CN-121987482-A - Multi-mode sensor fusion-based acupoint accurate positioning system and method

CN121987482ACN 121987482 ACN121987482 ACN 121987482ACN-121987482-A

Abstract

The invention discloses an acupoint accurate positioning system and method based on multi-mode sensor fusion, and relates to the technical field of traditional Chinese medicine diagnosis and treatment. The system integrates ultrasonic imaging, infrared thermal imaging, bioelectrical impedance and flexible pressure touch sensing, and synchronously acquires multidimensional physiological information of the acupoint areas. And outputting three-dimensional coordinates and confidence coefficient of the acupoints by a deep learning cross-mode attention fusion algorithm, wherein the positioning error is less than or equal to 2mm. And (3) guiding a needle to enter when the confidence coefficient is more than or equal to 80% by adopting a Monte Carlo Dropout uncertainty quantization technology, verifying the signal change rate in real time after needling, and confirming that positioning is effective when the comprehensive change rate is more than or equal to 15%, so as to form an intelligent closed loop. The system remarkably improves the positioning precision and repeatability (ICC is more than or equal to 0.95), improves the air rate by 40%, ensures the positioning time to be less than 30 seconds, is suitable for special patients, and promotes the acupuncture diagnosis and treatment to develop to accuracy, standardization and intellectualization.

Inventors

ZHOU QIN
WANG GUANFEN
LUO XI
LI FANG
LIU XIAOYU
ZHOU XIAOFENG
CHEN KEHAN

Assignees

四川省肿瘤医院

Dates

Publication Date: 20260508
Application Date: 20260126

Claims (10)

1. Multi-mode sensor fusion-based acupoint precise positioning system, which is characterized by comprising: the multi-mode sensor module is used for synchronously collecting ultrasonic images, infrared heat maps, multi-frequency bioelectrical impedance spectrums and pressure distribution data of the acupoint areas; the data processing module is used for carrying out feature extraction and fusion on the multi-mode data and outputting three-dimensional coordinates and confidence scores of the acupoints; the three-dimensional modeling and registering module is used for constructing an individualized three-dimensional body surface model of the patient and mapping the multi-mode data to a unified coordinate system; And the verification feedback module is used for retesting the multi-mode signals before and after needling, and verifying the positioning effectiveness through the signal change rate.
2. The system of claim 1, wherein the multi-modality sensor module comprises: The ultrasonic imaging unit has the frequency range of 10-20MHz and the axial resolution of less than or equal to 0.1mm; An infrared thermal imaging unit, the thermal sensitivity is less than or equal to 0.05 ℃, and the spatial resolution is more than or equal to 640 multiplied by 480 pixels; the bioelectrical impedance measuring unit supports 1kHz-1MHz multi-frequency scanning and adopts a four-electrode method; the flexible pressure touch sensor array has the spatial resolution less than or equal to 2mm and supports real-time pressure distribution acquisition.
3. The system of claim 1, wherein the data processing module comprises: The feature extraction submodule adopts a convolutional neural network to respectively extract high-dimensional feature vectors of all modes; The multi-mode fusion sub-module realizes feature weighted fusion based on a cross-mode attention mechanism; the positioning output sub-module outputs three-dimensional coordinates (x, y, z) of the acupoint center and a confidence coefficient P, wherein the confidence coefficient P is calculated based on Bayesian posterior probability, and the formula is as follows: ; Wherein, the For the number of monte carlo samples, For the neural network prediction function, The function is activated for Sigmoid.
4. The accurate acupoint positioning method based on multi-mode sensor fusion is characterized by comprising the following steps of: s1, acquiring three-dimensional body surface point cloud data of a patient, and identifying at least 10 osseous mark points; s2, calculating initial estimated coordinates of the target acupoint based on a bone-based size division method; s3, synchronously collecting ultrasonic, infrared, electrical impedance and pressure data in a region of 3cm multiplied by 3cm around the estimated coordinates; s4, preprocessing and extracting features of the data of each mode; s5, inputting the multi-modal characteristics into a pre-training deep learning model, and outputting an acupoint probability heat map; s6, extracting the maximum probability position as the final acupoint coordinates, and calculating the confidence coefficient; s7, if the confidence coefficient is more than or equal to 80%, guiding an operator to insert a needle; S8, retesting the multi-mode signals after needle insertion, and if the signal change rate is more than or equal to 15%, confirming that positioning is effective.
5. The method according to claim 4, wherein the preprocessing of bioelectrical impedance data in step S4 comprises: Fitting an electrical impedance spectrum based on a Cole-Cole model, and extracting a characteristic frequency fc and a phase angle phi; calculating the impedance difference ratio of the acupoint and surrounding tissue : ; Wherein, the Is the impedance of the point of the acupoint, Is the average impedance of the surrounding area.
6. The method of claim 4, wherein the processing of the infrared thermal imaging data in step S4 comprises: calculating the temperature difference between the target area and the surrounding 5cm annular area : ; Analysis of recovery time constant of temperature recovery curve after pressurization : ; Wherein, the acupoints are The values are typically 1.5-2 times smaller than the surrounding tissue.
7. The method of claim 4, wherein the processing of the pressure haptic data in step S4 comprises: Calculating tissue hardness index : ; Wherein, the In order to maximize the amount of applied pressure, Is the maximum deformation depth; The hardness index at the acupoint is typically 20-40% lower than that of the surrounding tissue.
8. The method according to claim 4, wherein the deep learning model in step S5 adopts a cross-modal attention fusion mechanism, and the attention weight calculation method is as follows: ; Wherein, the , , Respectively representing query, key and value matrix, and deriving from characteristic vectors of different modes.
9. The method of claim 4, wherein the confidence calculation in step S6 is based on an uncertainty quantization method, and the variance of the coordinate prediction is calculated by multiple reasoning with monte carlo Dropout : : Wherein, the For the number of samples to be taken, For a single prediction of the coordinates, Is the average coordinates.
10. The method according to claim 4, wherein the rate of change of signal in step S8 The calculation formula of (2) is as follows: ; Wherein, the And The characteristic values of a certain mode signal (such as impedance, temperature or pressure) before and after needling are respectively obtained.

