Noninvasive bacteria detection method and device based on combination of multi-wavelength photoacoustic imaging and deep learning

Abstract

The invention discloses a non-invasive bacteria detection method and device by combining multi-wavelength photoacoustic imaging with deep learning. According to the method, a multi-wavelength photoacoustic imaging system emits adjustable monochromatic nanosecond pulse laser to go deep into a bone joint to obtain hyperspectral photoacoustic high-dimensional map data, an information processing unit analyzes absorption spectrum differences at different three-dimensional positions in the bone joint based on the hyperspectral photoacoustic high-dimensional map data, and a deep learning algorithm is used for determining the position of the bone joint according to the absorption spectrum differences. And qualitatively identifying whether bacteria exist in the bone joints or not, if so, identifying the types of the bacteria, and quantitatively displaying the distribution and concentration of the bacteria. The device comprises a photoacoustic pulse tunable laser, an ultrasonic probe, an electric displacement table, a mechanical fixing device, a water tank, a data acquisition card and an information processing unit. According to the invention, the defects of long time period, low sensitivity and specificity and the like of the conventional examination means are overcome, and a solution which is easy to popularize and apply is provided for intelligent medical diagnosis.

Description

Technical Field

[0001] The present invention belongs to the field of non-invasive medical osteoarthritis detection, and particularly relates to a non-invasive bacteria detection method and device combining multi-wavelength photoacoustic imaging and deep learning. Background Art

[0002] Infection after orthopedic surgery is a common complication that can lead to catastrophic consequences. Previous studies have found that the infection rate after orthopedic joint replacement is close to 2%. According to the statistical data of the US health agency, the per capita cost for the treatment of infection after artificial joint replacement is nearly 100,000 US dollars per year, and even some patients die due to this complication. Once orthopedic infection occurs, it usually requires reoperation, long-term anti-infection treatment, etc., and even the risk of amputation, which has a significant impact on the quality of life of patients. However, there are still many difficulties in achieving effective diagnosis of postoperative infection clinically. At present, direct tissue culture of infection symptoms and clinical pathology detection are recognized as the gold standards for diagnosis by most experts and scholars. Among them, direct tissue culture requires both aerobic and anaerobic cultures, the culture time is at least 5-7 days, and multiple cultures are required, with a low positive detection rate; while clinical pathology detection mainly relies on intraoperative frozen section examination to observe the number of neutrophils in the periprosthetic tissue. However Limitations in the intraoperative sampling site and quality may lead to the inability of pathological examination to accurately identify the inflammatory response caused by some low-virulence pathogens. In recent years, the use of hyperspectral information for bacteria detection can shorten the bacteria classification and identification cycle [1,2] , but it is difficult to meet the depth requirements for osteoarthritis detection: Although an internal detection depth of 3-4 cm can be achieved, it is still unable to effectively detect bacteria related to osteoarthritis. Thus, conventional inspection methods have defects such as long detection cycle, low sensitivity and specificity, and are difficult to be applied in rapid diagnosis. Therefore, developing an early infection diagnosis technology with high sensitivity, specificity, strong multiplexing ability and suitable for large-scale promotion has important clinical value. The pathogen spectrum of deep infection after osteoarthritis surgery is complex, mainly including Gram-positive bacteria and Gram-negative bacteria. The most common pathogenic bacteria are Gram-positive bacteria, especially Staphylococcus aureus, including methicillin-resistant Staphylococcus aureus (MRSA). Other Gram-positive pathogenic bacteria also include Streptococcus spp. and Coagulase-negative staphylococci. In addition, Gram-negative bacteria such as Escherichia coli and Pseudomonas aeruginosa may also be the infecting bacteria.

[0003] Photoacoustic Imaging is a hybrid imaging technique that combines optical illumination and ultrasonic detection, and has achieved rapid development in the field of biomedicine in recent years. Photoacoustic imaging irradiates biological tissues with nanosecond pulsed lasers. After chromophores (such as hemoglobin, melanin, etc.) in the tissues absorb light energy, they generate thermoelastic expansion, and then generate ultrasonic signals. Since the scattering of sound waves in biological tissues is much smaller than that of light, photoacoustic imaging technology can achieve high-resolution imaging within a centimeter-scale depth.

