Portable quantization differential phase contrast microscopic module and deep learning reconstruction method thereof

Abstract

The invention provides a portable quantitative differential phase contrast microscopic module and a deep learning reconstruction method thereof, a detection light field passes through a modulation pattern generated by a light intensity modulation module, and then passes through a condensing lens to generate an off-axis light field, and the off-axis light field is projected to at least one detection position on an object to be detected, so that a detected object light field is generated; the image is guided to the image capturing device through the objective lens to generate a corresponding optical image. The corresponding optical image is subjected to deep learning of phase information of the corresponding detection position, and the surface topography of the object is reconstructed according to the phase information.

Description

Portable quantitative differential phase contrast microscopy module and its deep learning reconstruction method Technical Field

[0001] The present invention relates to a microscopy technology, and more particularly to a portable quantitative differential phase contrast microscopy module that utilizes deep learning to eliminate the existing tedious calculations to obtain phase information of an object under test, and a deep learning reconstruction method thereof. Background Art

[0002] Differential phase contrast (DPC) microscopy utilizes non-interference imaging to analyze object features. It produces phase contrast in label-free samples under asymmetric illumination. Due to its label-free nature, DPC has been widely used in commercial testing to observe biological specimens without photobleaching or phototoxicity.

[0003] In the field of non-visible light, there is a conventional technique for differential microscopy that uses an X-ray source and a grating to generate phase contrast. For example, Chinese Publication Application No. CN103348415 discloses an X-ray differential phase contrast imaging method according to the present invention. In order to enhance the information collected by phase contrast imaging, an analyzer grating for X-ray differential phase contrast imaging is provided with an absorption structure. The latter includes a first plurality of first regions and a second plurality of second regions, the first regions having a first X-ray attenuation, and the second regions having a second X-ray attenuation. The second X-ray attenuation is less than the first X-ray attenuation, and the first and second regions are periodically arranged in an alternating manner. A third plurality of third regions is provided with a third X-ray attenuation, which is within the range from the second X-ray attenuation to the first X-ray attenuation, wherein every second first region or every second second region is replaced by a third region.

[0004] In addition, in a typical DPC architecture using visible light for detection, a semicircular pattern is used to modulate the light source. The semicircular pattern, as shown in Figure 1A, can be modulated using a light intensity mask or a programmable LED array. Another approach is to use a spatial light modulator (SLM) or a liquid crystal panel placed in the Fourier plane of the objective lens to generate the detection light. The spectral modulation of the semicircular pattern's light intensity mask utilizes the Hilbert transform, a transformation technique that has been demonstrated to be capable of achieving an isotropic phase contrast response under the conditions of coherent laser illumination and a spatial light modulator placed in the Fourier plane of the objective lens. However, because the DPC transfer function is non-isotropic when measuring with a semicircular light intensity mask along only two axes (vertical and horizontal) under partially coherent illumination, measurements must be performed using a light intensity mask with up to twelve axial variations, such as masks 00 to 11 with different axial phase variations, as shown in Figure 1B , to increase stability and accuracy in phase reconstruction. Furthermore, the need for twelve axial variations in light intensity using a semicircular light intensity mask significantly reduces measurement efficiency, making it unsuitable for on-production line inspection.

[0005] Furthermore, existing technologies primarily focus on the design of pupil engineering algorithms to improve system image acquisition efficiency. This involves placing a digitally controlled pupil in the Fourier plane of the image and illuminating the sample with a custom-designed color pupil. The system's phase transfer function is calculated, and the sample phase is inferred from the light intensity information in each color channel, achieving phase recovery. However, for time-lapse observation of living cells, the collection of large amounts of image data is inconvenient due to the need to control different components using separate software. Furthermore, the sample's three-dimensional image information must be reconstructed using algorithms, and data processing is time-consuming.

[0006] Furthermore, prior art requires the expensive purchase of a dedicated quantitative differential phase contrast microscopy system for phase analysis. However, when not in use, the system sits unused, resulting in a waste of equipment costs. In summary, the present invention addresses both the equipment cost issue and the challenges inherent in DPC of visible light sources. Therefore, a quantitative differential phase contrast microscopy system and method are needed to address the shortcomings of prior art.

