A device for real-time detection and isolation of circulating tumor microemboli (CTMS), and an operating method of said device

Abstract

The invention relates to a lensless holographic microscope-integrated microfluidic device for the real-time detection and isolation of circulating tumor microemboli (CTMs), which are extremely rare in blood, and the operating method of said device. The device according to the invention operates on a microfluidic chip (10) using a semi-automated method based on filtration and image processing, and CTMs are imaged and automatically detected without any staining method by means of a trained neural network.

Description

[0001] DESCRIPTION

[0002] A DEVICE FOR REAL-TIME DETECTION AND ISOLATION OF CIRCULATING TUMOR MICROEMBOLI (CTMs), AND AN OPERATING METHOD OF SAID DEVICE

[0003] Technical Field of the Invention

[0004] The invention relates to a lensless holographic microscope-integrated microfluidic device for the real-time detection and isolation of circulating tumor microemboli (CTMs), which are extremely rare in blood, and to the operating method of said device. The device according to the invention operates on a microfluidic chip using a semi-automated method based on filtration and image processing, and CTMs are imaged and automatically detected without any staining method by means of a trained neural network.

[0005] State of Art

[0006] Cancer is responsible for the death of 1 in every 6 individuals globally, ranking as the second leading cause of disease-related deaths after cardiovascular diseases. Notably, approximately 90% of cancer-related deaths result from the disease metastasizing from the initial tumor tissue to other parts of the body. Cancer is currently diagnosed through radiographic imaging methods or serum tumor markers as part of health screenings. However, these methods are only used for diagnostic purposes after the onset of certain symptoms. It is believed that cancer cells detached from the primary cancer tissue enter the bloodstream long before symptoms appear. Existing methods are generally unable to detect cancer cells responsible for metastasis at the cellular level. These cells can be found in the blood as single cells, as circulating tumor cells (CTCs) or as circulating tumor microemboli (CTMs), which are clusters of two or more CTCs isolated from the cancer tissue. Circulating tumor microemboli (CTM) are structures composed of two or more tumor cells. CTMs are much rarer in the blood than circulating tumor cells (CTCs) and are known to be much more aggressive in cancer metastasis. Hence, in the early detection of cancer, the efficient and rapid detection and isolation of the cells with the highest metastatic potential in the blood plays a vital role in understanding the disease in depth and developing preventive methods and more effective drugs.

[0007] Circulating tumor microemboli (CTMs) represent the cancer cells that have entered the bloodstream and formed small emboli (clots) within blood vessels. These microemboli result from cancer cells detaching from the tumor and aggregating, leading to micro-level blockages in the circulatory system. CTMs play a critical role in the process of cancer spreading to different parts of the body, known as metastasis. One of the methods used to diagnose CTMs is liquid biopsy and CTMs are detected through blood samples. Liquid biopsy enables the examination of cancer cell DNA, extracellular vesicles, and other tumor- associated biomolecules, and this method is less invasive than normal biopsy and is often preferred for monitoring disease progression. However, CTMs are usually present in low concentrations in the bloodstream, making them difficult to detect. Liquid biopsy may lack the sensitivity necessary to detect these cells, especially during early-stage cancers or when only a few CTMs are present. Another diagnostic method, immunohistochemistry, detects CTMs using antibodies specific to certain tumor antigens (e.g. EpCAM). However, in some cases, CTMs express these antigens at low levels or not at all, making it difficult to detect cells and leading to false negative results. Next-generation sequencing (NGS) and PCR are also utilized for CTM detection in the state of the art. Nevertheless, advanced methods such as NGS are often costly, require sophisticated laboratory infrastructure, and demand highly specialized personnel. In addition, the amount of data generated by molecular genetic methods is huge, and the management, analysis and interpretation of this data is complex and time-consuming. In addition to the aforementioned techniques, microfluidic devices have been employed to isolate circulating tumor cells and microemboli with high precision. In addition, the integration of machine learning into microfluidic devices, which is now used in almost every field, increases the functionality and accuracy of these devices, making the detection of circulating tumor microemboli (CTM) more sensitive, fast and automated. This integration involves using machine learning algorithms to analyze large volumes of data collected by the microfluidic devices, recognize patterns, and accurately classify cells. In the state of the art, patent application US2023258554A1 discloses a technique for the holographic imaging and classification of cells. The system described in the aforementioned document includes a holographic imaging module configured and operable to display the flow of a heterogeneous population of cells; a cell sorting module for sorting the cell flow; a control unit configured to receive image data indicative of cell flow from the holographic imaging module; and a cell extraction module for automatically processing the image data, identifying a particular cell type during flow, and providing real-time, automated, label-free holography-activated sorting of cells upon identification of a particular cell type. Said system enables label-free imaging and sorting of circulating tumor cells in blood using a cell extraction module and label-free interferometric phase microscopy. Here, cell classification is achieved through machine learning. Furthermore, single-cell holograms are acquired during flow and analyzed in real time through image processing and machine learning. In this document, cells are detected and classified on a standard single-channel microchip. The cells are then subjected to dielectrophoresis-based separation while under flow. Patent application CN113588522A in the state of the art relates to a method and system for tumor detection and classification in circulation, based on microfluidics and image recognition within the field of image classification and recognition. In the present document, the circulating tumor intelligent recognition module is equipped with an image recognition module, and the image recognition module uses an artificial intelligence computing network (e.g., machine learning or deep learning) to detect circulating tumor cells. It is stated that cell staining is not required in this system. The imaging tool employed in this document is a light microscope and this imaging tool is not integrated into the device itself and is able to perform large-scale imaging through lenses. Circulating tumor cells detected by an artificial intelligence computing network (e.g. machine learning or deep learning) are diverted to a different outlet by flow control of a solenoid valve used without labels, without purification of microemboli.