[0004] [1] Zhu, H.; Luo, J.; He, S. Detecting Multiple Mixed Bacteria Using Dual-Mode Hyperspectral Imaging and Deep Neural Networks. Appl. Sci. 2024, 14(4), 1525. [2] Zhu H, Luo J, Liao J, et al. High-Accuracy Rapid Identification and Classification of Mixed Bacteria Using Hyperspectral Transmission Microscopic Imaging and Machine Learning[J]. Progress in Electromagnetics Research, 2023, 178. Summary of the Invention

[0005] In order to overcome the problems in the prior art, the object of the present invention is to provide a non-invasive bacteria detection method and device combining multi-wavelength photoacoustic imaging and deep learning.

[0006] The technical solution for the present invention to achieve its object is as follows: A non-invasive bacteria detection method combining multi-wavelength photoacoustic imaging and deep learning. A tunable monochromatic nanosecond pulsed laser is emitted by a multi-wavelength photoacoustic imaging system and penetrates deep into the bone joint to obtain hyperspectral photoacoustic high-dimensional map data. Based on the hyperspectral photoacoustic high-dimensional map data, the information processing unit analyzes the absorption spectrum differences at different three-dimensional positions inside the bone joint. Based on the deep learning algorithm, it qualitatively identifies whether there are bacteria inside the bone joint. If so, it identifies the types of bacteria and quantitatively displays the distribution and concentration of the bacteria.

[0007] The described monochromatic nanosecond pulsed laser excites the photoacoustic signal of bacteria in bone joints. The photoacoustic pulsed tunable laser is tuned to other wavelengths and irradiates the bone joint tissue again to obtain a hyperspectral photoacoustic high-dimensional map of the bacteria inside the bone joints. Based on the absorption characteristic differences of different types of bacteria for different spectra, the types of bacteria are identified, and the distribution and concentration of bacteria are quantitatively displayed.

[0008] The described deep learning algorithm is based on the CNN convolutional neural network structure. Through the deep convolutional module, high-dimensional features of the light absorption differences of different types of bacteria are extracted, and classification is performed in combination with image features to achieve the classification and distribution localization of bacteria inside the bone joints, as well as the quantitative analysis of the bacteria concentration.

[0009] The described multi-wavelength photoacoustic imaging system includes a photoacoustic pulsed tunable laser, an ultrasonic probe, an electric displacement stage, a mechanical fixing device, a water tank, a data acquisition card, and an information processing unit. The photoacoustic pulsed tunable laser emits monochromatic nanosecond pulsed laser and irradiates the inside of the bone joint tissue. When the laser irradiates the bacteria, the bacteria absorb light energy and generate thermal expansion, and then contract, thus generating an ultrasonic signal. This ultrasonic signal is transmitted through the water medium in the water tank to the ultrasonic probe, and the ultrasonic probe converts the ultrasonic signal into an electrical signal. The electrical signal is transmitted to the information processing unit through the data acquisition card to form photoacoustic data. The information processing unit controls the electric displacement stage to drive the displacement of the mechanical fixing device for two-dimensional photoacoustic imaging. The photoacoustic pulsed tunable laser can tune monochromatic light of different wavelengths, and the photoacoustic signal after laser irradiation is different according to the change of the wavelength of light. The information processing unit processes and forms photoacoustic spectral data. The absorption differences of bacteria in the deep tissue of the bone joint for the excitation light result in different ultrasonic signal propagation times. By calculating the sound velocity, the depth position of the bacteria in the bone joint tissue is obtained, thereby constructing a high-dimensional map of the bacteria distribution inside the bone joints.

