A hand-held retinal image acquisition and analysis device

Abstract

The present application relates to the field of ophthalmic medical apparatus and optical imaging technology, and particularly relates to a hand-held retinal image acquisition and analysis device, comprising: a shell composed of a handle and a support frame, an inclined support frame being arranged on the left side of the top end of the handle, and rubber blocks being arranged at the bottom end of the handle and the support frame; an imaging optical system shell group being installed above the handle and the support frame through a barrel seat, the imaging optical system shell group comprising, in sequence along the optical axis direction, a buffer silica gel barrel, an ocular objective lens, a liquid lens group, an imaging lens group and an image sensor; and the imaging optical system shell group further comprising a touch screen display screen. The present application adopts the handle and support frame shell and the imaging optical system shell group integrated along the optical axis, replaces the traditional mechanical focusing assembly with a single liquid lens, reduces the total number of parts from about 80 to about 45, and reduces the volume from about 300cm 3 to about 195cm 3 , thereby solving the problems of complex structure and large volume.

Description

Technical Field

[0001] This invention relates to the field of ophthalmic medical devices and optical imaging technology, and in particular to a handheld retinal image acquisition and analysis device. Background Technology

[0002] Current handheld fundus cameras generally employ traditional mechanical focusing structures for image focusing—achieving clear imaging of the human eye with different refractive powers through the relative movement of multiple lens groups along the optical axis. These mechanical focusing structures typically include movable focusing lens groups, precision guide rails or lead screw transmission mechanisms, stepper motors, and position sensors. They suffer from inherent problems such as complex structure, high assembly precision requirements, easy wear of transmission components, difficulty in reducing overall size and weight, and susceptibility to focus drift due to handheld vibrations.

[0003] In recent years, liquid lens technology has begun to be introduced into the field of fundus imaging. It achieves continuous changes in optical power by changing the curvature of the liquid interface through voltage, without the need for mechanical moving parts.

[0004] Current liquid lens fundus cameras focus solely on autofocus, neglecting its coordination with the illumination system. Under non-mydriatic imaging conditions (pupil diameter 2-3.5mm), different optical powers correspond to different imaging depths. Light attenuation is less in shallow imaging, while absorption and scattering by the cornea, lens, and other media are more significant in deep imaging, resulting in significantly different optimal illumination power requirements. Current devices use constant illumination power throughout autofocus and imaging, leading to wasted light energy and even overexposure in shallow imaging, and insufficient illumination and dark images in deep imaging. This also causes unnecessary glare for the patient and results in poor image quality consistency.

[0005] Meanwhile, in existing liquid lens fundus cameras, the liquid lens drive and illumination drive are independent of each other, and there is no function mapping relationship between them. They remain two parallel and independent control loops, unable to automatically match appropriate illumination conditions at different focal lengths and depths. Operators often need to manually adjust the illumination brightness or repeat the shooting, reducing examination efficiency and success rate. In addition, existing devices lack an active optimization mechanism for patient comfort and cannot dynamically adjust the illumination strategy according to the imaging depth to reduce the patient's light burden.

[0006] Furthermore, the liquid lens drive and illumination drive of existing liquid lens fundus cameras are independent of each other, and there is no function mapping relationship between them. They are still two parallel and independent control loops, which cannot automatically match appropriate illumination conditions at different focal lengths and depths. At the same time, the image analysis function of existing handheld fundus cameras mostly relies on external devices or cloud computing, making it difficult to achieve real-time, low-power intelligent screening of fundus lesions locally on the device, thus limiting their application in environments without network or in primary healthcare scenarios.

[0007] Existing devices lack proactive optimization mechanisms for patient comfort. During fundus examinations, patients should receive as little light as possible. However, different examination requirements, such as macular area detection versus peripheral retinal detection, necessitate different imaging depths and focusing conditions. Current technologies cannot dynamically adjust illumination strategies based on imaging depth to reduce the patient's light burden. Therefore, a handheld retinal image acquisition and analysis device is needed to address these issues. Summary of the Invention

[0008] The purpose of this invention is to solve the problems of existing handheld retinal image acquisition devices mentioned in the background art, such as complex mechanical focusing structures, independent focusing and illumination control of liquid lenses, and reliance on external devices for image analysis.

[0009] To achieve the above objectives, the present invention adopts the following technical solution: A handheld retinal image acquisition and analysis device, comprising: The housing consists of a handle and a support frame. An inclined support frame is provided on the top left side of the handle, and a rubber block is provided at the bottom of the handle and the support frame. An imaging optical system housing is mounted above the handle and support frame via a base. Along the optical axis, the housing comprises, in sequence, a buffer silicone tube, an eyepiece objective, a liquid lens group, an imaging lens group, and an image sensor. The housing also includes a touchscreen display. The liquid lens group consists of a first lens barrel and liquid lens plates disposed within it. The imaging lens group consists of a second lens barrel, a biconvex lens, a spacer, and a meniscus lens. The lighting system, located inside the eyepiece objective lens, includes a ring-shaped LED lighting panel, a light diffuser, and an lighting drive circuit. A liquid lens driving circuit, electrically connected to the liquid lens sheet, is used to apply a driving voltage to the liquid lens sheet to adjust its optical power; The signal processing and control unit is electrically connected to the liquid lens driving circuit, the illumination driving circuit, and the image sensor. The signal processing and control unit includes a sharpness evaluation subunit and a collaborative control subunit. The sharpness evaluation subunit is used to calculate the sharpness evaluation index value of the fundus image output by the image sensor. The collaborative control subunit stores a preset optical power-illumination power mapping function, which defines the target illumination light power value that the illumination system should output under different optical powers. A power module, built into the handle, supplies power to the aforementioned electrical components. The signal processing and control unit is configured to: during autofocus, generate a liquid lens drive voltage adjustment command based on feedback from the sharpness evaluation subunit to change the optical power of the liquid lens in the direction of making the image sharp; and generate an illumination power adjustment command based on the current optical power value and the optical power-illumination power mapping function to dynamically adjust the output light power of the illumination system as the optical power changes.