[0007] Summary of the Invention

[0008] The present invention aims to provide a portable quantitative differential phase contrast microscopy module and its deep learning reconstruction method. The module is portable and can be combined with an optical system, such as a microscope system. By utilizing the reprogrammability and parallel computing capabilities of the processing module, the various modules are integrated and controlled by a portable module based on a single processing module. This allows the modules with different functions to operate in a unified environment interface and the same programming language, making operation easier and facilitating system adjustments when replacing optical components. In addition, since the portable quantitative differential phase contrast microscopy module is externally attached to the microscope, it can be removed when phase analysis is not being performed, allowing the microscope to use its original functions. The portable and detachable design allows users to convert a conventional microscope into a quantitative differential phase contrast microscopy system without having to purchase a dedicated quantitative differential phase contrast microscopy system. Moreover, when differential phase analysis is not required, the module can be removed to restore the original microscope function, significantly reducing the cost of purchasing the equipment.

[0009] Another object of the present invention is to provide a portable quantitative differential phase contrast microscopy module and its deep learning reconstruction method, which uses the phase image obtained by traditional differential phase calculation to perform deep learning to establish a phase learning model. The phase learning model then uses the image of the object to be tested obtained by the image capture device as the input of the phase learning model. Through the deep learning process, the phase information corresponding to the detection position of the object to be tested is reconstructed to achieve rapid detection and improve analysis efficiency.

[0010] In one embodiment, the present invention provides a portable quantitative differential phase contrast microscopy module and its deep learning reconstruction method, comprising a light intensity modulation module, a focusing lens, an image capture module, and a processing module. The light intensity modulation module is configured to generate at least one modulation pattern with varying light amplitude in response to a control signal to modulate an incident light field, thereby forming a detection light field. The focusing lens is disposed on one side of the light intensity modulation module, such that the light intensity modulation module is located on the Fourier plane of the focusing lens. The focusing lens is configured to receive the detection light field and generate an off-axis light field that is projected onto an object to be measured, thereby generating a measurement light field. The image capture module is configured to receive the measurement light field and generate an optical image corresponding to the varying light amplitude. The processing module is electrically connected to the image capture module and the light intensity modulation module and controls the image capture module and the light intensity modulation module using control instructions generated in a single programming language.

[0011] In one embodiment, the operation processing module performs a phase recovery deep learning calculation on the optical image captured by the image capture module. When performing the phase recovery deep learning calculation, the operation processing module removes background information from at least one optical image generated by the image capture module, and then inputs each processed optical image into a learning model, so that the learning model processes the optical image, and then can restore at least one phase image corresponding to the surface morphology of the object to be measured.

[0012] In one embodiment, the light amplitude change is a change in the light amplitude gradient, and the light intensity modulation module selects one of a first gradient pupil mask in which the light intensity gradually decreases along a first axis, a second gradient pupil mask in which the light intensity gradually increases along the first axis, a third gradient pupil mask in which the light intensity gradually decreases along a second axis, and a fourth gradient pupil mask in which the light intensity gradually increases along the second axis. The operation processing module also receives a first optical image, a second optical image, a third optical image, and a fourth optical image of the first gradient pupil mask, the second gradient pupil mask, the third gradient pupil mask, and the fourth gradient pupil mask to perform the phase recovery deep learning to obtain a phase image of each detection position on the object to be measured.

[0013] In one embodiment, the light amplitude change is a change in the light amplitude gradient, and the light intensity modulation module generates the radial light amplitude gradient change in a manner that takes the optical axis of the incident light field as the center and a specific length as the radius to generate the modulation pattern with the light amplitude gradient change along the radial direction of the light intensity modulation module.

[0014] In one embodiment, the modulation pattern generated by the light intensity modulation module is a symmetrical black and white semicircle pattern.

