Miniature fluorescence microscopic imaging and automatic counting system for onychomycosis detection

Abstract

The invention belongs to the technical field of medical instruments, and particularly discloses a miniature fluorescence microscopic imaging and automatic counting system for onychomycosis detection. The system comprises an integrated handheld sampling imaging terminal, a cloud intelligent analysis server and a user interaction terminal. The terminal integrates sampling, fluorescence labeling and microscopic imaging functions, automatically completes sample processing and image acquisition, and uploads the sample processing and image acquisition to the cloud. And the server performs automatic identification, segmentation and counting of fungal targets on the image through a deep learning algorithm to generate a detection report. According to the invention, mechanical sampling, microfluidic fluorescence labeling and miniature fluorescence microscopic imaging are highly integrated in the handheld terminal, a closed and automatic sample processing and imaging assembly line is constructed, and the problem of leak detection caused by insufficient sampling depth and inconsistent dyeing operation in a traditional method is solved.

Description

Technical Field

[0001] This invention belongs to the field of medical device technology, specifically relating to a miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis. Background Technology

[0002] In the field of medical diagnostics and testing, microscopic imaging technology is a core tool for observing and analyzing microscopic targets such as cells and microorganisms, and its development has driven progress in pathology, microbiology, and clinical testing. Fluorescence microscopy, due to its high sensitivity and specificity, plays a crucial role in pathogen detection, enabling precise identification and localization of targets by labeling them with specific fluorescent substances.

[0003] The etiological diagnosis of onychomycosis (nail fungus) is an important technical area in dermatology and laboratory medicine. Onychomycosis is caused by dermatophytes, yeasts, or non-dermatophyte molds. Accurate and rapid diagnosis is crucial for guiding clinical medication, controlling infection, and evaluating treatment efficacy. Traditional diagnostic methods aim to identify and confirm the presence of fungi from nail shavings samples.

[0004] Current technologies mainly rely on methods such as KOH wet mount microscopy, fungal culture, and immunofluorescence. While KOH wet mount microscopy is simple to perform, its positive rate is highly dependent on sample quality, making it prone to missed detections in early infections or when the fungal load is low. Furthermore, the samples often contain background impurities, and manual microscopic interpretation is highly subjective, resulting in poor consistency of results. Although fungal culture methods are more reliable, they are time-consuming, taking several weeks, and have stringent requirements for culture conditions, failing to meet the clinical needs for rapid screening and immediate diagnosis.

[0005] Immunofluorescence technology offers high sensitivity, but its reagents and equipment are expensive, its operation is complex, and it heavily relies on specialized laboratory environments, making it difficult to widely apply in primary healthcare institutions, communities, or home settings. These methods generally suffer from technical bottlenecks such as inaccurate sampling leading to high false negative rates, lengthy testing cycles, and over-reliance on professional personnel and laboratory conditions, posing a serious challenge to the early, rapid, and convenient diagnosis of onychomycosis. Summary of the Invention

[0006] The purpose of this invention is to provide a miniature fluorescence microscopy imaging and automatic counting system for onychomycosis detection, so as to solve the technical contradictions in existing onychomycosis detection technologies, such as high false negative rates due to inaccurate sampling, long detection cycles, and excessive reliance on professional personnel and laboratory conditions.

[0007] This invention provides a miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis, comprising:

[0008] An integrated handheld sampling and imaging terminal is used to perform physical collection, fluorescent labeling, microscopic imaging, and uploading of raw image data of nail dander samples;

[0009] A cloud-based intelligent analysis server is used to receive and process image data from the integrated handheld sampling and imaging terminal, and to perform fungal target identification and automatic counting analysis.

[0010] User interaction terminal, used to display test result reports to users;

[0011] The cloud-based intelligent analysis server is equipped with a fungal intelligent identification and counting algorithm based on a deep convolutional neural network. It first preprocesses the received raw fluorescence image.

[0012] The preprocessing includes background noise subtraction based on dark field images, non-uniform illumination correction, and contrast-limited adaptive histogram equalization enhancement. The preprocessed image is then input into a specially trained fungal target detection neural network model.