[0010] Preferably, the optical power-illumination power mapping function is: P_target=P_base+k_focus·|D|+k_depth·L(D); Where P_target is the target illumination power, P_base is the reference illumination power, k_focus is the gain coefficient related to the amplitude of the change in optical power, D is the current optical power of the liquid lens, k_depth is the gain coefficient related to the imaging depth, and L(D) is the imaging depth compensation function.

[0011] Preferably, the illumination system includes a near-infrared guiding illumination source and a visible light exposure illumination source; during autofocus, the illumination system adopts a near-infrared guiding illumination mode; After focusing is completed and the exposure shooting stage begins, the illumination system switches to visible light exposure illumination mode, and the light power of the visible light exposure illumination is determined according to the current optical power value.

[0012] Preferably, the signal processing and control unit further includes a pupil diameter detection module, used to automatically calculate the pupil diameter of the subject based on the image acquired by the image sensor, and to perform amplitude limiting correction on the target illumination power output by the optical power-illumination power mapping function based on the pupil diameter.

[0013] Preferably, the signal processing and control unit further includes an AI analysis module, used to perform real-time analysis of the acquired retinal images based on a lightweight convolutional neural network model, and to provide preliminary screening and prompts for fundus diseases; The AI ​​analysis module is also configured to evaluate the image quality of the acquired images and send a feedback signal to the collaborative control subunit when it determines that the image quality does not meet the preset conditions. The collaborative control subunit automatically adjusts the parameters of the optical power-illumination power mapping function according to the feedback signal, triggering secondary optimization acquisition.

[0014] Preferably, the device further includes a touch screen display and operation buttons disposed on the imaging optical system housing, the operation buttons including a shutter button and a mode switching button; The signal processing and control unit is configured to switch between fully automatic and semi-automatic modes in response to the operation of the mode switching button: in the fully automatic mode, the signal processing and control unit autonomously performs focusing and illumination power adjustment. In the semi-automatic mode, the user can manually set the optical power of the liquid lens using the operation button, and the signal processing and control unit will automatically match the illumination power according to the manually set optical power.

[0015] Preferably, the device has multiple preset working modes, including a rapid screening mode and a high-precision diagnostic mode, with different working modes corresponding to different optical power-illumination power mapping function parameters.

[0016] Preferably, the liquid lens is an electrowetting type double liquid lens with an optical power adjustment range of -5D to +13D.

[0017] Preferably, the imaging optical system housing is mounted above the handle and support frame via a cylindrical base; The eyepiece is disposed at the front end of the imaging optical system housing; The liquid lens group and the imaging lens group are sequentially disposed inside the imaging optical system housing; The buffer silicone sleeve is disposed between the imaging lens group and the image sensor; The touchscreen display is located on the rear end face of the imaging optical system housing.

[0018] Preferably, the annular LED lighting panel of the lighting system is sleeved on the outside of the eyepiece and located on the inner wall of the eyepiece, and the light-diffusing sheet covers the front of the annular LED lighting panel; The liquid lens driving circuit is fixed to the side of the imaging optical system housing and is connected to the liquid lens and the signal processing and control unit respectively via flexible cables.

[0019] The present invention has at least the following beneficial effects: 1. This invention, by employing a handle and support frame housing and an imaging optical system housing integrated along the optical axis, replaces the traditional mechanical focusing assembly with a single liquid lens, reducing the number of parts in the entire device from approximately 80 to approximately 45, and the volume from approximately 300 cm³. 3 Shrink to approximately 195cm 3 This solved the problems of complex structure and large size.

[0020] 2. This invention uses a signal processing and control unit to pre-store the optical power-illumination power mapping function. During autofocus, the illumination power dynamically adapts to the optical power, reducing the standard deviation of fundus image brightness from ±3.2dB to ±1.1dB. The success rate of a single shot is increased by approximately 23.8 percentage points compared to existing equipment, thus solving the problem of poor image quality without mydriasis.

[0021] 3. This invention, by setting the switching between near-infrared guiding illumination and visible light exposure illumination, uses 850nm near-infrared light, which is insensitive to the human eye, for automatic focusing throughout the process. Patients experience no flickering and their pupils do not constrict, effectively improving examination comfort and ensuring the continuity of non-mydriatic conditions.

[0022] 4. This invention monitors the pupil diameter in real time through a pupil diameter detection module and limits and corrects the illumination power. In scenarios with small pupils, the upper limit of illumination is forcibly reduced, which significantly reduces patient discomfort and avoids the adverse effects of further pupil constriction on imaging.

[0023] 5. This invention integrates an AI analysis module and runs a lightweight neural network model on the local NPU to achieve real-time classification of fundus lesions, i.e., DR staging accuracy of 91.3%, and has image quality evaluation and secondary optimization acquisition functions, thus solving the problem of dependence on the cloud.

[0024] 6. This invention offers two modes: fully automatic and semi-automatic. The fully automatic mode is suitable for screening at the grassroots level, while in the semi-automatic mode, doctors can manually set the optical power and the system automatically matches the illumination power. For patients with high myopia, the average focusing time is reduced from 3.2 seconds to 1.8 seconds, taking into account both ease of use and professional needs.

[0025] 7. This invention uses a preset rapid screening mode and a high-precision diagnostic mode. Different modes correspond to different optical power-illumination power mapping function parameters. In the rapid screening mode, the focusing speed is fast and the AI ​​only outputs binary classification results, which is suitable for large-scale screening. In the high-precision diagnostic mode, the focusing accuracy is high and the AI ​​performs local magnification analysis, which is suitable for detailed examination of confirmed patients.

[0026] 8. This invention employs an electrowetting dual-liquid lens, with an optical power adjustment range covering -5D to +13D, which is sufficient to meet the non-mydriatic fundus imaging needs of the vast majority of people. Moreover, the adjustment speed reaches the millisecond level, which is faster and more reliable than traditional mechanical focusing.