[0008] Limitations and inadequacies of existing technical solutions includes low sensitivity in CTM isolation; high cost and prolonged processing times of the used methods; inability to detect cancer cells responsible for metastasis at the cellular level; determination of morphological characteristics of circulating tumor microemboli by staining; CTM identification based solely on size; low-efficiency sorting and consequent limitation of use in personalized therapies; as well as the inability to always distinguish CTMs from other blood cells due to false positive / false negative results in image processing methods used in a simple way; lack of standardization for different devices and assay media; background noise generated by other cells and particles in the images, and these reasons necessitated an improvement in the devices used for the detection and separation of circulating tumor microemboli.

[0009] Brief Description and Objects of the Invention

[0010] This invention describes a lensless holographic microscope-integrated microfluidic device for the real-time detection and isolation of circulating tumor microemboli (CTMs), which are extremely rare in blood, and the operating method of said device. The lensless holographic microscopy integrated device according to the device semi-automatically filters, purifies and separates circulating microemboli.

[0011] The object of the invention is the real-time detection and isolation of circulating tumor microemboli present in the blood. In this invention, CTMs captured on a microfluidic filter are detected in real time through image processing algorithms and transferred to a collection reservoir. Detection of microemboli is mainly based on the algorithm's detection of the inactivity in which CTMs, which are mobile in the area from the cell inlet channel to the filter structures, get stuck in the filters depending on their size when they arrive on the filter and finally remain immobile. The developed algorithm employs image processing tools developed through numpy, glob, serial, time and openCV libraries. The developed image processing algorithm consists of 5 main steps: 1. Identifying and filling CTMs as solid objects based on their morphology, 2. Detecting all moving objects and displaying them with bounding boxes, 3. Determining immobilization of moving objects on the filter, 4. Initiating forward and reverse washing upon detection of immobility, 5. Repeating the entire process.

[0012] Another object of the invention is to enable high-efficiency and high-sensitivity detection of cancer cells responsible for metastasis at the cellular level. In the device according to the invention, CTMs, which are separated according to their size in the filters on the microfluidic chip, are detected in real time with the image processing algorithm. Upon detection, the valve structures and syringe pumps on the microfluidic chip are automatically controlled in processes I, II and III. These washing processes ensure that CTMs are isolated with high efficiency and precision.

[0013] Another object of the invention is to identify CTMs based on both their size and morphological characteristics. In the device according to the invention, CTMs are determined not only by size, but also by shape characteristics by using YOLOv5-based deep learning algorithms that can perform image analysis and CTM segmentation, trained specific to CTMs and developed by us with the ability to identify CTMs according to their shape, structure and size.

[0014] An object of the invention is the real-time detection and separation of circulating tumor microemboli (CTM) at low cost. The present invention is not dependent on bulky and expensive microscopes. In the present invention, image analysis is performed with a lensless holographic microscope integrated into a microfluidic chip. The components of a lensless holographic microscope consist of an image sensor, a light source and a pinhole. It does not include expensive components such as objectives, lens and mirror structures. In this way, it is a system with low production costs, low energy consumption and no need for trained personnel.

[0015] An object of the invention is to identfy CTMs without the need for staining. By employing the lensless holographic microscopy system in the invention, CTMs can be identified based on their morphological characteristics without requiring staining procedures. Descriptions of the Figures

[0016] Figure 1. (A) Schematic view of the microfluidic chip and the structures managed during the semi-automated cell separation, purification and isolation protocols, (B) Top view of the filter structure.