[0013] The model outputs a probability map of each pixel belonging to the fungal target category;

[0014] The probability map is binarized with a fixed threshold to generate a preliminary fungal target mask. Then, connected component analysis based on 8 neighborhoods and morphological post-processing are performed to calculate the morphological features of each connected component and filter out false targets according to a preset threshold. The confirmed valid fungal targets are statistically analyzed to obtain the total number of hyphae and spores, and the fungal density per unit field of view is calculated. The fungal load of the entire sampling area is estimated by combining the total area of ​​multiple spliced ​​fields of view. A digital detection report containing the total number of fungi, density distribution, typical target marker images, and infection degree assessment is automatically generated.

[0015] Preferably, the integrated handheld sampling and imaging terminal includes a mechanical sampling module, a microfluidic fluorescence labeling module, a miniature fluorescence microscopy imaging module, a main control and communication module, and a power supply module;

[0016] The mechanical sampling module is used to perform standardized nail dandruff sample collection actions, including a replaceable disposable sampling head. The sampling head has a micro-scraping structure at the front end, which is made of biocompatible material. The scraping angle and pressure threshold are controlled by preset mechanical limits and elastic elements. The sampling head integrates a micro-sample storage cavity.

[0017] The microfluidic fluorescent labeling module is connected to the sample storage chamber of the mechanical sampling module through a microchannel. It is used to perform automated fluorescent staining on the collected nail shavings samples. The module is pre-encapsulated with dried lyophilized fluorescent staining reagent powder. The fluorescent staining reagent is a specific fluorescent dye with high affinity for fungal chitin or ergosterol. The microfluidic fluorescent labeling module also includes a micro piezoelectric pump and an independent buffer chamber, which are used to quantitatively inject buffer into the microchannel to dissolve the lyophilized powder and form a staining liquid stream. The staining liquid stream is mixed and incubated with nail shavings fragments under preset temperature control conditions.

[0018] The micro-fluorescence microscopy imaging module is used to acquire fluorescence microscopic images of stained samples. It adopts an inverted fluorescence optical path design and includes an excitation source, an excitation filter, an objective lens, a dichroic mirror, an emission filter, and a high-sensitivity complementary metal-oxide-semiconductor image sensor. The objective lens is a miniaturized fixed-focus lens with a high numerical aperture. The entire optical module is fixed on a three-dimensional nano-displacement platform driven by a piezoelectric ceramic actuator.

[0019] The main control and communication module includes an embedded microprocessor and a wireless communication unit. The embedded microprocessor is used to coordinate and control the operation timing of the mechanical sampling module, the microfluidic fluorescence labeling module and the micro fluorescence microscopy imaging module. The wireless communication unit is used to encrypt and upload the acquired raw fluorescence image data, sampling position encoding information and device identification code to the cloud intelligent analysis server.

[0020] The power supply module provides power to all electronic components within the integrated handheld sampling imaging terminal.

[0021] Preferably, the user interaction terminal is a dedicated application installed on a smart mobile device, which establishes a secure connection with the cloud-based intelligent analysis server through authorization and authentication, receives and visualizes the digital test report, and provides historical data query and management functions.

[0022] Preferably, the fungal target detection neural network model adopts an encoder-decoder architecture. The encoder part uses a residual network structure to extract multi-scale feature maps of the input image, and the decoder part fuses deep semantic information and shallow detail information through upsampling and skip connections.

[0023] Preferably, the training process of the fungal target detection neural network model uses a large-scale labeled nail clipping fluorescence microscopic image dataset. During training, the cross-entropy loss function and the focus loss function optimized for small target detection are used, and an online hard example mining strategy is adopted.

[0024] Preferably, the morphological features of the connected domains calculated in the morphological post-processing include area, perimeter, roundness, mean and variance of fluorescence intensity, and the preset threshold range for filtering out false targets is set for area, shape regularity and fluorescence intensity distribution.

[0025] Preferably, the estimation process for the fungal load is as follows: Count all effective fungal targets to obtain the total number. Calculate the total area of ​​all stitched fields of view. Fungal density The calculation formula is Based on clinical diagnostic criteria, fungal density is classified into different infection levels.

[0026] Preferably, the fungal target detection neural network model of the cloud-based intelligent analysis server is compressed using knowledge distillation technology to obtain a lightweight student model. The lightweight student model is deployed in the main control module of the integrated handheld sampling and imaging terminal to perform local real-time preliminary analysis and target pre-screening in an environment without network connection.