[0027] 9. This invention precisely defines the axial sequence and matching relationship of the buffer silicone tube, eyepiece objective, liquid lens group, imaging lens group, and image sensor, and places the annular LED illumination plate on the outside of the eyepiece objective and covers its front with a light-diffusing sheet. While achieving optical axis coaxiality, illumination uniformity, and assembly compactness, the buffer silicone tube absorbs hand vibrations, providing a stable optical and mechanical support structure for the realization of the above-mentioned multi-module collaborative control function. Attached Figure Description

[0028] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0029] Figure 1 This is a schematic diagram of the external structure of a handheld retinal image acquisition and analysis device proposed in this invention; Figure 2 This is a schematic diagram of the disassembly and assembly structure of a handheld retinal image acquisition and analysis device proposed in this invention; Figure 3 This is a three-dimensional disassembled structural diagram of the liquid lens group in a handheld retinal image acquisition and analysis device proposed in this invention. Figure 4 This is a three-dimensional disassembled structural diagram of the imaging lens group in a handheld retinal image acquisition and analysis device proposed in this invention. Figure 5 This is a three-dimensional structural diagram of the buffer silicone tube in a handheld retinal image acquisition and analysis device proposed in this invention. Figure 6 This is a schematic diagram of the workflow of a handheld liquid lens retinal image acquisition device proposed in this invention. Figure 7 This is the sub-process of the optical power-illumination coordinated control of the present invention.

[0030] In the picture: 1. Imaging optical system housing; 11. Eyepiece objective lens; 12. Liquid lens group; 121. Lens barrel one; 122. Liquid lens plate; 13. Imaging lens group; 131. Lens barrel two; 132. Biconvex lens; 133. Spacer ring; 134. Meniscus lens; 14. Buffer silicone tube; 15. Touch screen display; 2. Handle; 3. Support frame. Detailed Implementation

[0031] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0032] The basic structural features are as follows: Reference Figure 1-7 A handheld retinal image acquisition and analysis device, comprising: The housing consists of a handle 2 and a support frame 3. An inclined support frame 3 is provided on the top left side of the handle 2, and rubber blocks are provided at the bottom of the handle 2 and the support frame 3 for cushioning and support. The imaging optical system housing 1 is mounted on the handle 2 and the support frame 3 via a cylindrical base. The imaging optical system housing 1 includes, in sequence along the optical axis, a buffer silicone tube 14, an eyepiece objective lens 11, a liquid lens group 12, an imaging lens group 13, and an image sensor. The imaging optical system housing 1 also includes a touch screen display 15. The liquid lens group 12 consists of a lens barrel 121 and a liquid lens sheet 122 disposed inside the lens barrel 121; The imaging lens group 13 consists of a lens barrel 131, a biconvex lens 132, a spacer 133 which uses a 0.1mm copper gasket, and a meniscus lens 134. The illumination system, located inside the eyepiece objective lens 11, includes a ring-shaped LED illumination panel, a light diffuser, and an illumination driving circuit. The liquid lens driving circuit is electrically connected to the liquid lens plate 122 and is used to apply a driving voltage to the liquid lens plate 122 to adjust its optical power. The signal processing and control unit is electrically connected to the liquid lens 122 driving circuit, the illumination driving circuit, and the image sensor. The signal processing and control unit includes a sharpness evaluation subunit and a cooperative control subunit. The sharpness evaluation subunit is used to calculate the sharpness evaluation index value of the fundus image output by the image sensor. The cooperative control subunit stores a preset optical power-illumination power mapping function, which defines the target illumination light power value that the illumination system should output under different optical powers. The power module, which is built into the handle 2, supplies power to the aforementioned electrical components. The signal processing and control unit is configured to: during autofocus, generate a liquid lens drive voltage adjustment command based on feedback from the sharpness evaluation subunit to change the optical power of the liquid lens in the direction of making the image sharp; and generate an illumination power adjustment command based on the current optical power value and the optical power-illumination power mapping function to dynamically adjust the output light power of the illumination system as the optical power changes.

[0033] The optical power-illumination power mapping function is: P_target=P_base+k_focus·|D|+k_depth·L(D); Where P_target is the target illumination power, P_base is the reference illumination power, k_focus is the gain coefficient related to the amplitude of the change in optical power, D is the current optical power of the liquid lens (122), k_depth is the gain coefficient related to the imaging depth, and L(D) is the imaging depth compensation function.

[0034] The illumination system includes a near-infrared guiding illumination source and a visible light exposure illumination source. During autofocus, the illumination system uses the near-infrared guiding illumination mode. After focusing is completed and the exposure shooting stage begins, the illumination system switches to the visible light exposure illumination mode, and the light power of the visible light exposure illumination is determined according to the current optical power value.

[0035] The signal processing and control unit also includes a pupil diameter detection module, which is used to automatically calculate the pupil diameter of the subject based on the image acquired by the image sensor, and to limit and correct the target illumination power output by the optical power-illumination power mapping function based on the pupil diameter.

[0036] The signal processing and control unit also includes an AI analysis module, which performs real-time analysis of the acquired retinal images based on a lightweight convolutional neural network model and provides preliminary screening and alerts for fundus diseases. The AI ​​analysis module is also configured to evaluate the image quality of the acquired images and send a feedback signal to the collaborative control subunit when the image quality does not meet the preset conditions. The collaborative control subunit automatically adjusts the parameters of the optical power-illumination power mapping function based on the feedback signal to trigger secondary optimization acquisition.

[0037] The device also includes a touchscreen display and operation buttons mounted on the housing, including a shutter button and a mode switching button. The signal processing and control unit is configured to switch between fully automatic and semi-automatic modes in response to the operation of the mode switching button. In fully automatic mode, the signal processing and control unit autonomously completes focusing and illumination power adjustment. In semi-automatic mode, the user can manually set the optical power of the liquid lens through the operation buttons, and the signal processing and control unit automatically matches the illumination power according to the manually set optical power.

[0038] The device has multiple preset working modes, including a rapid screening mode and a high-precision diagnostic mode. Different working modes correspond to different optical power-illumination power mapping function parameters.

[0039] The liquid lens is an electrowetting type double liquid lens with an optical power adjustment range of -5D to +13D.

[0040] The imaging optical system housing 1 is mounted above the handle 2 and the support frame 3 via a cylindrical base; The buffer silicone tube 14 is located at the front end of the imaging optical system housing 1 and is used to fit the patient's eye socket; The eyepiece 11 is positioned behind the buffer silicone tube 14; The liquid lens group 12 and the imaging lens group 13 are sequentially arranged behind the eyepiece objective lens 11; The image sensor is located behind the imaging lens group 13; The touch screen display 15 is located on the rear end face of the imaging optical system housing 1.