[0017] Figure 2. Schematic representation of the lensless holographic microscope setup.

[0018] Figure 3. (A) CTM separation efficiency and (B) separation efficiency of U937 cells at different flow rates of the microfluidic filter.

[0019] Figure 4. Time required to remove leukocytes from the microfluidic channel (Process II) at different flow rates.

[0020] Figure 5. (A) Injection of the cell solution into the microfluidic channel at 1 mL / h (Process I) (B) Removal of leukocytes from the microfluidic channel at 0.5 mL / h to increase the separation purity of CTMs (Process II) (C) Directing the CTMs retained in the filter to the reservoir at 1 mL / h without loss of integrity.

[0021] Figure 6. Automatic CTM isolation protocol on the filter structure positioned above the image sensor (A) Introduction of the cell mixture at 1 mL / h into the microfluidic channel (B) Automatic detection of CTMs trapped on the filter using image processing algorithms (C) Enhancement of CTM separation purity by washing leukocytes from the microfluidic channel (D) Transfer of the trapped CTMs to the reservoir by reverse washing.

[0022] Figure 7. CTM separation efficiencies. Non-automated and automated microfluidic filter efficiencies at flow rates of 0.25-4 mL / h.

[0023] Figure 8. Detection of CTMs transferred to the reservoir with deep learning algorithms (A) Holographic microscope image of CTMs and U937 cells in the reservoir, (B) Fluorescent microscope image of CTMs. (C) Classification and validation of CTMs and U937 cells using a deep learning based algorithm.

[0024] Figure 9. Detection of (A) 12% CTM pure dilute (B) 7% pure concentrated samples collected in the reservoir using deep learning algorithms.

[0025] Figure 10. Confusion matrix for CTM, U937 and background.

[0026] Description of the Reference Numbers in Figures

[0027] 1. Syringe pump (S) 2. Valve (V)

[0028] (V1)

[0029] (V2)

[0030] (V3)

[0031] (V4)

[0032] (V5)

[0033] 3. Outlet channel (W)

[0034] 4. Microfilter

[0035] 5. Reservoir

[0036] 6. Microemboli (CTM)

[0037] 7. Leukocyte

[0038] 8. Light source

[0039] 9. Pinhole

[0040] 10. Microfluidic chip

[0041] 11. Image sensor

[0042] Detailed Description of the Invention

[0043] The invention relates to a lensless holographic microscope-integrated microfluidic device for the real-time detection and isolation of circulating tumor microemboli (CTMs), which are extremely rare in blood, and to the operating method of said device. The lensless holographic microscope-integrated device accoding to the invention enables semiautomated filtration, purification, and isolation of circulating microemboli.

[0044] The device according to the invention offers a semi-automated system based on filtration and image processing on a microfluidic chip (10) to separate CTMs from leukocytes based on their size and morphology. The CTMs, which are filtered according to their size in microfilters (4) in the microfluidic chip (10), are instantly observed under a lensless holographic microscope system and transferred to the collection reservoir (5) on the microfluidic chip to prevent the CTMs from degrading their structure under flow. For this process, the CTM in the sample trapped in the filter is automatically determined by image processing algorithms and the sample flow on the microfluidic chip is stopped. Then, the leukocytes in the sample in the vicinity of the microfilter (4) are removed with washing solution and the CTMs are transferred to the collection reservoir. The automation of these processes is achieved through the syringe pumps and active valve (2) structures (V1 , V2, V3, V4, and V5) on the microfluidic chip (10) and through the software for their automatic control. This software automatically repeats these processes until the sample is depleted. During these processes, some residual leukocytes (7) that cannot be fully removed from the area surrounding the microfilters (4) may be transferred to the collection reservoir (5) along with CTMs. To address this, the reservoir (5) is imaged using the lensless holographic microscopy system, and deep learning algorithms are used to automatically detect the CTMs. Identified CTMs can then be collected with the aid of a micromanipulator.

[0045] A semi-automated device based on filtering and image processing for real-time detection and differentiation of rare circulating tumor microemboli (CTM) in blood according to the invention comprises at least three syringe pumps (1 ) for injecting solutions; at least five valves (2) for controlling the flow direction of the solutions; at least one outlet channel (3); at least two microfilters (4) for the removal of leukocytes; at least one reservoir (5) on the microfluidic chip (10) in which the CTMs filtered according to their size in microfilters (4) in the microfluidic chip (10) are instantaneously observed under a lensless holographic microscope system and transferred to prevent the CTMs from degrading their structure under flow; at least one light source (7) for illuminating the sample for imaging; at least one pinhole (8) for spatial filtering of the light; at least one microfluidic chip (10) on which filtering and image processing is performed to isolate CTMs from leukocytes according to their size and morphology and at least one image sensor (1 1 ) for capturing holographic images of the sample. The light source (8), image sensor (11 ), and pinhole (9) collectively constitute the lensless holographic microscopy system.