[0027] Preferably, the integrated handheld sampling imaging terminal caches the pre-screening results of the lightweight student model along with the original image data, and uploads them to the cloud-based intelligent analysis server after the network is restored. The cloud-based intelligent analysis server then uses the complete teacher model to perform detailed analysis and result verification.

[0028] Preferably, the dedicated application of the user interaction terminal integrates augmented reality guidance function. When the sampling process is started, the mobile device camera is invoked to overlay a virtual sampling area frame and a pressure indicator bar on the real-time video screen to guide the user to align the sampling head with the damaged area of ​​the deck and control the applied pressure.

[0029] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0030] 1. This invention integrates mechanical sampling, microfluidic fluorescent labeling, and miniature fluorescence microscopy into a handheld terminal, constructing a closed, automated sample processing and imaging pipeline. This fundamentally eliminates the missed detection problems caused by insufficient sampling depth and inconsistent staining procedures in traditional methods. Standardized sampling actions and precisely controlled fluorescent labeling processes ensure consistent and reliable sample preparation quality for each sample, laying a solid foundation for subsequent high-sensitivity imaging and accurate analysis, and significantly improving the positive rate and repeatability of detection.

[0031] 2. This invention utilizes a deep convolutional neural network for fully automated analysis of fluorescence microscopic images, achieving precise identification and quantitative counting of fungal targets. It can distinguish specific fluorescence signals from complex sample background noise, exhibiting excellent detection capabilities for sparse and weak hyphae and spores in early infections, overcoming the drawbacks of subjective and inefficient manual microscopy. The entire process from sampling to report generation is shortened, meeting the urgent clinical needs for rapid diagnosis and screening.

[0032] 3. The integrated handheld terminal of this invention is ingeniously designed and easy to operate. Furthermore, its cloud-based intelligent analysis mode reduces reliance on local computing resources, making the overall system cost controllable. This allows the technology to overcome the dependence of traditional fluorescence detection on professional laboratory environments, enabling its widespread application in primary healthcare institutions, community clinics, and even home settings. This promotes the accessibility and affordability of onychomycosis diagnosis technology, providing a powerful technical tool for the early detection and timely intervention of the disease. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the overall technical solution architecture of the present invention;

[0034] Figure 2 This is a schematic diagram of the core principle framework of the fungal intelligent identification and counting algorithm based on deep convolutional neural networks in this invention;

[0035] Figure 3 This is a logical flow diagram of the internal modules of the integrated handheld sampling and imaging terminal in this invention;

[0036] Figure 4 This is a schematic diagram illustrating the principle framework of the collaborative operation of microfluidic fluorescent labeling and micro-fluorescence microscopy in this invention;

[0037] Figure 5 This is a schematic diagram of the multi-level interaction relationship and data flow between the cloud-based intelligent analysis server and the terminal in this invention. Detailed Implementation

[0038] refer to Figures 1 to 5 The miniature fluorescence microscopy imaging and automatic counting system for onychomycosis detection proposed in this invention consists of three main parts: an integrated handheld sampling and imaging terminal, a cloud-based intelligent analysis server, and a user interaction terminal. These three components form a closed-loop data stream through an encrypted wireless communication protocol. This system achieves full-chain automation in both physical structure and logical flow, from sample collection, fluorescence labeling, and microscopic imaging to intelligent analysis and result presentation, completely avoiding the variability and subjectivity introduced by manual operation in traditional onychomycosis detection.

[0039] The integrated handheld sampling and imaging terminal is the core hardware carrier of this system. Its internal module layout and logic control flow are shown in the attached figure. Figure 3 As shown, this integrated handheld sampling and imaging terminal integrates a mechanical sampling module, a microfluidic fluorescence labeling module, a micro fluorescence microscopy imaging module, a main control and communication module, and a power supply module. All modules are encapsulated in an ergonomically designed shell, and the entire device weighs less than 350 grams, making it easy to operate with one hand.