[0041] The liquid lens 122 is an electrowetting type dual liquid lens with an oil-water mixed structure. Its packaging structure is as follows: a sealed cavity is formed by a first transparent glass substrate, a second transparent glass substrate and an annular sealing frame. The cavity is filled with immiscible conductive liquid (electrolyte) and insulating liquid (non-conductive silicone oil). A transparent conductive film (such as ITO) is deposited on the inner surface of the first substrate as a common electrode. An annular independent control electrode is provided inside the sealing frame. The electrode surface and the inner wall of the sealing frame are fully covered with a hydrophobic insulating dielectric layer.

[0042] The zoom principle is as follows: The liquid lens driving circuit applies a driving voltage to the conductive liquid through the common electrode and the control electrode. Based on the electrowetting effect, the surface hydrophobicity of the hydrophobic insulating dielectric layer decreases as the voltage increases, the contact angle between the conductive liquid and the solid wall decreases, and the radius of curvature of the meniscus between the two liquids is forced to decrease, thus increasing the optical power of the liquid lens and shortening the focal length. The driving voltage and the optical power have an approximately square relationship. By adjusting the amplitude of the driving voltage (recommended voltage range 0~110V), the optical power can be continuously varied in the range of -5D to +13D, with a response time of less than 50ms.

[0043] The following analysis and description of this solution are based on multiple embodiments; Example 1; This embodiment is applicable to non-mydriatic fundus examinations in ophthalmology departments of primary hospitals or mobile screening vehicles. The operator is a nurse or a technician who has undergone short-term training. The examinee is an ordinary adult patient (refractive error range -6D to +6D, natural pupil diameter of about 2.5 to 4.0 mm). The examination environment is indoor normal lighting (illuminance 300 to 500 lux), and mydriasis is not required.

[0044] The mechanical structure of this device is divided into two main parts according to function: a handheld support unit and an optical imaging unit.

[0045] The handheld support unit consists of a handle 2 and a support frame 3. The handle 2 is cylindrical with a hollow interior and an opening at the bottom for inserting a power module. The lithium battery has a capacity of 3000mAh. The top left of the handle 2 has an integrally injection-molded inclined support frame 3. The upper surface of the support frame 3 is a horizontal platform with a 12mm diameter cylinder mounting hole in the center of the platform. The inner wall of the hole has a radial set screw thread. Both the handle 2 and the support frame 3 have 2mm thick silicone rubber blocks attached to their bottom ends for cushioning and preventing slippage during placement.

[0046] The optical imaging unit is an independent imaging optical system housing 1, which is cylindrical in shape, with a total length of 150mm and an outer diameter of 45mm. A touch screen display 15 is provided on the rear end. The bottom of the imaging optical system housing 1 is provided with a metal cylinder base made of 6061 aluminum alloy. The cylinder base is fixed to the mounting hole at the upper end of the support frame 3 by interference fit with an interference amount of 0.02~0.05mm and M2 radial set screw. The cylinder base is also provided with a flexible ribbon cable channel with a diameter of 3mm, which is used to lead the signal line inside the imaging optical system housing 1 to the main control PCB inside the handle 2.

[0047] The imaging optical system housing 1 is installed sequentially along the optical axis from front to back, i.e., from the patient side to the operator side: The buffer silicone tube 14 is a trumpet-shaped silicone elastomer with a Shore A hardness of 30A. It has a front outer diameter of 50mm, a rear outer diameter of 42mm, and an axial length of 15mm. Its front end protrudes 10mm beyond the frontmost part of the shell assembly. It is used to conform to the patient's eye socket, providing cushioning, positioning, and blocking external stray light. The rear end of the buffer silicone tube 14 is embedded in an annular groove on the inner wall of the front end of the shell assembly via a slot structure and is further secured with medical-grade instant adhesive.

[0048] Eyepiece 11: Located at the rear center of the buffer silicone tube 14, it is screwed into the internal thread hole at the front end of the housing assembly through an M12×0.75 fine thread. A 1.5mm gap is left between its front end face and the rear end face of the buffer silicone tube 14 to avoid direct contact. The eyepiece 11 is a cemented doublet achromatic convex lens with a focal length of 20mm and a light aperture of 10mm, used to collect reflected light from the fundus.

[0049] Liquid lens assembly 12: Composed of lens barrel 121 and liquid lens plate 122. Lens barrel 121 is a POM (polyoxymethylene) injection molded part with an outer diameter of 14mm, an inner diameter of 6mm, and a length of 8mm. The outer wall has two O-ring sealing grooves and is fitted with a 1.5mm silicone ring. It is fixed in the first-stage stepped hole in the inner cavity of the housing by a pressure ring with a pressure ring torque of 0.2N·m. Liquid lens plate 122 is an electrowetting type double liquid lens, model Varioptic Arctic416, which is inserted into the annular groove on the inner wall of lens barrel 121. Its 0.5mm pitch flexible circuit board is led out from the opening on one side wall of the lens barrel and connected to the liquid lens drive circuit board fixed on the side of the housing by a miniature FPC connector, model Molex503480.

[0050] Imaging lens group 13: It consists of lens barrel 2 131, biconvex lens 132, spacer 133 (i.e., 0.1mm thick copper gasket), and meniscus lens 134. Lens barrel 2 131 is an aluminum alloy CNC part with an outer diameter of 12mm and a length of 12mm. Its outer wall slides with the second-stage stepped hole in the inner cavity of the shell assembly with a gap of 0.01~0.02mm. The rear end is pressed by a wave spring with a compression of 0.5mm. Biconvex lens 132 has a diameter of 8mm and a center thickness of 3mm and is made of BK7 material. Meniscus lens 134 has a diameter of 6mm and a center thickness of 2mm and is made of SF5 material. They are placed in the lens barrel 2 in sequence and separated by spacer 133. The front end is locked with an M8×0.5 pressure ring.