[0010] A non-invasive detection device based on multi-wavelength photoacoustic imaging combined with deep learning adopts the described method. The device includes a photoacoustic pulsed tunable laser, an ultrasonic probe, an electric displacement stage, a mechanical fixing device, a water tank, a data acquisition card, and an information processing unit. The described photoacoustic pulsed tunable laser can tune monochromatic light of different wavelengths and emits monochromatic nanosecond pulsed laser to irradiate the inside of the bone joint tissue to generate an ultrasonic signal. The water medium in the described water tank is used to transmit the ultrasonic wave to the ultrasonic probe. The described ultrasonic probe is used to convert the ultrasonic signal into an electrical signal. The described data acquisition card collects the electrical signal and transmits it to the information processing unit to form photoacoustic data. The described information processing unit controls the electric displacement stage to drive the displacement of the mechanical fixing device.

[0011] The beneficial effects of the present invention: The present invention discloses a method for non-invasive detection of deep infection bacteria after osteoarthritis surgery using a multi-wavelength photoacoustic imaging system. By using the multi-wavelength photoacoustic imaging system, it is possible to non-invasively detect the high-dimensional photoacoustic spectra of different types of bacteria in osteoarthritis. Based on the differences in light absorption of different types of bacteria in osteoarthritis at different wavelengths, combined with deep learning algorithms, it can accurately identify the types, concentrations, and distributions of different types of bacteria in osteoarthritis, thus providing an important analysis means for the progression of bacterial infections in osteoarthritis.

[0012] The present invention overcomes the problems of traditional osteoarthritis detection methods. When detecting traditional osteoarthritis, a large surgical window needs to be opened to obtain bacterial samples from joint fluid and then culture the bacteria in a petri dish, which takes 5 to 7 days. Moreover, the accuracy of bacterial classification is low, and the sampling site and quality during the operation are limited, resulting in the inflammation caused by some low-virulence pathogens may not be accurately detected by pathological examination. It can be seen that conventional detection methods have deficiencies such as long time cycles, low sensitivity, and low specificity, and it is difficult to demonstrate their advantages in rapid diagnosis. Therefore, the development of a non-invasive detection device based on multi-wavelength photoacoustic imaging combined with deep learning has important application prospects and value. Brief Description of the Drawings

[0013] Figure 1 It is a schematic structural diagram of a non-invasive detection device based on multi-wavelength photoacoustic imaging combined with deep learning.

[0014] In the figure, there are a photoacoustic pulse tunable laser 1, an ultrasonic probe 2, an electric displacement stage 3, a mechanical fixing device 4, a water tank 5, a data acquisition card 6, and an information processing unit 7. Detailed Embodiments

[0015] The following further elaborates on the present invention in conjunction with the drawings and embodiments.

[0016] A non-invasive bacteria detection method combining multi-wavelength photoacoustic imaging and deep learning. A tunable monochromatic nanosecond pulsed laser is emitted by a multi-wavelength photoacoustic imaging system and penetrates deep into the joint bone to obtain hyperspectral photoacoustic high-dimensional spectral data. The information processing unit analyzes the absorption spectral differences at different three-dimensional positions inside the joint bone based on the hyperspectral photoacoustic high-dimensional spectral data. Based on deep learning algorithms, it qualitatively identifies whether there are bacteria inside the joint bone. If so, it identifies the types of bacteria and quantitatively displays the distribution and concentration of the bacteria. This method can be used to accurately identify the types, concentrations, and three-dimensional spatial distributions of deep infection bacteria after osteoarthritis surgery.

[0017] The described monochromatic nanosecond pulsed laser excites the photoacoustic signal of bacteria in bone joints. The photoacoustic pulsed tunable laser is tuned to other wavelengths and irradiates the bone joint tissue again to obtain the hyperspectral photoacoustic high-dimensional map of bacteria inside the bone joints. Based on the absorption characteristic differences of different types of bacteria for different spectra, the types of bacteria are identified, and the distribution and concentration of bacteria are quantitatively displayed.

[0018] The described deep learning algorithm is based on the CNN convolutional neural network structure. It extracts the high-dimensional features of the light absorption differences of different types of bacteria through the deep convolutional module, and combines the image features for classification to achieve the classification and distribution positioning of bacteria inside the bone joints, as well as the quantitative analysis of the bacteria concentration.