[0040] The mechanical sampling module, located at the front end of the terminal, performs standardized nail shavings sample collection. This module includes a replaceable disposable sampling head with a micro-scraping structure at its tip. This micro-scraping structure is precision injection molded from polyetheretherketone (PEEK) material, featuring a wedge-shaped cutting edge with a 25-degree leading edge angle and a surface texture of periodically arranged micron-level grooves, 8 micrometers deep and 15 micrometers apart. The micro-scraping structure is designed to effectively penetrate the surface keratin of the nail plate and scrape away deep nail shavings containing fungal hyphae when a vertical pressure of 0.8 to 1.2 Newtons is applied, while avoiding damage to the nail bed tissue.

[0041] The scraping mechanism is driven by a built-in micro stepper motor, which converts the rotational motion into linear reciprocating motion via a gear set and linkage mechanism. The stroke length is 2 mm, and the speed is 3 reciprocations per second. The sampling head has a sample storage chamber with a volume of 5 microliters. The chamber wall is coated with a superhydrophobic coating composed of fluorinated silica nanoparticles. The static water contact angle is greater than 150 degrees, and the dynamic roll-off angle is less than 5 degrees, ensuring that the collected nail clippings sample is free of residue during subsequent transfer, thus eliminating the risk of cross-contamination.

[0042] Once the sampling process is completed and confirmed by the main control module, the system automatically triggers the microfluidic fluorescent labeling process. The sample storage chamber of the microfluidic fluorescent labeling module and the mechanical sampling module are connected by a PTFE microchannel with an inner diameter of 200 micrometers. The microfluidic fluorescent labeling module is pre-encapsulated with lyophilized powder of a fluorescent staining reagent, CalcofluorWhiteM2R, which has a high specific affinity for chitin in fungal cell walls and emits bright blue-white fluorescence under ultraviolet or blue light excitation.

[0043] The lyophilized powder is sealed in an independent reagent chamber with a volume of 10 μL. Upon receiving a staining command, the main control module drives a micro-piezoelectric pump with a driving voltage of 12 V and an operating frequency of 50 Hz to quantitatively inject phosphate buffer pre-stored in the buffer chamber into the microchannel at a volume of 8 μL and a flow rate of 1 μL per second. The buffer solution has a pH of 7.4 and an ionic strength of 150 mmol / L, and is used to dissolve the lyophilized powder and form a 0.1% staining solution.

[0044] The staining solution is thoroughly mixed with the nail fragments in the sample storage chamber within the microchannel, and incubated at 37 degrees Celsius for 3 minutes under the control of the temperature control unit. The temperature control unit consists of a miniature thermoelectric cooler and a high-precision temperature sensor with a resolution of 0.1 degrees Celsius. The feedback control loop ensures a temperature control accuracy of ±0.5 degrees Celsius. After incubation, the main control module initiates the waste liquid discharge procedure, driving another miniature peristaltic pump to guide the waste liquid into a 20-microliter waste liquid collection chamber. This waste liquid collection chamber employs a disposable sealed design to prevent leakage of biological contamination.

[0045] The stained sample is then pushed by air pressure into a transparent observation microcavity on the stage of the micro fluorescence microscopy module. This transparent observation microcavity is made of optical glass with a thickness of 170 micrometers, an inner diameter of 1 millimeter, a depth of 50 micrometers, and its bottom is directly coupled to the imaging optical path.

[0046] The miniature fluorescence microscopy module employs an inverted fluorescence optical path design to minimize its vertical dimensions. The excitation source is a high-power light-emitting diode with a peak wavelength of 470 nm, an output optical power of 50 mW, and a full width at half maximum (FWHM) of 20 nm. The excitation light passes sequentially through a collimating lens and an excitation filter with a center wavelength of 470 nm and a bandwidth of 20 nm, forming a uniform parallel excitation beam. This parallel excitation beam is reflected by a dichroic mirror and then illuminates the sample within the observation microcavity through the objective lens.

[0047] The objective lens employs a miniature fixed-focus lens with a numerical aperture of 0.65, a working distance of 0.8 mm, and a focal length of 4 mm. The optical design incorporates aberration correction for the 525 nm emission wavelength. The fluorescence signal emitted by the excited sample passes through the objective lens, is transmitted through a dichroic mirror, and then through an emission filter with a center wavelength of 525 nm and a bandwidth of 40 nm before being received by a high-sensitivity complementary metal-oxide-semiconductor (CMOS) image sensor. This high-sensitivity CMOS image sensor has a 1920×1080 pixel array, a pixel size of 1.4 μm, a full-well capacity of 12,000 electrons, a readout noise of less than 2 electrons, supports a global shutter mode, and allows for dynamic adjustment of the exposure time from 1 ms to 500 ms.