[0051] Image sensor: A CMOS sensor, model Sony IMX287, with a resolution of 720×540 and a pixel size of 6.9μm. It is mounted on a 20mm×20mm square FR-4 PCB, and the four corners of the PCB are fixed to the inner boss at the rear end of the housing assembly with M1.6 screws. The photosensitive surface of the sensor is perpendicular to the optical axis and located at the focal plane of the imaging lens group 13. The distance between the photosensitive surface and the rear end face of the meniscus lens 134 is 2.0±0.05mm. This distance is determined by Zemax optical design and fine-tuned during assembly using 0.1mm stainless steel shims.

[0052] The 15-inch TFT LCD touchscreen is a 2.8-inch TFT LCD with a resolution of 320×240. It features a capacitive touchscreen and is embedded in a recessed area on the rear side of the housing. It is connected to the PCB adapter of the image sensor via a 40-pin 0.5mm pitch FPC cable, and is also connected to the main control PCB inside the controller via another FPC cable.

[0053] The illumination system is configured as follows: A ring-shaped LED illumination plate is fitted onto the outside of the eyepiece objective 11. Its inner diameter is 12.2 mm, outer diameter is 26 mm, and thickness is 1.0 mm. It fits the outer diameter of the eyepiece objective with a 0.1 mm gap on each side. It is fixed to the root of the eyepiece objective 11 by three evenly distributed positioning posts with a diameter of 1.5 mm and a height of 2 mm. The upper surface of the ring-shaped LED illumination plate is fitted with eight near-infrared LEDs with a wavelength of 850 nm and a power of 20 mW each, and six white LEDs with a color temperature of 5000 K and a power of 50 mW each, arranged in a staggered ring. The light-diffusing sheet is a ring-shaped frosted PC sheet with an outer diameter of 26 mm, an inner diameter of 12 mm, and a thickness of 0.5 mm, covering the front of the ring-shaped LED illumination plate. The illumination drive circuit is integrated on a small 15 mm × 10 mm PCB and fixed to the side wall of the imaging optical system housing 1 with M1.4 screws. It is connected to the contacts of the ring-shaped LED illumination plate via gold-plated spring pins.

[0054] Control and power supply connection: The handle 2 contains a lithium battery, model Panasonic NCR18650B, with a capacity of 3400mAh. The positive and negative terminals are connected to the power input socket of the main control PCB through nickel-plated phosphor bronze spring contacts. The signal processing and control unit is integrated on the main control PCB, which is fixed to four 8mm high support pillars inside the handle 2 cavity with M1.6 screws. The main control PCB is connected to the circuit boards in the imaging optical system housing 1 through a multi-core shielded cable with 10 cores, each 0.1mm thick, running through the base. 2 The total length of the connection is 150mm, and the interface adopts DF13-10S.

[0055] Work process: Step 1: Equipment startup and initialization. The operator presses the power button on the side of the handle 2 to lightly touch the switch. The signal processing and control unit is powered on. The system first performs a self-test, which takes less than 0.5 seconds: checking whether the output voltage of the liquid lens drive circuit is zero, whether the illumination LED is functioning properly through photosensitive feedback, and whether the image sensor outputs a valid image and checks the frame validity signal. After the self-test passes, the output voltage of the liquid lens drive circuit is set to 0V via the I2C bus, so that the optical power D of the liquid lens 122 is 0D, corresponding to normal vision. Near-infrared guiding illumination is then turned on, and the eight 850nm LEDs are driven by a constant current of 50mA. The touch screen display 15 displays a real-time preview image at 30fps, with the image scaled to 320×240.

[0056] Step 2: The operator holds the device, gently pressing the front end of the buffer silicone tube 14 against the patient's eye socket, and adjusts the angle so that the eyepiece 11 is aligned with the pupil. The image sensor acquires fundus images at a rate of 30 frames per second. Each frame is transmitted to the main control PCB via the MIPICSI-2 interface. The sharpness evaluation subunit uses a modified Brenner gradient function to calculate the image sharpness value S: S = Σ x，y [(I(x+2,y)-I(x,y)) 2 +(I(x,y+2)-I(x,y)) 2 The calculation area is limited to 70% of the diameter of the center of the image (excluding edge stray light).

[0057] If S is less than the preset threshold S_th=0.65, the drive control subunit sends a stepping command to the liquid lens drive circuit via I2C, causing the drive voltage to change in 5V steps, corresponding to an optical power step of approximately 0.5D. The direction is determined by the initial focus direction search algorithm, first searching in the positive direction, and then reversing if the sharpness decreases. After each step, after waiting for 50ms of liquid lens response time, the collaborative control subunit reads the actual voltage of the liquid lens drive circuit via ADC and converts it into the current optical power D (pre-calibration curve: D=0.1×V-5, where V is voltage (V) and D is in D units), and substitutes it into the preset mapping function P_target=P_base+k_focus·|D|+k_depth·L(D) to calculate the target illumination power.

[0058] In this embodiment, P_base = 2.5mW / cm 2 , k_focus=0.3, k_depth=0.15, L(D)=D 2 / 10 (unit: mW / cm) 2 The lighting drive circuit uses PID control to change the duty cycle of the white LED driver PWM, ensuring that the actual output light power reaches P_target within 10ms. This cycle of changing the optical power → adjusting the lighting power → acquiring a new frame → calculating the sharpness is repeated, iterating up to 20 times, until the sharpness value S ≥ 0.65. At this point, the current optical power D_lock and lighting power P_lock are locked, and a focus completion icon is displayed on the touchscreen.

[0059] Step 3: Exposure and Image Storage. The operator presses the shutter button, which is a two-stage button. Half-pressing locks the parameters, and full-pressing triggers the exposure. The signal processing and control unit immediately switches the illumination system from near-infrared mode to visible light mode: the near-infrared LED is turned off, and the white LED is driven by P_lock power (controlled by PWM duty cycle). At the same time, the image sensor is triggered to perform a single-frame exposure (exposure time set to 20ms, electronic rolling shutter). After the exposure is completed, the white LED is immediately turned off, and the system switches back to near-infrared guided illumination mode. The original fundus image (720×540 resolution, 12-bit RAW format) is converted to 8-bit JPEG after ISP processing and stored in the eMMC chip (32GB) on the main control PCB.

[0060] Step 4: Image display. The image is displayed synchronously on the touch screen 15. The operator can zoom with two fingers and drag with one finger to view details. They can also browse historical images through the "Album" button in the lower left corner of the touch screen.