[0015] In one embodiment, the present invention provides a quantitative differential phase contrast microscopy deep learning reconstruction method, comprising the following steps: first, establishing a phase learning model. Then, combining a portable quantitative differential phase contrast microscopy module with an optical system to form a quantitative differential phase contrast microscopy system. Next, placing an object under test in the quantitative differential phase contrast microscopy system. Then, using the image capture module, capturing an optical image of the object under at least one axis pupil mask generated by the light intensity modulation module. Finally, using the computational processing module, using the optical image as the input image for the phase learning model, the phase learning model generates a phase image based on the input optical image.

[0016] In one embodiment, establishing the learning model also includes the following steps: first, providing an initial learning model. Then, using a plurality of pupil masks about an axis, and using an image capture module to capture a first optical image of the object to be tested as an input image of the initial learning model. Then, using the initial learning model to learn based on the input optical image to obtain a first phase image. Then, using the pupil mask about a plurality of axes, using the image capture module to capture a second optical image of the pupil mask about the plurality of axis phases of the object to be tested. Then, using a phase algorithm to calculate the second optical image to obtain a second phase image as a reference truth. Then, the second phase image is trained with the reference truth to obtain a loss value. Finally, a weight value of the initial learning model is adjusted according to the loss value to obtain the phase learning model. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] 1A and 1B are schematic diagrams of a semicircular pupil mask and its phase changes along different axes.

[0018] FIG2A is a schematic diagram of a portable quantitative differential phase contrast microscopy module according to the present invention.

[0019] FIG. 2B is a schematic diagram of an embodiment of a portable quantitative differential phase contrast microscopy system formed by applying the portable quantitative differential phase contrast microscopy module to a microscope.

[0020] FIG. 2C is a schematic diagram of the optical architecture of the differential phase contrast microscopy system of the present invention.

[0021] 3A to 3D are schematic diagrams of a mask showing a light intensity gradient distribution along a first axis and a second axis according to the present invention.

[0022] 3E to 3H are schematic diagrams of optical images captured by a pupil mask according to the present invention with light intensity gradient distributions along the first and second axes.

[0023] FIG. 4A is a flow chart illustrating an embodiment of a quantitative differential phase contrast microscopy method according to the present invention.

[0024] FIG. 4B and FIG. 4C are schematic diagrams of different embodiments of establishing a learning model.

[0025] 5A to 5F are schematic diagrams of different modulation patterns generated by the light intensity modulation module.

[0026] 6A to 6D are schematic diagrams showing the intensity gradients of different first and second color lights generated by the light intensity modulation module along different axes to form a colorful dual-axis gradient pupil mask.

[0027] Explanation of reference numerals: 2-portable quantitative differential phase contrast microscopy module; 2a-differential phase contrast microscopy system; 20-light source; 200-incident light field; 21-light intensity modulation module; 210-detection light field; 22-condensing lens; 220-off-axis light field; 221-measurement light field; 23-objective lens; 24-image capture module; 25-tubular lens; 26-processing unit; 3-phase recovery deep learning algorithm; 30-36-steps; 430-434-steps; 530-534-steps; S-object to be measured; 90-carrying platform; 91-first gradient pupil mask; 92-second gradient pupil mask; 93-third gradient pupil mask; 94-fourth gradient pupil mask; S-object to be measured; 4-microscope; 40-carrying platform; 41-object to be measured; 42-support structure; 43-eyepiece; 44-light source; 45-objective lens system; 46-support base. DETAILED DESCRIPTION

[0028] Various exemplary embodiments will be more fully described below with reference to the accompanying drawings, some of which are shown in the accompanying drawings. However, the concepts of the present invention may be embodied in many different forms and should not be construed as limited to the exemplary embodiments described herein. Rather, these exemplary embodiments are provided so that the present invention will be detailed and complete and will fully convey the scope of the concepts of the present invention to those skilled in the art. Like numbers indicate like elements throughout. The portable quantitative differential phase contrast microscopy module and its deep learning reconstruction method will be described below using multiple embodiments in conjunction with the accompanying drawings. However, the following embodiments are not intended to limit the present invention.