[0019] As Figure 1 shown, a non-invasive detection device based on multi-wavelength photoacoustic imaging combined with deep learning includes a photoacoustic pulsed tunable laser 1, an ultrasonic probe 2, an electric displacement stage 3, a mechanical fixing device 4, a water tank 5, a data acquisition card 6, and an information processing unit 7. The photoacoustic pulsed tunable laser 1 emits monochromatic nanosecond pulsed excitation light and injects it into the bone joint tissue. The bacteria infected with osteoarthritis irradiated by the excitation light absorb the light energy and generate thermal expansion, and then quickly contract, thus generating an ultrasonic signal. The ultrasonic signal is transmitted through the water medium in the water tank 5 to the ultrasonic probe 2 on the mechanical fixing device 4. The ultrasonic probe 2 converts the ultrasonic signal into an electrical signal, and the signal is collected by the data acquisition card 6 and transmitted to the information processing unit 7 to form photoacoustic data. The information processing unit 7 realizes two-dimensional photoacoustic imaging by controlling the displacement movement of the electric displacement stage 3 to drive the mechanical fixing device 4. The photoacoustic pulsed tunable laser 1 can tune monochromatic light of different wavelengths and inject it into the bone joint tissue. Different types of bacteria in osteoarthritis absorb light of different wavelengths, so the generated photoacoustic signal will change due to wavelength tuning, and after being processed by the information processing unit 7, photoacoustic spectral data is formed. There are differences in the time when the bacteria at the depth of the osteoarthritis tissue absorb the excitation light signal and generate an ultrasonic signal that propagates to the ultrasonic probe 2. The depth position of the osteoarthritis bacteria can be calculated through the sound velocity, thereby constructing a high-dimensional map of the distribution of osteoarthritis bacteria.

[0020] Among them, for the described deep learning algorithm, a multi-wavelength photoacoustic imaging system is used to collect the hyperspectral photoacoustic high-dimensional map data inside osteoarthritis. The hyperspectral photoacoustic high-dimensional map data reflects the light absorption difference characteristics of different types of bacteria at different three-dimensional positions inside osteoarthritis. The deep learning algorithm uses a deep neural network based on the CNN convolutional structure. After the deep convolutional module, the high-dimensional features of the light absorption differences of different types of bacteria are extracted, and then combined with the image features to achieve high-accuracy classification of different types of bacteria inside osteoarthritis and three-dimensional classification of quantitatively analyzing the concentrations of different types of bacteria inside.

[0021] Among them, the photoacoustic pulse tunable laser 1 emits a monochromatic wavelength and irradiates it into the osteoarthritis. The photoacoustic signal of the bacteria inside the osteoarthritis is excited. By tuning the photoacoustic pulse tunable laser 1 to other wavelengths and irradiating them into the bone joint, different types of bacteria, including Gram-positive bacteria and Gram-negative bacteria, have different absorption characteristics for light of different wavelengths. Therefore, the ultrasonic signal intensities of different types of bacteria excited by light of different wavelengths are different. Therefore, the photoacoustic pulse tunable laser 1 can be tuned to different wavelengths to detect the hyperspectral photoacoustic high-dimensional map of the bacteria inside the osteoarthritis, and use this to classify the types of bacteria inside the osteoarthritis.