[0061] A 5V step voltage for the liquid lens corresponds to 0.5D optical power. This relationship was obtained from the Varioptic official datasheet and actual calibration measurements on this device. The calibration method involved using a model eye ranging from -10D to +10D, measuring the optical power at which the image was sharpest under different voltages on the optical bench, and then linearly fitting the R0 value. 2 =0.998.

[0062] The sharpness threshold of 0.65 was determined through a preliminary experiment involving 50 patients with different refractive errors. The median of the sharpness distribution was subtracted from the standard deviation by 0.1, and it can be recalibrated during factory calibration according to the actual optical system.

[0063] Near-infrared LEDs have a wavelength of 850nm. The spectral sensitivity of the human eye at this wavelength is less than 1%, and the pupil constriction response is negligible. Experiments have verified that continuous irradiation for 1 minute did not cause a measurable change in pupil diameter.

[0064] Exposure time 20ms: When shooting handheld, the physiological hand tremor frequency is about 1~3Hz, and the maximum image shift within 20ms is about 0.7 pixels, calculated based on a field of view of 40° and a working distance of 20mm.

[0065] The compression of the cushioning silicone tube: When pressed against the eye socket, the silicone tube compresses by about 2mm, generating a restoring force of about 1N. This can adapt to the differences in the orbital bone contour of different patients, while avoiding hard contact that may cause discomfort to the patient.

[0066] This embodiment replaces the mechanical focusing assembly with a liquid lens, reducing the number of parts in the entire device from approximately 80 in the traditional solution to approximately 45, and reducing the volume from approximately 300 cm³ in the traditional solution. 3 Shrink to approximately 195cm 3 Dynamic illumination synergy reduced the standard deviation of fundus image brightness from ±3.2dB in the traditional constant illumination scheme to ±1.1dB in 30 clinical trials (refractive power range -8D to +6D). The success rate of a single shot (image clarity ≥0.65 and no local overexposure / underexposure) was about 23.8 percentage points higher than that of existing handheld fundus cameras.

[0067] Example 2; Based on Example 1, this embodiment adds an AI analysis module and a pupil diameter detection module, and introduces multiple working mode switching, making it suitable for more professional ophthalmological diagnostic scenarios and sensitive populations (such as children and diabetic patients).

[0068] This embodiment is used in ophthalmology clinics of tertiary hospitals or community screening sites for diabetic retinopathy (DR). The operator is an ophthalmologist or a trained screening staff member. The subjects include patients with a history of diabetes for more than 5 years (requiring high-precision diagnosis) and children aged 3 to 10 years (sensitive to light and poor cooperation). The examination environment is a clinic or screening vehicle, and low illumination (<100 lux) is allowed to avoid pupil dilation.

[0069] Based on all the structures in Example 1, an NPU chip, 2GB LPDDR4 memory, and 32GB eMMC storage were additionally soldered onto the main control PCB of the signal processing and control unit. The AI ​​analysis module runs a lightweight MobileNetV3-Small model, which is trained from the VGG16 teacher network through knowledge distillation. The number of parameters is compressed to 4.2M, the input size is 224×224, and the output categories include: normal, mild NPDR, moderate NPDR, severe NPDR, PDR, glaucoma-like changes, macular degeneration, and unqualified image quality.

[0070] The pupil diameter detection module does not add hardware and is implemented through software algorithms: using near-infrared fundus images acquired by an image sensor (under 850nm illumination, the contrast between the iris and pupil is high), the dark pupil region is first segmented using the OTSU method for binarization, then the boundary points are extracted using Canny edge detection (with high and low thresholds of 50 and 150, respectively), and finally the pupil diameter (in mm) is output by fitting a circle using the least squares method. This algorithm has a processing latency of <5ms under a 30fps image stream.

[0071] The touchscreen display 15 features a virtual mode switching button located in the upper right corner, corresponding to quick screening mode, high-precision diagnostic mode, and child mode. Mode switching is achieved by modifying configuration parameters in the EEPROM on the main control PCB.

[0072] In terms of electrical connection, both the AI ​​analysis module and the pupil diameter detection module access the image sensor data through the internal bus of the main control PCB, without adding any mechanical parts or external wiring.

[0073] Work process: The working process of this embodiment is based on the steps of Embodiment 1, with the following steps added and inserted during the second step of autofocusing and after the third step of exposure and shooting: During the second step of autofocusing, pupil diameter monitoring is performed in parallel. When the system acquires preview images using near-infrared illumination, the pupil diameter detection module calculates the pupil diameter d every 5 frames. If d < 2.5mm (small pupil), the collaborative control subunit automatically compares the output P_target of the mapping function with an upper limit value P_limit: P_limit = 1.5 + 0.5 × d (unit: mW / cm). 2The formula is derived from a comfort threshold experiment of 50 patients with different pupil diameters. If P_target>P_limit, then P_target=P_limit is forced, and a yellow pop-up window is displayed on the touch screen (15) indicating "pupil is too small, and the lighting power has been automatically reduced". If d>4.0mm (mydriatic state or extremely dark environment), then the upper limit is canceled and the standard mapping function is used. If d>6.0mm (pathological mydriasis), an additional prompt is given that the pupil is too large, and attention should be paid to excluding diseases such as glaucoma.

[0074] After the third step of exposure and shooting, an AI analysis step is added; The image sensor captures fundus images (720×540), which are first scaled down to 224×224 by the ISP, and then transmitted to the NPU. The NPU loads the MobileNetV3-Small model, completes inference within 0.8 seconds, and outputs the following results: Image quality score (0-100 points, based on the probability transformation of the "unacceptable quality" class in the output of the penultimate softmax layer of the model); If the image quality score is below 60 (e.g., due to localized overexposure caused by uneven lighting, or blurring caused by patient eye movement), the AI ​​analysis module sends a feedback signal to the collaborative control subunit via an internal message queue. The feedback signal includes the category of the quality problem (overexposure / underexposure / blur / shadow). The collaborative control subunit automatically adjusts the mapping function parameters according to the problem category. Overexposure: Reduce P_base by 0.3 mW / cm 2 ; Underexposure: Increase P_base by 0.5mW / cm 2 ; Blurry: Reduce the optical power step of the liquid lens from 0.5D to 0.2D and re-execute autofocus.