Description

Multi-mode sensor fusion-based acupoint accurate positioning system and method Technical Field The invention relates to the technical field of traditional Chinese medicine diagnosis and treatment, in particular to an acupoint accurate positioning system and method based on multi-mode sensor fusion. Background Acupuncture is used as an important component of traditional Chinese medicine, and has unique curative effects in the fields of pain management, nerve function rehabilitation, chronic disease conditioning and the like. The World Health Organization (WHO) has officially acknowledged the effectiveness of acupuncture for a variety of diseases and established international positioning regulations including 361 standard acupuncture points. However, the stable exertion of the acupuncture effect is highly dependent on the accurate positioning of the acupuncture points, and the traditional acupuncture point selection method and the modern electronic auxiliary equipment have obvious technical bottlenecks, so that the standardized popularization and clinical repeatability of the acupuncture treatment are severely restricted. The acupoint positioning in the current clinical practice mainly depends on two main methods, namely an empirical body surface positioning method which is inherited for thousands of years, including a bone size dividing method, a body surface marking method and a finger same body size method, and an electronic acupoint apparatus based on an electrical detection principle. The traditional method is highly dependent on experience accumulation and hand feeling judgment of operators, and has remarkable subjectivity and uncertainty. Study data shows that the consistency coefficient (ICC) of different acupuncture operators to locate the same acupoint is only 0.45-0.65, which is far below the clinically acceptable 0.80 threshold. Even if the same physician operates at different times, the positioning error can be as much as 5-10 mm. For obese (BMI > 30), edematous, muscular dystrophy or patients with anatomical variation (about 15-20% of the population), the failure rate of localization in the conventional method is as high as 40-60%. In addition, the deep acupoints (such as ring jump and rank edge, and depth of 8-10 cm) can not be reached through the body surface, and positioning completely depends on rough anatomical inference, so that treatment risks and treatment uncertainty are further increased. In order to overcome the defect of manual positioning, various electronic acupoint detection devices are appeared on the market, and the main technology is a low-frequency bioelectrical impedance measurement method (such as a good vein conduction instrument). This type of device detects impedance changes through skin electrodes to locate an acupoint based on the assumption that the resistance at the acupoint is lower than surrounding tissue. However, the technology has the fundamental limitations that firstly, external factors such as skin humidity, sweat secretion, electrode contact pressure and the like greatly interfere with measurement results, so that false positive rate is as high as 30-40%, the trust degree of clinicians is generally low, secondly, the penetration depth of low-frequency current (usually less than 10 kHz) is limited, the electric characteristics of 2-3 millimeters of the skin surface layer can only be reflected, the true anatomical position and physiological state of deep acupuncture points can not be detected, furthermore, the electric characteristics change can be possibly caused by local compression of non-acupuncture point areas, true acupuncture points and false acupuncture points are difficult to distinguish by equipment, and finally, the equipment such as an electrocardiograph monitor, an electrotome and the like widely existing in a hospital environment can generate electromagnetic interference, so that the stability and accuracy of low-frequency impedance measurement are seriously affected. In the individuation medical age, the prior art system also exposes two major problems of insufficient adaptability and lack of verification mechanism. On the one hand, the standard acupoint coordinate graph is established based on a standard human body model, and the individual differences caused by age (the difference between children and adults can reach 40%), sex, body type (the thickness of subcutaneous fat of an obese person can be increased by 5-10 times), disease state and even body position change are ignored. On the other hand, the existing equipment only provides positioning advice before needle insertion, and cannot carry out real-time objective verification on whether the needle insertion is actually carried out on middle acupuncture points or not and whether the due physiological reaction of qi is caused or not (such as local myoelectric activation and blood flow increase or not). The treatment effect can only depend on subjective description of a patient and retrospect