[0022] Example 1: Construction and data acquisition of a multi-wavelength photoacoustic imaging system The photoacoustic pulse tunable laser 1 (wavelength range 680 - 1300 nm, monochromatic light resolution ±2 nm) emits nanosecond-level pulsed laser (repetition frequency 10 Hz, pulse energy ≤20 mJ / cm²) to penetrate the surface of the bone joint tissue. Since different types of bacteria (such as Gram-positive bacteria and Gram-negative bacteria) have significantly different light absorption characteristics for specific wavelengths of light (for example, Staphylococcus aureus has a characteristic absorption peak at 720 nm, while Escherichia coli has strong absorption at 850 nm), different bacteria can be excited by wavelength tuning. After the laser penetrates the tissue, the light energy absorbed by the bacteria is converted into heat energy, causing local thermal expansion and generating broadband ultrasonic signals (frequency range 1 - 15 MHz). The ultrasonic signals are coupled to the ultrasonic probe 2 through the deionized water in the water tank 5. Traditional ultrasonic imaging only relies on the acoustic characteristics of tissues and is difficult to distinguish the types of bacteria. However, in this invention, by exciting the specific light absorption of bacteria with multi-wavelength lasers and combining high-sensitivity ultrasonic probes, the spectral characteristics of bacteria in deep tissues (up to 5 cm) are captured for the first time, breaking through the limitations of single-wavelength imaging.

[0023] Example 2: High-dimensional map construction and depth localization The information processing unit 7 receives the ultrasonic signals through the data acquisition card 6 and synchronously controls the electric displacement stage 3 to drive the ultrasonic probe 2 for two-dimensional scanning. By measuring the time difference of ultrasonic signal propagation and combining the sound speed (about 1500 m / s in the water medium), the depth position of the bacteria is calculated. Further, the photoacoustic pulse tunable laser 1 switches wavelengths in 5 nm steps, and three-dimensional photoacoustic signals are collected at each wavelength, and finally fused to form a hyperspectral photoacoustic high-dimensional map. Beneficial effects: Traditional photoacoustic imaging is mostly single-wavelength or broadband imaging. In this invention, through fine wavelength tuning and spatio-temporal signal fusion, a high-dimensional map (spatial + spectral dimension) of bacteria distribution is constructed for the first time, providing high-dimensional feature data for subsequent deep learning classification.

[0024] Example 3: Bacteria classification and concentration quantification based on CNN The information processing unit 7 adopts a pre-trained deep convolutional neural network (CNN), and its structure includes: an input layer: receiving high-dimensional spectral slices; a feature extraction module: convolution + max pooling; a classification head: a fully connected layer outputs the types of bacteria (such as Staphylococcus, Streptococcus, etc.) and the relative concentration (0-100%); a three-dimensional reconstruction module: a deconvolution layer fuses multi-wavelength features to generate a three-dimensional heat map of bacterial concentration. Traditional methods rely on culture or PCR detection, which is time-consuming and unable to locate. However, the present invention fuses spectral and spatial features through CNN, and for the first time realizes non-invasive and real-time bacterial species identification and three-dimensional quantification, providing a new tool for dynamic monitoring of postoperative infections.

[0025] Example 4: Clinical application and disease course analysis Place the joint area (such as the knee joint) of the patient in the water tank 5. After obtaining the high-dimensional spectrum through multi-wavelength scanning, the information processing unit 7 automatically marks the infected area and outputs the distribution curve of bacterial concentration with depth. For example, an increase in the concentration of Staphylococcus epidermidis (characteristic wavelength 780 nm) in the subchondral bone area (depth 2.1 cm) was detected within 2 weeks after surgery, indicating early deep infection and guiding precise antibiotic intervention. Compared with traditional X-rays or MRIs that can only show structural changes, the present invention can detect subclinical infections 1-2 weeks earlier through dynamic high-dimensional spectra, significantly improving the treatment effect. In summary, through the deep integration of multi-wavelength photoacoustic imaging and deep learning algorithms, the present invention has overcome the problem of non-invasive detection of deep infection bacteria in osteoarthritis, and is significantly superior to the prior art in terms of specificity, sensitivity, and three-dimensional localization ability, providing a new solution for intelligent medical diagnosis.

[0026] The implementation schemes described above can be further combined or replaced. Moreover, the implementation schemes only describe the preferred embodiments of the present invention, and do not limit the concept and scope of the present invention. Without departing from the design idea of the present invention, various changes and improvements made by those of ordinary skill in the art to the technical solutions of the present invention all belong to the protection scope of the present invention. The protection scope of the present invention is given by the appended claims and any equivalent technical solutions.