[0075] After adjustment, a second optimization acquisition is triggered—steps two and three are retried, with a maximum of two retries. If it is still unsatisfactory, a prompt will appear asking you to retake the photo.

[0076] If the image quality score is ≥60, the AI ​​analysis module will display the category with the highest probability of lesion classification and its confidence level (output when the confidence level is ≥0.7, otherwise output "uncertain") on the touch screen display 15. At the same time, annotations will be superimposed on the image (such as marking suspicious areas such as microaneurysms and bleeding points with red boxes, and the annotation positions will be obtained by mapping back to the original image through Grad-CAM heat map). Finally, the image and AI diagnosis results will be saved together in DICOM format, including structured report fields.

[0077] Operating mode switching: The operator selects the mode using virtual buttons on the touchscreen display 15. Different modes correspond to different configuration parameters (stored in EEPROM): Rapid screening mode: Sharpness threshold S_th=0.60, focus step size 1.0D (voltage step size 10V), illumination power change step size, AI inference only outputs normal / abnormal binary classification (abnormal includes any lesion category), average focus time ≤1 second, suitable for large-scale rapid screening.

[0078] High-precision diagnostic mode: S_th=0.75, focus step size 0.2D (voltage step size 2V), fine adjustment of illumination power (PWM resolution increased to 12-bit), and addition of local magnification analysis of the macula and optic disc during AI analysis (80×80 sub-images are extracted and input into the model respectively), suitable for confirmed patients.

[0079] Kids Mode: The P_base value in the mapping function is reduced by 30%, and the maximum exposure time is limited to 20ms (to reduce micro-movements in children's eyes). During autofocus, a "smile animation" (a 20×20 pixel smiley face bitmap that moves every frame) plays in the center of the touchscreen to attract children's attention. If a child's gaze is detected to deviate by more than 5°, the system will emit a soft prompt sound.

[0080] This embodiment, based on Embodiment 1, implements local AI real-time analysis and pupil-adaptive illumination limitation, and has been tested in 100 clinical cases (including 30 children, 40 diabetic patients, and 30 healthy adults): The AI ​​model achieved an accuracy rate of 91.3% in DR staging (compared with the image readings of experts from tertiary hospitals), and an AUC of 0.94 in glaucoma visual disc ratio assessment.

[0081] In scenarios with small pupils (<2.5mm), automatic amplitude limiting reduces patient discomfort.

[0082] AI-driven quality feedback loops improve the final diagnostic rate after secondary optimization and data collection.

[0083] In Kids Mode, the percentage of children who actively cooperated (continuously staring at the screen for more than 3 seconds) increased from 38% to 72%.

[0084] Example 3; This embodiment, based on Embodiment 1, adds physical operation buttons and manual adjustment functions, making it suitable for scenarios where experienced ophthalmologists wish to finely control focus parameters.

[0085] This embodiment is used in high-end ophthalmology clinics or research institutions. The operator is a senior ophthalmologist, and the examinee is a patient with complex refractive error (such as high myopia >8D) or who needs detailed examination of specific areas, such as the peripheral retina. The doctor wants to manually intervene in the focusing process to obtain the best imaging effect.

[0086] Based on the structure of Embodiment 1, a rotary encoder knob, EC11 type, with a push switch and a two-stage shutter button is added to the side of the imaging optical system housing 1, 30mm from the front end. The half-press travel is 0.5mm, the full-press travel is 1.2mm, and the trigger force is 2N.

[0087] The knob is connected to the GPIO of the main control PCB via a 4-pin ribbon cable, and the button is connected to the external interrupt pin of the main control PCB via a 2-pin ribbon cable. A virtual switch "A / M" is added to the touch screen display 15 interface to switch between "fully automatic mode" and "semi-automatic mode". The switching state is stored in the EEPROM on the main control PCB.

[0088] Work process: The fully automatic mode is exactly the same as in Example 1.

[0089] In semi-automatic mode, the signal processing and control unit no longer automatically adjusts the optical power of the liquid lens based on sharpness evaluation when selecting the semi-automatic mode via a virtual touchscreen switch. The user manually sets the optical power D by rotating the encoder knob: each clockwise rotation (30°) increases the optical power by 0.25D; each counterclockwise rotation decreases the optical power by 0.25D. The current optical power value is displayed in real time in the top status bar of the touchscreen display (15) (e.g., "D=-3.25D"). The knob has a built-in push switch; pressing it resets the optical power to 0D. The optical power adjustment range is limited to -10D to +10D. If the range is exceeded, the knob generates damping feedback (achieved by adding a PWM drive electromagnet).

[0090] The output light power of the illumination system still automatically matches the currently manually set D value according to the mapping function P_target=F(D), without user intervention. In this mode, the user can first coarsely adjust the light to be close to the subject's refractive power, for example, based on the patient's glasses prescription, and then half-press the shutter button. The system will then activate fine autofocus adjustment—at this time, autofocus searches only within ±1D of the current D value in 0.1D steps, and the illumination power changes synchronously with the fine adjustment. Full-pressing the shutter button will trigger exposure and shooting.

[0091] Multi-level working mode preset: This embodiment presets four fixed working modes, which can be switched cyclically by pressing and holding the mode switch button for 2 seconds. Each mode corresponds to a set of preset mapping function parameters (P_base, k_focus, k_depth, L(D) form), which are stored in different sectors of EEPROM.

[0092] The four modes are: Mode 1: Standard Macular Setting – Optical Power Range 0 to +2D, Low Illumination Power (P_base = 2.0mW / cm²) 2 (k_focus=0.2) to suppress foveal reflection; Mode 2: Visual inspection mode – Optical power range -2 to 0D, medium illumination power (P_base = 2.5mW / cm²) 2 (k_focus=0.3), to enhance contrast and display the edge of the display disc; Mode 3: Peripheral scan mode – Optical power range -6 to -2D, illumination power relatively high (P_base = 3.0mW / cm²) 2 (k_focus=0.4) to compensate for ambient light attenuation; Mode 4: Deep choroidal setting – Optical power range +2 to +6D, using near-infrared enhanced illumination (70% 850nm LED, 30% white light), for observing choroidal vessels.