[0029] Please refer to Figures 2A to 2C, where Figure 2A is a schematic diagram of a portable quantitative differential phase contrast microscopy module of the present invention; Figure 2B is a schematic diagram of an embodiment of a quantitative differential phase contrast microscopy system formed by applying the portable quantitative differential phase contrast microscopy module to a microscope; and Figure 2C is a schematic diagram of another embodiment of the optical architecture of the quantitative differential phase contrast microscopy system of the present invention. In the embodiment of Figure 2A, the portable quantitative differential phase contrast microscopy module 2 includes a light intensity modulation module 21, a focusing lens 22, an image capture module 24, and an arithmetic processing module 26. The operation of each component will be described below. The portable quantitative differential phase contrast microscopy module 2 can be combined with an optical system to form the structure shown in Figure 2B or Figure 2C. In one embodiment, the optical system can be a microscope, as shown in Figure 2B, or any device used to observe the microscopic structure of an object to be tested, such as that shown in Figure 2C.

[0030] In Figure 2B, a microscope 4 has a stage 40, on which an object to be measured 41 is mounted. A support structure 42 is provided on one side of the microscope 4, on which a light source 44 is mounted for generating a light beam. In this embodiment, the light intensity modulation module 21 and the focusing lens 22 in the portable quantitative differential phase contrast microscopy module 2 are mounted on a support base 46. The support base 46 is detachably mounted on the support structure 42, such that the light intensity modulation module 21 and the focusing lens 22 are positioned between the light source 44 and the object to be measured 41. An image capture module 24, a camera in this embodiment, is mounted on an eyepiece 43. The image capture module 24 and the light intensity modulation module 21 are electrically connected to the processing module 26. The light beam emitted by the light source 44 passes through the light intensity modulation module 21 and the focusing lens 22, and is then projected onto the object to be measured 41. The light beam passes through the object to be measured 41 and is received by the objective lens system 45 disposed below the object to be measured 41, where it is formed into an image. The image is then captured by the image capture device 24 on the eyepiece 42. The microscope 4 and its objective lens system 45 in FIG. 2B are merely schematic representations. The components of the microscope 4 and the lens structure of the objective lens system 45 are well known to those skilled in the art and are not described in detail here.

[0031] In this embodiment, the support base 46 with the light intensity modulation module 21 and the focusing lens 22, as well as the image capture module 24, are removable, thus preserving the original functionality of the microscope 4. If phase analysis is required, the support base 46 with the light intensity modulation module 21 and the focusing lens 22, as well as the image capture module 24, are reassembled onto the support structure 42. This eliminates the need for expensive, dedicated quantitative differential phase contrast microscopy systems, saving equipment costs. In addition to its portability, the processing module 26 in this embodiment controls the image capture module 24 and the light intensity modulation module 21 via control instructions generated by a single programming language. In the prior art, the computation processing module 26 is a module composed of various hardware and control languages. Although it can control the image capture module 24, the light intensity modulation module 21, and perform optical image reconstruction, because different control platforms are combined together, for example, the light intensity modulation module 21 is controlled by an embedded hardware platform such as Arduino, the reconstruction algorithm is executed on a computer using a program such as Matlab, and the image capture module 24 is controlled by a Vimba control interface. Therefore, switching between platforms and computation processing programs from controlling the mask and capturing the image to reconstructing the phase image takes a considerable amount of time.

[0032] In this embodiment, the computational processing module 26 is a field-programmable gate array (FPGA). Because optical microscopy involves extensive image processing and hardware performance is affected by hardware, the reprogrammability and parallel computing capabilities of the FPGA facilitate system development with diverse hardware components and facilitate programming adjustments in response to technological advancements. Its data processing capabilities also support the conversion of large amounts of image information. By directly integrating all hardware controls into a single control interface and using a single programming language to control the image capture module 24, the light intensity modulation module 21, and the optical image reconstruction algorithms, both software and hardware performance can be improved. Integrating the required component functions into a unified environment and programming language simplifies operation and facilitates system adjustments when optical components are replaced. In one embodiment, the programming language can be Python, but this is not a limitation.