[0048] The entire optical module is rigidly fixed on a three-dimensional nano-displacement platform, which consists of three orthogonal piezoelectric ceramic actuators, each with a stroke of 100 micrometers and a positioning resolution better than 1 nanometer. The main control module, according to a preset scanning strategy, controls the platform to perform a grid-like scan in the XY plane with a step size of 800 micrometers and a coverage area of ​​4 mm × 4 mm. Simultaneously, multi-layer focusing is performed in the Z-axis direction with an interlayer spacing of 2 micrometers, acquiring a total of 5 layers of images for subsequent depth-of-field fusion processing. The imaging process takes place in a completely light-proof dark chamber structure, with the inner wall of the chamber coated with a black matte coating with an absorbance greater than 99%, ensuring complete suppression of ambient light interference.

[0049] The main control and communication module is based on a 600 MHz embedded microprocessor, which integrates a floating-point unit and a hardware encryption engine. The firmware is responsible for coordinating the timing control of each module.

[0050] First, the sampling command issued by the user through the terminal interface is parsed, and the mechanical sampling module is started;

[0051] After sampling is completed, the staining and temperature control process of the microfluidic module is triggered;

[0052] After staining is completed and the sample is confirmed to be in place, the light source, sensor and displacement platform of the imaging module are initialized;

[0053] After imaging is completed, the original image data is initially compressed (using the lossless JPEG-LS algorithm with a compression ratio of approximately 2:1), and a sampling location code (calculated by the fusion of the built-in inertial measurement unit and the geomagnetic sensor), a unique device identifier (based on a 256-bit device certificate generated by the security element), and a timestamp are added. The data is then uploaded to the cloud-based intelligent analysis server via Wi-Fi 6 or Bluetooth 5.3 protocol through a TLS 1.3 encrypted channel.

[0054] The power supply module uses a 2000mAh lithium polymer battery, along with a multi-channel power management integrated circuit, to provide stable DC voltages of 3.3V, 5V and 12V to each module, and the whole machine can work continuously for more than 4 hours.

[0055] The cloud-based intelligent analytics server is deployed on a cloud infrastructure with high availability and elastic scalability. Its core functional modules are listed below. Figure 2 As shown. The server first preprocesses the received raw fluorescence image. The preprocessing procedure includes:

[0056] The first step is to perform background noise reduction using dark-field images acquired under the same imaging conditions, that is, to subtract the pixel values ​​of the dark-field images from the original images.

[0057] The second step is to use a non-uniform illumination correction algorithm based on polynomial fitting to eliminate the image brightness gradient caused by non-uniform excitation light.

[0058] The third step involves applying a contrast-limited adaptive histogram equalization algorithm with a neighborhood window size of 64×64 pixels and a cropping threshold of 2.0 to enhance the local contrast of weakly fluorescent targets without amplifying noise.

[0059] The preprocessed image is input into a fungal target detection neural network model. This fungal target detection neural network model adopts the U-Net++ architecture. The encoder part consists of 5 stacked residual blocks, each containing 2 convolutional layers and skip connections. The convolutional kernel size is 3×3, and the number of channels is 64, 128, 256, 512, and 1024 respectively. The decoder part performs upsampling through transposed convolution and forms dense skip connections with the feature maps of the corresponding layers of the encoder to preserve fine spatial details.

[0060] The output of the fungal target detection neural network model is a probability map of the same size as the input image, where each pixel value represents the probability that the location belongs to a fungal target. The model was trained using a large-scale labeled dataset containing over 100,000 fluorescence microscopic images of nail scales from different patients and different stages of infection. Each image was independently labeled and cross-validated by three senior dermatologists. The training loss function is a weighted sum of cross-entropy loss and focus loss, with weight coefficients of 1.0 and 2.0, respectively. The focus parameter γ for the focus loss was set to 2.0 to enhance attention to small, difficult-to-distinguish targets. An online hard example mining mechanism was also employed during training, dynamically selecting the top 20% of samples with the highest loss values ​​for focused optimization in each iteration.

