random walk trajectories of a plurality of virtual photons in the volume units when generating a three-dimensional thermal field distribution; in the light transmission simulation subprogram, the motion path of each virtual photon is determined by probability distribution generated by a random number generator, and the step length and the scattering direction are regulated by the scattering coefficient and the anisotropic factor; And the energy deposition calculation module is configured to calculate the energy deposition amount according to the absorption coefficient of a certain volume unit when the virtual photon passes through the unit, and the physical information neural network is configured to learn the random simulation result and convert the random simulation result into a deterministic heat source distribution function, and predict the transient evolution process of the tissue temperature at different time points based on the space accumulation effect of photon deposition energy.
  6. 6. The artificial intelligence computer vision based skin laser treatment assist robot system of claim 5 wherein the laser parameter optimization controller is configured with dual-objective dynamic optimization logic comprising: a first target optimizing unit configured to protect the epidermis layer, requiring that a highest temperature at an interface of the epidermis and the dermis does not exceed a preset safety threshold, in order to prevent blisters or scars caused by thermal damage; A second target optimizing unit configured to ensure a treatment effect, requiring an accumulated thermal integration value of a target lesion area to reach or exceed a preset treatment threshold to induce thermal coagulation or disruption of a lesion tissue; A weighted solution module configured to combine the requirements of the first target optimization unit and the second target optimization unit into a scalar optimization function by a weighted function, wherein the weight of the epidermal safety target is set to be higher than the weight of the therapeutic effect target so as to embody a safety priority principle; And the parameter searching module is configured to search a global optimal solution in a multidimensional parameter space formed by pulse width, light spot diameter, energy density and pulse frequency through a cyclic iterative algorithm, and generate a binary control instruction conforming to a communication protocol.
  7. 7. The artificial intelligence computer vision based skin laser treatment auxiliary robot system of claim 6, wherein the laser actuator comprises: a laser generator module configured to output laser pulses of a specific wavelength according to a control instruction; The beam transmission and shaping system consists of a multi-axis mechanical arm, a reflecting mirror group and a zoom lens and is configured to guide laser to a target position and adjust the size of a light spot; The real-time feedback adjustment module is configured to continuously receive a feedback signal from the hyperspectral vision acquisition device in the laser emission process, compare the actual thermal response with the predicted thermal field, and trigger a parameter fine adjustment mechanism or a safety fusing mechanism to reduce energy output or terminate laser emission if the actual temperature rise speed exceeds a predicted value or the deviation exceeds a preset tolerance; Extracting the total energy value of laser pulse, obtaining the round area value corresponding to the diameter of the light spot, and dividing the total energy value by the round area value; the calculation logic of the thermal integration is to calculate the sum of the products of the temperature values and the differences between the reference temperatures at each time step over a predetermined period of action.
  8. 8. The skin laser treatment auxiliary robot system based on artificial intelligence computer vision according to claim 7, wherein a mechanical arm control part of the laser executing mechanism adopts a precise industrial robot structure with 6 degrees of freedom, a uniform world coordinate system is established between the laser executing mechanism and the hyperspectral vision acquisition device, the mechanical arm control part is configured to adjust the pose of the end effector in real time according to coordinate information identified by the vision acquisition device by executing a hand-eye calibration algorithm, ensure that a laser beam is always perpendicular to the skin surface to irradiate so as to reduce energy loss caused by Fresnel reflection, a pressure sensor and a distance sensor are arranged on the mechanical arm, the mechanical arm control part is configured to maintain a constant distance between the end effector and the skin surface in a scanning process, ensure stability of a laser irradiation focal plane, and the zoom lens group is configured with a piezoelectric driver with high frequency response and is configured to rapidly switch the spot size in the same pulse sequence, so that high-energy focusing of a core area and low-energy transitional irradiation of an edge area are realized.
  9. 9. The artificial intelligence computer vision based skin laser treatment assist robot system of claim 8, further comprising: The anomaly detection sub-module is configured to perform dimension reduction analysis on the acquired spectral data by using a self-encoder network, mark the spectral features of the region as a suspected high-risk region if the spectral features deviate from a preset normal skin spectral distribution library, and require the laser parameter optimization controller to apply more severe safety boundary conditions to the region; A time series prediction module configured to predict a cooling time of the skin tissue according to a heat accumulation effect under continuous irradiation of the laser, and to optimize a repetition frequency of the laser pulses accordingly, ensuring that a time interval between two continuous pulses is sufficient for the epidermis layer to emit heat by thermal conduction, while a temperature of the dermis layer is maintained at a therapeutic level; and the hardware monitoring loop is logically independent of the software control layer, is directly connected to the power supply end of the laser generator module, measures the total joule number and the pulse width of laser output in real time through a physical sensor, and cuts off the power supply through hardware interlocking if the output energy is detected to exceed the preset safety limit value.