[0033] The following describes the operation of the portable quantitative differential phase contrast microscopy module 2, combined with an optical system to form a quantitative differential phase contrast microscopy system 2a, as shown in Figure 2C. The quantitative differential phase contrast microscopy system 2a includes a light source 20, an objective lens 23 with adjustable magnification, and a support stage 90 for supporting an object S. In this embodiment, the light intensity modulation module 21, focusing lens 22, image capture module 24, and processing module 26 are integrated into a portable modular architecture. Users can carry the portable quantitative differential phase contrast microscopy module 2 to any optical system and, when combined with the optical system, achieve the effect of quantitative differential phase contrast microscopy analysis. The light source 20 is used to generate an incident light field 200. In this embodiment, the light source is a broadband visible light source, but this is not limiting. For example, a monochromatic visible light source can also be used. The light intensity modulation module 21, disposed on one side of the light source 20, generates a light intensity gradient distribution that modulates the incident light field 200 according to a control signal. In one embodiment, the light intensity modulation module 21 is a liquid crystal module (TFT shield) or a reflective liquid crystal on silicon (LCoS) module that can control the intensity of light transmission. These modules are characterized by their ability to control the orientation of the liquid crystal, thereby varying the intensity of transmitted or reflected light, thereby generating pupil masks with varying axial orientations. For example, in one embodiment, at least one modulation pattern with varying light amplitude is generated in response to a control signal, such as the semicircular pupil mask shown in FIG1A .

[0034] In another embodiment of the present invention, the light intensity modulation module 21 can generate a pupil mask with a gradient distribution of light amplitude along different axes according to a control signal to modulate the incident light field 200 into a detection light field 210 with an intensity gradient distribution, for example, as shown in Figures 3A to 3D. Figure 3A shows a first gradient pupil mask with gradually decreasing light intensity along a first axis, Figure 3B shows a second gradient pupil mask with gradually increasing light intensity along the first axis, Figure 3C shows a third gradient pupil mask with gradually decreasing light intensity along a second axis, and Figure 3D shows a fourth gradient pupil mask with gradually increasing light intensity along the second axis. In this embodiment, the first axis is the X-axis and the second axis is the Y-axis, but this is not a limitation. The mask with a light intensity gradient distribution can solve the problem of intensity jump (amplitude cross) generated in the middle axis when measuring each axis using a semicircular bright and dark mask. It should be noted that the light intensity modulation module 21 can also generate various semicircular pupil masks as shown in FIG. 1A and FIG. 1B according to the control signal.

[0035] The condenser lens 22 is disposed on one side of the light intensity modulation module 21 to receive the detection light field 210 modulated by the light modulation module 21. The light intensity modulation module 21 is located on the Fourier plane of the condenser lens 22. The condenser lens 22 is used to receive the detection light field 210 and generate an off-axis light field 220 that is projected onto the object to be detected S on the support platform 90, thereby generating a detection object light field 221. The object to be detected can be a transparent cell or a transparent living cell, etc. The objective lens 23 is disposed on one side of the condenser lens 22 so that the object to be detected S is located at the focal length of the objective lens 23. The objective lens 23 receives the detection object light field 221 that penetrates the object to be detected S. It should be noted that the architecture of the present invention can generate optical images with partially coherent illumination. In one embodiment, the condition for generating partially coherent illumination is that the condenser lens 22 has a first numerical aperture (NA) value, the objective lens has a second NA value, and the ratio of the first to the second NA value is 1 or approximately 1. Utilizing partially coherent illumination can produce effects superior to coherent illumination in the prior art, such as enhanced resolution and reduced coherent speckle noise.

[0036] The image capture module 24 is coupled to the objective lens 23 to receive the object light field 221 and generate an optical image corresponding to the intensity gradient distribution. The distance between the image plane of the image capture module 24 and the object S to be measured is twice the focal length of the objective lens 23 and twice the focal length of the lens 25. In this embodiment, the image capture module 24 is coupled to the objective lens 23 via a tube lens 25. The objective lens 23 and tube lens 25 in this embodiment are components of a microscope system. The microscope system can be a commercial microscope system, such as, but not limited to, a Leica DMI3000 device.