[0061] The probability map output by the model is binarized with a threshold of 0.5 to generate a preliminary fungal target mask. Subsequently, the system performs connected component analysis and morphological post-processing. First, the binary mask is labeled with 8-neighbor connected components to obtain several independent target regions. Second, morphological features are calculated for each connected component, including area (unit: square micrometers), perimeter (unit: micrometers), roundness (defined as 4π × area / perimeter²), mean fluorescence intensity (unit: gray value), and variance (unit: gray value²).

[0062] Based on prior knowledge derived from extensive clinical data, the following screening thresholds were set for effective fungal targets: area between 50 and 5000 square micrometers, sphericity greater than 0.3, and mean fluorescence intensity greater than 3 standard deviations of the background mean. Connected regions that do not meet any of the above conditions are identified as false targets and discarded. The remaining connected regions are the confirmed effective fungal targets, including hyphal fragments and spores.

[0063] The counting and report generation unit counts all valid fungal targets to obtain the total number of tests conducted. Simultaneously, the system calculates the fungal density per unit field of view. The calculation formula is:

[0064]

[0065] The total area of ​​all spliced ​​fields of view is expressed in square millimeters. Fungal density, combined with clinical diagnostic criteria, was used to estimate the fungal load in the entire sampling area and classified into four infection levels: "negative," "mild," "moderate," and "severe."

[0066] The location coordinates, morphological parameters, and statistical results of all identified targets are stored in a structured cloud database, and a digital detection report is automatically generated. This digital detection report includes the total number of fungi, a density distribution histogram, marked images of typical targets (highlighted with colored bounding boxes on the original image), an assessment of the degree of infection, and a recommended follow-up period.

[0067] The user interaction terminal is a dedicated application installed on a smartphone or tablet, and its interaction with the cloud server is shown in the attached figure. Figure 5 As shown. After the application starts, it first completes user authentication via the OAuth2.0 protocol and establishes an end-to-end encrypted secure session. Once the detection task is completed, the application automatically retrieves the latest detection report and displays key indicators in a card-style layout on the main interface. Users can click on any marked image to enter a details view, which supports two-finger zoom and panning to view the high-resolution original image. The application also provides a history function, allowing users to compare the trend of fungal density changes across different tests along a timeline and generate a disease progression curve.

[0068] In addition, the application integrates augmented reality guidance: when the user is ready to take a sample, the front-facing camera of the mobile device is invoked, and a virtual green sampling area frame is superimposed on the real-time video screen. The position of the green sampling area frame is determined by the device's attitude sensor matching the preset deck anatomy model. At the same time, a pressure indicator bar is displayed on the right side of the screen. The height is fed back in real time by the pressure sensor built into the integrated handheld sampling imaging terminal. When the pressure is within the ideal range of 0.8 to 1.2 Newtons, the indicator bar is green, and it turns red when it is outside the range, thereby guiding the user to complete the standardized sampling.

[0069] The entire system workflow is as follows: The user initiates a sampling task through the user interaction terminal. The application uses augmented reality guidance to help the user align the integrated handheld sampling imaging terminal with the affected nail area. The user presses the sampling button on the terminal, and the mechanical sampling module performs a scraping action to collect nail scraping samples. After sampling, the microfluidic fluorescent labeling module automatically completes staining and cleaning. The miniature fluorescence microscopy imaging module performs multi-field and multi-focal plane imaging on the stained sample. The main control module encrypts and uploads the raw image data to the cloud. The cloud server completes intelligent analysis and generates a report within 30 seconds. The user interaction terminal receives and displays the report instantly.

[0070] Based on the aforementioned embodiments, this embodiment further introduces a cloud-edge collaborative intelligent analysis architecture to address usage scenarios with unstable network signals or no network coverage.

[0071] Specifically, after training the fungal target detection neural network model, the cloud-based intelligent analysis server uses knowledge distillation technology to compress the model. The teacher model uses the aforementioned complete U-Net++ architecture, while the student model is designed as a lightweight MobileNetV3-UNet hybrid structure, with only 8% of the parameters and 15% of the computational cost. During distillation, the student model not only learns the probability distribution output by the teacher model but also the attention weights of the feature maps to preserve key spatial semantic information. After training, the lightweight student model is deployed in the main control module of the integrated handheld sampling and imaging terminal.