[0093] When switching modes, the touchscreen displays the name of the current mode and the corresponding recommended application scenarios.

[0094] This embodiment provides professional doctors with manual intervention capabilities while retaining the convenience of automatic illumination matching. After trial by 10 ophthalmologists, for patients with high myopia (below -8D), manual coarse adjustment followed by half-pressing the shutter for fine focusing reduced the average focusing time from 3.2 seconds in fully automatic mode to 1.8 seconds. The preset multi-level working modes improve the image quality scores for different fundus structures, such as the macula, optic disc, peripheral vision, and choroid. The damping feedback of the rotary encoder knob and the real-time touchscreen display provide a good operating feel.

[0095] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A handheld retinal image acquisition and analysis device, characterized in that, include: The housing is composed of a handle (2) and a support frame (3). An inclined support frame (3) is provided on the left side of the top of the handle (2), and rubber blocks are provided at the bottom of the handle (2) and the support frame (3). The imaging optical system housing (1) is mounted above the handle (2) and the support frame (3) via a tube base. The imaging optical system housing (1) includes, in sequence along the optical axis, a buffer silicone tube (14), an eyepiece (11), a liquid lens group (12), an imaging lens group (13), and an image sensor. The imaging optical system housing (1) also includes a touch screen display (15), the liquid lens group (12) is composed of a first lens barrel (121) and a liquid lens sheet (122) disposed inside the first lens barrel (121), and the imaging lens group (13) is composed of a second lens barrel (131), a biconvex lens (132), a spacer (133), and a meniscus lens (134); The lighting system, located inside the eyepiece (11), includes a ring-shaped LED lighting panel, a light diffuser, and an lighting drive circuit; A liquid lens driving circuit is electrically connected to the liquid lens sheet (122) and is used to apply a driving voltage to the liquid lens sheet (122) to adjust its optical power; The signal processing and control unit is electrically connected to the liquid lens driving circuit, the illumination driving circuit, and the image sensor. The signal processing and control unit includes a sharpness evaluation subunit and a collaborative control subunit. The sharpness evaluation subunit is used to calculate the sharpness evaluation index value of the fundus image output by the image sensor. The collaborative control subunit stores a preset optical power-illumination power mapping function, which defines the target illumination light power value that the illumination system should output under different optical powers. The power module is built into the handle (2) and supplies power to the above-mentioned electrical components; The signal processing and control unit is configured to: during autofocus, generate a liquid lens drive voltage adjustment command based on feedback from the sharpness evaluation subunit to change the optical power of the liquid lens (122) in the direction of making the image sharp; and generate an illumination power adjustment command based on the current optical power value and the optical power-illumination power mapping function to dynamically adjust the output light power of the illumination system with the change in optical power.

2. The handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The optical power-illumination power mapping function is: P_target=P_base+k_focus·|D|+k_depth·L(D); Where P_target is the target illumination power, P_base is the reference illumination power, k_focus is the gain coefficient related to the amplitude of the change in optical power, D is the current optical power of the liquid lens (122), k_depth is the gain coefficient related to the imaging depth, and L(D) is the imaging depth compensation function.

3. The handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The lighting system includes a near-infrared guiding illumination source and a visible light exposure illumination source; During autofocus, the illumination system employs a near-infrared guided illumination mode; After focusing is completed and the exposure shooting stage begins, the illumination system switches to visible light exposure illumination mode, and the light power of the visible light exposure illumination is determined according to the current optical power value.

4. The handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The signal processing and control unit also includes a pupil diameter detection module, which is used to automatically calculate the pupil diameter of the subject based on the image acquired by the image sensor, and to perform amplitude limiting correction on the target illumination power output by the optical power-illumination power mapping function based on the pupil diameter.

5. The handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The signal processing and control unit also includes an AI analysis module, which is used to perform real-time analysis of the acquired retinal images based on a lightweight convolutional neural network model and to provide preliminary screening and prompts for fundus diseases. The AI ​​analysis module is also configured to evaluate the image quality of the acquired images and send a feedback signal to the collaborative control subunit when it determines that the image quality does not meet the preset conditions. The collaborative control subunit automatically adjusts the parameters of the optical power-illumination power mapping function according to the feedback signal, triggering secondary optimization acquisition.

6. The handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The device also includes a touch screen display (15) and operation buttons disposed on the imaging optical system housing (1), the operation buttons including a shutter button and a mode switching button; The signal processing and control unit is configured to switch between fully automatic and semi-automatic modes in response to operation of the mode switching button. In the fully automatic mode, the signal processing and control unit autonomously completes focusing and illumination power adjustment; In the semi-automatic mode, the user can manually set the optical power of the liquid lens (122) through the operation button, and the signal processing and control unit will automatically match the illumination power according to the manually set optical power.

7. The handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The device has multiple preset working modes, including a rapid screening mode and a high-precision diagnostic mode. Different working modes correspond to different optical power-illumination power mapping function parameters.

8. The handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The liquid lens (122) is an electrowetting type double liquid lens with an optical power adjustment range of -5D to +13D.

9. A handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The imaging optical system housing (1) is mounted above the handle (2) and the support frame (3) via a cylindrical base; The buffer silicone tube (14) is located at the front end of the imaging optical system shell assembly (1) and is used to fit the patient's eye socket; The eyepiece (11) is located behind the buffer silicone tube (14); The liquid lens group (12) and the imaging lens group (13) are sequentially disposed behind the eyepiece objective (11); The image sensor is located behind the imaging lens group (13); The touch screen display (15) is disposed on the rear end face of the imaging optical system housing (1).

10. A handheld retinal image acquisition and analysis device according to claim 1, characterized in that, The ring-shaped LED lighting plate of the lighting system is sleeved on the outside of the eyepiece (11) and located on the inner wall of the eyepiece (11), and the light-diffusing sheet covers the front of the ring-shaped LED lighting plate. The liquid lens driving circuit is fixed to the side of the imaging optical system housing (1) and is connected to the liquid lens (122) and the signal processing and control unit respectively via flexible cables.

Images