Claims

1. A non-invasive bacterial detection method combining multi-wavelength photoacoustic imaging with deep learning, characterized in that: Through the multi-wavelength photoacoustic imaging system, an adjustable monochromatic nanosecond pulse laser is emitted deep into the bone joint to obtain hyperspectral photoacoustic high-dimensional atlas data. The information processing unit analyzes the difference in absorption spectra at different three-dimensional positions inside the bone joint based on the hyperspectral photoacoustic high-dimensional atlas data, and qualitatively identifies whether there are bacteria in the bone joint based on the deep learning algorithm. If so, the type of bacteria is identified, and the distribution and concentration of the bacteria are quantitatively displayed.

2. The method according to claim 1, characterized in that: The monochromatic nanosecond pulse laser excites the photoacoustic signal of bacteria in the bone joints, which is then tuned to other wavelengths by a photoacoustic pulse tunable laser and irradiated again to obtain a high-spectral photoacoustic high-dimensional map of the bacteria inside the bone joints. The types of bacteria are identified based on the differences in the absorption characteristics of different types of bacteria for different spectra, and the distribution and concentration of the bacteria are quantitatively displayed.

3. The method according to claim 1, characterized in that: The deep learning algorithm is based on the CNN convolutional neural network structure. The high-dimensional features of light absorption differences of different types of bacteria are extracted through the deep convolution module, and classified in combination with image features to achieve the classification and distribution positioning of bacteria inside bone joints, as well as quantitative analysis of bacterial concentrations.

4. The method according to claim 1, characterized in that: The multi-wavelength photoacoustic imaging system comprises a photoacoustic pulse tunable laser (1), an ultrasonic probe (2), an electric displacement stage (3), a mechanical fixing device (4), a water tank (5), a data acquisition card (6) and an information processing unit (7); the photoacoustic pulse tunable laser (1) emits monochromatic nanosecond pulse laser light to be incident on the inside of bone joint tissue; when the laser light irradiates bacteria, the bacteria absorb light energy to generate thermal expansion and then contract, thereby generating an ultrasonic signal; the ultrasonic signal is transmitted to the ultrasonic probe (2) through the water medium in the water tank (5); the ultrasonic probe converts the ultrasonic signal into an electrical signal, and the electrical signal passes through the data acquisition card (6) The information processing unit (7) is used to form photoacoustic data; the information processing unit (7) controls the electric displacement stage (3) to drive the displacement of the mechanical fixing device (4) to perform two-dimensional photoacoustic imaging; the photoacoustic pulse tunable laser (1) can tune monochromatic light of different wavelengths, and the photoacoustic signal after laser irradiation varies according to the wavelength of the light, and the information processing unit (7) processes and forms photoacoustic spectrum data; the difference in the absorption of the excitation light by bacteria deep in the bone joint tissue leads to the difference in the propagation time of the ultrasonic signal, and the depth position of the bacteria in the bone joint tissue is obtained by calculating the sound speed, thereby constructing a high-dimensional map of the distribution of bacteria in the bone joint.

5. A non-invasive detection device based on multi-wavelength photoacoustic imaging combined with deep learning, characterized in that: The method according to claim 1 is adopted, wherein the device comprises a photoacoustic pulse tunable laser (1), an ultrasonic probe (2), an electric translation stage (3), a mechanical fixing device (4), a water tank (5), a data acquisition card (6) and an information processing unit (7); The photoacoustic pulse tunable laser (1) can tune monochromatic light of different wavelengths and emit monochromatic nanosecond pulse laser to be incident on the inside of bone joint tissue to generate ultrasonic signals; The water medium in the water tank (5) is used to transmit ultrasound to the ultrasound probe (2); The ultrasonic probe (2) is used to convert ultrasonic signals into electrical signals; The data acquisition card (6) acquires the electrical signal and transmits it to the information processing unit (7) to form photoacoustic data; The information processing unit (7) controls the electric displacement platform (3) to drive the displacement of the mechanical fixing device (4).

Images