  10. 10. The artificial intelligence computer vision based skin laser treatment auxiliary robot system of claim 9, wherein the system employs a distributed architecture based on edge computing in conjunction with cloud computing, comprising: The on-site acquisition terminal integrates the hyperspectral vision acquisition device and the image compression module, and the image compression module compresses hyperspectral data streams by using a sparse representation algorithm and transmits the hyperspectral data streams through a wireless link; The cloud computing cluster is used for carrying out physical information neural network prediction engine and running high-precision Monte Carlo simulation examples and ultra-high-resolution three-dimensional thermal field evolution simulation, and sending the calculated optimal parameter set to the edge processing gateway through an encryption channel; the cloud computing cluster is further configured to maintain a global skin optical characteristic database, and perform incremental learning on the physical information neural network prediction engine by using anonymized clinical feedback results to continuously optimize the prediction accuracy of skin thermal responses of different skin colors and anatomical parts.

Description

Skin laser treatment auxiliary robot system based on artificial intelligence computer vision Technical Field The invention belongs to the technical field of crossing of medical robots and artificial intelligence, and particularly relates to a skin laser treatment auxiliary robot system based on artificial intelligence computer vision. Background Along with the wide penetration of artificial intelligence technology in the field of medical auxiliary robots, the identification of skin lesions based on computer vision and the precision of laser treatment have become important directions for the development of skin medicine. Modern medical systems can assist physicians in completing a complex series of operations from lesion localization to treatment parameter setting by integrating image sensors with intelligent algorithms. In high-precision skin laser surgery, a robot system needs to sense physiological characteristics and pathological distribution of skin tissues in real time, and provides scientific decision support for the emission of laser energy, so that huge application potential is shown in the aspects of improving treatment effect and reducing surgery risk. The subcutaneous injury prediction technology based on hyperspectral imaging and generative physical modeling is a core for assisting a robot in realizing intelligent treatment. The technology aims at acquiring biochemical indexes and optical parameters of deep skin tissues by a non-invasive visual acquisition means, and simulating the propagation track and energy conversion process of laser beams in the multi-layer skin tissues by using a calculation model. By converting the abstract physical interaction process into visual three-dimensional thermal field distribution, the system can predict the accumulated influence of different laser parameters on epidermis and dermis tissues, and realize the fine regulation and control of the treatment depth and the damage range. The computer vision system in the prior art is mostly limited to two-dimensional extraction of pigment characteristics on the surface of the skin, and is difficult to accurately invert the scattering and absorption dynamic process of photons in the skin, so that the energy matching between epidermis protection and dermis focus removal is unbalanced. The traditional analysis method lacks the capability of dynamically embedding multi-source heterogeneous biochemical data, cannot realize the accurate association of data and physical semantics in a complex skin tissue environment, and is difficult to break through the physical limitation that vision cannot penetrate the skin surface layer. The existing energy prediction logic often depends on linear simplified analysis, and cannot capture nonlinear thermal damage state evolution in the skin under extremely short pulses in real time, so that a prediction model is deviated from a real clinical thermal effect. Accordingly, a skin laser treatment assisted robotic system based on artificial intelligence computer vision is desired. Disclosure of Invention The invention aims to provide a skin laser treatment auxiliary robot system based on artificial intelligence computer vision, which can solve the problems in the background technology. In order to achieve the above purpose, the technical scheme adopted by the invention is as follows: The utility model provides a skin laser treatment auxiliary robot system based on artificial intelligence computer vision, includes hyperspectral vision collection device, skin biochemical characteristics analysis unit, physical information neural network prediction engine, laser parameter optimization controller and laser actuating mechanism, wherein: The hyperspectral vision acquisition device is configured to perform non-contact scanning on a target skin area, acquire hyperspectral image data covering visible light to near infrared wave bands, and generate a spatial distribution map reflecting the concentration of hemoglobin, the distribution of melanin and the oxygen saturation in skin tissues based on the data; The skin biochemical characteristic analysis unit is in communication connection with the hyperspectral vision acquisition device and is configured to perform image segmentation and spectrum unmixing treatment on the hyperspectral image data, extract tissue optical characteristic parameters of an epidermis layer, a dermis superficial layer and a dermis deep layer, including absorption coefficients, scattering coefficients and anisotropic factors, and construct a biochemical-optical combined characterization model of a multilayer skin structure; The physical information neural network prediction engine is coupled with the skin biochemical characteristic analysis unit and is configured to receive the biochemical-optical combined characterization model as input, call a pre-trained generation type artificial intelligent model, and combine physical priori knowledge of Monte Carlo light transmission simulation