[0037] The processing module 26 is electrically connected to the light intensity modulation module 21 to generate control signals to control the light intensity modulation module 21 to generate masks along different axial light intensity gradients, such as a first gradient pupil mask with gradually decreasing light intensity along a first axial direction, a second gradient pupil mask with gradually increasing light intensity along the first axial direction, a third gradient pupil mask with gradually decreasing light intensity along a second axial direction, and a fourth gradient pupil mask with gradually increasing light intensity along the second axial direction, as shown in FIG3A to FIG3D . Furthermore, the processing module 26 is also electrically connected to the image capture module 24 to receive the first, second, third, and fourth optical images of the first, second, third, and fourth gradient pupil masks 91, 92, 93, and 94 captured by the image capture module 24 to perform calculations to obtain the phase of each detection position on the object to be detected, thereby reconstructing the surface topography or internal structural features of the object to be detected.

[0038] In one embodiment, as shown in FIG4A , the figure is a flow chart of an embodiment of the quantitative differential phase contrast microscopy method of the present invention. In this embodiment, unlike the prior art, the phase image reconstruction algorithm after the optical image is intercepted in this embodiment does not utilize the existing phase image reconstruction algorithm, because the existing reconstruction algorithm is time-consuming to calculate and requires good computer equipment, which increases the burden on the user in terms of time and money costs. Therefore, in this embodiment, the deep learning method used, through a pre-established learning model, can convert the asymmetric optical phase image (anisotropic quantitative phase images) of the image captured by the image capture module 24 into a symmetric phase image (isotropic quantitative phase image) without using cumbersome algorithms, and directly output the phase image corresponding to the optical image with the learning model, thereby improving the inefficiency and cost of the existing algorithm.

[0039] When the processing module 26 performs the phase recovery deep learning algorithm 3, it first performs step 30 to establish a phase learning model. The learning model of this embodiment is a U-Net learning model. Establishing the learning model depends on the pupil mask to be used. For example, if the pupil mask to be used is the existing 12-axis semicircular mask, as shown in Figures 1A and 1B, then the learning model must be established using the semicircular mask; if a gradually changing pupil mask is to be used, as shown in Figures 3A to 3D, then the learning model is established using the gradually changing pupil mask.

[0040] First, the learning model established by the semicircular pupil mask shown in Figures 1A and 1B is described. As shown in Figure 4B, this figure is a flow chart of an embodiment of establishing a learning model. First, step 430 is performed using a symmetrical semicircular pupil mask with an open axis (for example, the symmetrical (00 and 06 or 01 and 07) in Figure 1B), and an image capture module is used to capture an optical image of the object to be tested, such as a thin transparent cell or other living cell. Then, a phase image is obtained through a phase algorithm as the input image of the initial learning model. Then, step 431 is performed, and the initial learning model learns based on the input image to obtain a first phase image.

[0041] Next, step 432 is performed, where the image capture module 24 captures optical images of the thin transparent cell relative to the 12-axis pupil mask shown in FIG1B . The image capture module 24 then uses a phase algorithm, such as quantized differential contrast (qDPC), to calculate the 12 optical images, resulting in a second phase image serving as the ground truth. The calculation method is conventional and will not be described in detail here. Next, step 433 is performed, where the first phase image and the second phase image are trained to obtain a loss evaluation. Next, step 434 is performed, where a weight value of the initial learning model is adjusted based on the loss value to obtain a phase learning model. The traditional single-axis semicircle is used to establish a U-net learning model. This involves capturing one-axis (two images) reconstruction as input and 12-axis (24 images) reconstruction as the ground truth to train the U-net learning model. The resulting phase image is then used to improve accuracy. This has the advantage of reducing the time required to capture optical images. For example, existing technology requires 12 axes (24 images), while the establishment of a learning model only requires 2 optical images.

[0042] In another embodiment of establishing a phase learning model, a learning model is established using a gradient pupil mask, as shown in Figures 3A to 3D. Figure 4C is a schematic flow chart of an embodiment of a method for establishing a learning model. The method first proceeds to step 530, capturing optical images of a thin transparent cell or living cell with respect to a biaxial gradient pupil mask, such as the upper two images or the lower two images of the pupil mask in Figures 3A to 3D, as input images for the initial learning model. Next, step 531 is performed, where the initial learning model is subjected to computational noise removal and phase algorithm calculations to obtain a first phase image.