[0072] In local analysis mode, when the terminal detects a network connection interruption or the user actively selects offline mode, the main control module will invoke the locally deployed student model to perform real-time preliminary analysis of the acquired raw fluorescence images. Because the student model has been specifically optimized, the detection rate for typical fungal targets can still reach over 85%. The analysis process includes rapid preprocessing (performing only background noise subtraction and simple histogram stretching) and target detection, outputting preliminary fungal target masks and quantity estimates. The pre-screening results are cached along with the raw image data in the terminal's local memory.

[0073] Once the network is restored, the terminal immediately uploads the cached data to the cloud. The cloud server uses a complete teacher model to perform detailed analysis of the original images and compares the final results with the local pre-screened results. If the difference exceeds a preset threshold (e.g., a quantity error greater than 20%), a manual review process is triggered. This cloud-edge collaborative mechanism ensures both basic availability in offline environments and high accuracy of the final diagnostic results, significantly improving the system's robustness and applicability.

[0074] Furthermore, this embodiment incorporates a redundant design for the microfluidic fluorescent labeling module. In addition to the original single reagent chamber, a spare reagent chamber is added, containing FilipinIII, a fluorescent dye with high affinity for ergosterol. The main control module can switch between different staining reagents based on instructions from the cloud server or user selection.

[0075] For example, when a preliminary analysis suspects a non-dermatophyte fungal infection, the system can automatically activate FilipinIII for secondary staining verification, as it has a better staining effect on non-filamentous fungi such as yeast.

[0076] Both staining procedures share the same microfluidic system and temperature control system, but the flow path is switched via independent micro-solenoid valves to ensure no cross-contamination between reagents. This design enhances the system's ability to differentiate between complex or mixed infection cases.

[0077] On the user interaction terminal side, this embodiment adds a multimodal data fusion function. The application not only displays the fluorescence image analysis results of this test, but can also associate them with user-inputted clinical symptom descriptions (such as the degree of itching, the extent of nail discoloration, and the duration of the illness). Through a lightweight clinical decision support model, it integrates imaging and symptomatological information to generate a more comprehensive infection risk assessment. This clinical decision support model uses an integrated approach of logistic regression and random forest, runs on the local device, and protects user privacy. The assessment results are presented in the form of a visual dashboard to help users understand the severity of their condition and provide a reference for subsequent medical treatment.

[0078] Through the above improvements, this embodiment significantly enhances the system's environmental adaptability, diagnostic breadth, and user experience while maintaining the core functions, enabling it to better serve diverse clinical and home application scenarios.

[0079] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0080] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis, characterized in that, include: An integrated handheld sampling and imaging terminal is used to perform physical collection, fluorescent labeling, microscopic imaging, and uploading of raw image data of nail dander samples; A cloud-based intelligent analysis server is used to receive and process image data from the integrated handheld sampling and imaging terminal, and to perform fungal target identification and automatic counting analysis. User interaction terminal, used to display test result reports to users; The cloud-based intelligent analysis server is equipped with a fungal intelligent identification and counting algorithm based on a deep convolutional neural network. It first preprocesses the received raw fluorescence image. The preprocessing includes background noise subtraction based on dark field images, non-uniform illumination correction, and contrast-limited adaptive histogram equalization enhancement. The preprocessed image is then input into a specially trained fungal target detection neural network model. The model outputs a probability map of each pixel belonging to the fungal target category; The probability map is binarized with a fixed threshold to generate a preliminary fungal target mask. Then, connected component analysis based on 8 neighborhoods and morphological post-processing are performed to calculate the morphological features of each connected component and filter out false targets according to a preset threshold. The confirmed valid fungal targets are statistically analyzed to obtain the total number of hyphae and spores, and the fungal density per unit field of view is calculated. The fungal load of the entire sampling area is estimated by combining the total area of ​​multiple spliced ​​fields of view. A digital detection report containing the total number of fungi, density distribution, typical target marker images, and infection degree assessment is automatically generated.

2. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 1, characterized in that, The integrated handheld sampling and imaging terminal includes a mechanical sampling module, a microfluidic fluorescence labeling module, a miniature fluorescence microscopy imaging module, a main control and communication module, and a power supply module. The mechanical sampling module is used to perform standardized nail dandruff sample collection actions, including a replaceable disposable sampling head. The sampling head has a micro-scraping structure at the front end, which is made of biocompatible material. The scraping angle and pressure threshold are controlled by preset mechanical limits and elastic elements. The sampling head integrates a micro-sample storage cavity. The microfluidic fluorescent labeling module is connected to the sample storage chamber of the mechanical sampling module through a microchannel. It is used to perform automated fluorescent staining on the collected nail shavings samples. The module is pre-encapsulated with dried lyophilized fluorescent staining reagent powder. The fluorescent staining reagent is a specific fluorescent dye with high affinity for fungal chitin or ergosterol. The microfluidic fluorescent labeling module also includes a micro piezoelectric pump and an independent buffer chamber, which are used to quantitatively inject buffer into the microchannel to dissolve the lyophilized powder and form a staining liquid stream. The staining liquid stream is mixed and incubated with nail shavings fragments under preset temperature control conditions. The micro-fluorescence microscopy imaging module is used to acquire fluorescence microscopic images of stained samples. It adopts an inverted fluorescence optical path design and includes an excitation source, an excitation filter, an objective lens, a dichroic mirror, an emission filter, and a high-sensitivity complementary metal-oxide-semiconductor image sensor. The objective lens is a miniaturized fixed-focus lens with a high numerical aperture. The entire optical module is fixed on a three-dimensional nano-displacement platform driven by a piezoelectric ceramic actuator. The main control and communication module includes an embedded microprocessor and a wireless communication unit. The embedded microprocessor is used to coordinate and control the operation timing of the mechanical sampling module, the microfluidic fluorescence labeling module and the micro fluorescence microscopy imaging module. The wireless communication unit is used to encrypt and upload the acquired raw fluorescence image data, sampling position encoding information and device identification code to the cloud intelligent analysis server. The power supply module provides power to all electronic components within the integrated handheld sampling imaging terminal.

3. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 2, characterized in that, The user interaction terminal is a dedicated application installed on a smart mobile device. It establishes a secure connection with the cloud-based intelligent analysis server through authorization and authentication, receives and visualizes the digital test report, and provides historical data query and management functions.

4. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 3, characterized in that, The fungal target detection neural network model adopts an encoder-decoder architecture. The encoder part uses a residual network structure to extract multi-scale feature maps of the input image, and the decoder part fuses deep semantic information and shallow detail information through upsampling and skip connections.

5. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 4, characterized in that, The training process of the fungal target detection neural network model uses a large-scale labeled fluorescence microscopic image dataset of nail clippings. During training, the cross-entropy loss function and the focus loss function optimized for small target detection are used, and an online hard example mining strategy is adopted.

6. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 5, characterized in that, The morphological features of the connected domains calculated in the morphological post-processing include area, perimeter, roundness, mean and variance of fluorescence intensity. The preset threshold range for filtering out false targets is set for area, shape regularity and fluorescence intensity distribution.

7. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 6, characterized in that, The process for estimating the fungal load is as follows: All effective fungal targets are counted to obtain the total number. Calculate the total area of ​​all stitched fields of view. Fungal density The calculation formula is Based on clinical diagnostic criteria, fungal density is classified into different infection levels.

8. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 7, characterized in that, The fungal target detection neural network model of the cloud-based intelligent analysis server is compressed using knowledge distillation technology to obtain a lightweight student model. This lightweight student model is deployed in the main control module of the integrated handheld sampling and imaging terminal to perform local real-time preliminary analysis and target pre-screening in an environment without network connection.

9. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 8, characterized in that, The integrated handheld sampling and imaging terminal caches the pre-screening results of the lightweight student model along with the original image data, and uploads them to the cloud-based intelligent analysis server after the network is restored. The cloud-based intelligent analysis server then uses the complete teacher model to perform detailed analysis and result verification.

10. The miniature fluorescence microscopy imaging and automatic counting system for detecting onychomycosis according to claim 9, characterized in that, The user interaction terminal's dedicated application integrates augmented reality guidance functionality. When the sampling process is initiated, the mobile device's camera is invoked to overlay a virtual sampling area frame and a pressure indicator bar onto the real-time video screen, guiding the user to align the sampling head with the damaged area of ​​the deck and control the applied pressure.

Images