[0043] Then, step 532 is performed to obtain four optical images of a biaxially gradient pupil mask for the thin transparent cell, as shown in Figures 3E-3H. A second phase image is obtained through a phase algorithm as a ground truth. Then, step 533 is performed to train the first phase image and the second phase image to obtain a loss value. Then, step 534 is performed to adjust a weight value of the initial learning model based on the loss value to obtain a phase learning model. It should be noted that the above two methods for establishing the phase learning model are not limited to the pupil masks of Figures 1A-1B or Figures 3A-3D. For example, in one embodiment, the light intensity modulation module 21 can also be controlled to generate a pupil mask that varies in a circumferential direction, as shown in Figures 5A-5F, which are schematic diagrams of a pupil mask with a circumferential variation. The light intensity modulation module 21 generates a radial light amplitude gradient variation by generating a modulation pattern with a light amplitude gradient variation along the circumference of the light intensity modulation module 21, centered around the optical axis of the incident light field and with a radius of a specific length. In another embodiment, the color dual-axis gradient pupil mask shown in Figures 6A-6D is similar to Figures 3A-3D, except that different colored light fields are added to give the pupil mask a gradient color change. Using the dual-axis gradient pupil mask to establish a U-net phase learning model involves extracting one axis (two images) after background removal (without reconstruction) as input and reconstructing two axes (four images) as the baseline truth to train the U-net phase learning model. This model can replace the traditional method of directly reconstructing images using mathematical software. This has the advantage of reducing the number of acquisitions, eliminating the time required for parameter adjustment and reconstruction.

[0044] Returning to FIG. 4A , in this embodiment, the phase learning model in step 30 is the phase learning model established in FIG. 4C . Next, step 31 is performed to combine the portable quantitative differential phase contrast microscopy module with an optical system to form a quantitative differential phase contrast microscopy system. In one embodiment, the portable quantitative differential phase contrast microscopy module can be the portable quantitative differential phase contrast microscopy module 2 shown in FIG. 2A . The optical system can be a microscope or other optical device that can be used for microscopic observation. After the optical system is combined with the portable quantitative differential phase contrast microscopy module 2, in one embodiment, it can be the quantitative differential phase contrast microscopy system 2a shown in FIG. Next, step 32 is performed to place an object to be measured, such as a thin transparent cell or a transparent living cell, in the quantitative differential phase contrast microscopy system 2 shown in FIG. Next, step 33 is performed to capture an optical image of the thin transparent cell along one axis of any two-axis gradient pupil mask using the image capture module 24, such as one of FIG. 3A to FIG. 3D . Next, step 34 is performed, where the processing module 26 performs background and noise removal on the at least one optical image generated by the image capture module 24. Next, step 35 is performed, where the processed optical image is used as the input image for the phase learning model. The phase learning model generates a phase image based on the input optical image. Finally, step 36 is performed, where the surface topography of the thin transparent cell sample is reconstructed using the phase image.

[0045] The foregoing merely describes preferred embodiments or examples of the technical means employed by the present invention to address the problems to be solved, and is not intended to limit the scope of the present invention. In other words, all equivalent variations and modifications consistent with the scope of the present invention or within the scope of the present invention are encompassed by the present invention.

Claims

1. A portable quantitative differential phase contrast microscopy module, characterized in that: include: a light intensity modulation module for generating at least one modulation pattern with light amplitude variation according to a control signal to modulate an incident light field so that the incident light field forms a detection light field; a condenser lens disposed on one side of the light intensity modulation module so that the light intensity modulation module is located on the Fourier plane of the condenser lens, the condenser lens being used to receive the detection light field and generate an off-axis light field to be projected onto an object to be measured, thereby generating a detection light field; an image capture module for receiving the object light field and generating an optical image corresponding to the change in the light amplitude; and A calculation processing module is electrically connected to the image capture module and the light intensity modulation module. The calculation processing module controls the image capture module and the light intensity modulation module through control instructions formed by a single programming language.

2. The portable quantitative differential phase contrast microscopy module according to claim 1, wherein: The operation processing module performs a phase recovery deep learning calculation on the optical image captured by the image capture module.

3. The portable quantitative differential phase contrast microscopy module according to claim 2, wherein: When performing the phase recovery deep learning calculation, the operation processing module removes background information from at least one of the optical images generated by the image capture module, and then inputs each processed optical image into a phase learning model, so that the phase learning model processes the optical image and can restore at least one phase image corresponding to the surface morphology of the object to be tested.

4. The portable quantitative differential phase contrast microscopy module according to claim 1, wherein: The light intensity modulation module is a liquid crystal module or a light reflective liquid crystal module for controlling light penetration intensity, and has a liquid crystal unit therein for changing light transmission according to a control signal.

5. The portable quantitative differential phase contrast microscopy module according to claim 1, wherein: The condenser lens has a first numerical aperture value, the objective lens has a second numerical aperture value, and a ratio of the first numerical aperture value to the second numerical aperture value is within a range of partially coherent illumination.

6. The portable quantitative differential phase contrast microscopy module according to claim 6, wherein: The ratio of the first numerical aperture value to the second numerical aperture value is 1.

7. The portable quantitative differential phase contrast microscopy module according to claim 1, wherein: The light amplitude change is a change in the light amplitude gradient, and the pupil mask is one of a first gradient pupil mask in which the light intensity gradually decreases along a first axis, a second gradient pupil mask in which the light intensity gradually increases along the first axis, a third gradient pupil mask in which the light intensity gradually decreases along a second axis, and a fourth gradient pupil mask in which the light intensity gradually increases along the second axis.

8. The portable quantitative differential phase contrast microscopy module according to claim 1, wherein: The light amplitude variation is a variation of a light amplitude gradient, and the light intensity modulation module generates a pupil mask having a light amplitude gradient variation along a circumferential radial direction.

9. The portable quantitative differential phase contrast microscopy module according to claim 1, wherein: The modulation pattern generated by the light intensity modulation module is a pupil mask of a symmetrical black and white semicircle pattern.

10. A quantitative differential phase contrast microscopy deep learning reconstruction method, characterized in that: Includes: Establish a one-phase learning model; Combining a portable quantitative differential phase contrast microscopy module with an optical system to form a quantitative differential phase contrast microscopy system; placing an object to be tested in the quantitative differential phase contrast microscopy system; capturing an optical image of the object under test under the at least one-axis pupil mask generated by the light intensity modulation module using the image capturing module; and The operation processing module uses the optical image as an input image of the phase learning model, and the phase learning model generates a phase image according to the input optical image.

11. The quantitative differential phase contrast microscopy deep learning reconstruction method according to claim 10, characterized in that: Building this learning model also includes: Providing an initial learning model; Using a plurality of pupil masks about an axis, and using an image capture module to capture a first optical image of the object to be tested as an input image of an initial learning model; Using the initial learning model to learn according to the input optical image to obtain a first phase image; Using the pupil mask along a plurality of axes, the image capture module captures a second optical image of the object under test with respect to the pupil mask along the plurality of axes; Using a phase algorithm to calculate the second optical image to obtain a second phase image as a reference truth; Training the second phase image with the ground truth to obtain a loss value; as well as A weight value of the initial learning model is adjusted according to the loss value to obtain the phase learning model.

12. The quantitative differential phase contrast microscopy deep learning reconstruction method according to claim 10, wherein: The light amplitude change is a change in the light amplitude gradient, and the pupil mask is one of a first gradient pupil mask in which the light intensity gradually decreases along a first axis, a second gradient pupil mask in which the light intensity gradually increases along the first axis, a third gradient pupil mask in which the light intensity gradually decreases along a second axis, and a fourth gradient pupil mask in which the light intensity gradually increases along the second axis.

13. The quantitative differential phase contrast microscopy deep learning reconstruction method according to claim 10, characterized in that: The light amplitude variation is a variation of a light amplitude gradient, and the light intensity modulation module generates a pupil mask having a light amplitude gradient variation along a circumferential radial direction.

14. The quantitative differential phase contrast microscopy deep learning reconstruction method according to claim 10, characterized in that: The modulation pattern generated by the light intensity modulation module is a pupil mask of a symmetrical black and white semicircle pattern.

Images