[0046] The operating method of the lensless holographic microscope-integrated microfluidic device of the invention comprises the following process steps: i. spatially filtering the light by a pinhole (8) placed in front of the light source (8) in the lensless holographic microscope, ii. automatically controlling the syringe pump (1 ) and the valve (2) structure via the microcontroller using image processing algorithms, iii. switching all syringe pumps (S1 , S2 and S3) and valve (2) structures (V1 , V2, V3, V4 and V5) to the closed state in the system ready position, iv. injecting (process I) the sample containing the cells into the system with S1 by switching the structures V2 and V3 on the microfluidic chip (10) to the opened state, v. real-time processing the microfluidic filters by image processing algorithms through a lensless holographic microscope, vi. automatically stopping the injection of S1 and switching the structure V3 to the closed state by inserting at least one CTM into the microfilter (4) structures on the microfluidic chip (10), vii. injecting the solution via S2 by automatically switching the structure V5 to the opened state for the purification (process II) of CTMs captured on the microfilter (4) within the microfluidic channel from leukocytes, viii. automatically stopping S2 and switching V5 and V2 to the closed state at the end of process II, ix. injecting the solution with S3 by automatically switching the structures V1 and V4 to the opened state in order to direct the CTMs captured in the microfilter (4) structures on the microfluidic chip (10) to the reservoir (process HI), x. switching the system to the initial state by automatically stopping S3 at the end of process III and switching the structures V1 and V4 to the closed state, xi. repeating this cycle in the process steps (i-x) until the cell-containing sample is exhausted, xii. manually placing the reservoir (5) region on the microfluidic chip (10) onto the image sensor (1 1 ) after the cell solution sample is exhausted, xiii. identifying CTMs by deep learning algorithms while observing the reservoir (5) in real time with a lensless holographic microscope (process IV).

[0047] The microfluidic chip (10) used in the separation process of the invention consists of three layers of polydimethylsiloxane (PDMS) and a glass surface. There are at least five active valves (5) that can be deformed by air pressure to control the liquid flow in the channels connecting 1 inlet, 3 outlets and 1 reservoir outlet on the microfluidic chip to the main channel where the filters are located. The top PDMS layer contains air pressure-driven control channels for valve regulation. The middle and bottom layers contain the fluidic channels. The bottom PDMS layer houses the fluid transfer passages (facilitating liquid transition between the middle layer and bottom layer) and the microfilter structures for separating CTMs. The filter structures in the microfluidic chip are designed to allow leukocytes consisting of single cells (15 pm) to pass through the microfluidic flow while allowing CTMs consisting of 2-100 cells (>25 pm) to attach (Figure 1 B). The final spacing between oval-shaped filters is set to 10-20 pm, with an optimum of 18 pm, and branching structures are used to ensure homogeneous distribution of cells in the channels coming into the filter zone. To fabricate the final microfluidic chip (10), SU-8 micro-patterns on Si flakes were fabricated in a clean room by photolithography technique from mask layers containing designed fluid inlet channels, valve control chambers, microfluidic filter structures and passages. Using these molds, a three-layer PDMS chip was fabricated through casting / spin coating. The PDMS layers and glass were bonded together by air plasma treatment and the channel inlets / outlets (1 mm) and reservoir (5) (4 mm) were drilled with a hole punch.

[0048] As shown in Figure 1 A, the microfluidic chip (10) includes 3 inlets, 1 outlet, and a reservoir (5). Each of the 3 inlets is connected to a syringe pump (1 ), and the syringe pump 1 (S1 ) contains the cell mixture, while the syringe pump 2 (S2) and the syringe pump 3 (S3) contain culture medium solutions. Flow direction in each channel is controlled by valves (V). In the valve composed of 3 PDMS layers, the flow in the microfluidic channel is ensured by the PDMS membrane opening and closing the channel by controlling the compressed air in the valve control room. By controlling these structures, the cell solution visualized on the microfluidic chip (10) with a lensless holographic microscope platform will be detected as the CTMs are captured while passing through the filter and the flow will be stopped, the channel will be washed free of other cells with cleaning solution to increase the separation purity, and the captured cells will be directed to the reservoir by reverse flow. After the sample is fully processed, the collected cells in the reservoir (5) are imaged and the CTMs will be collected by capillary connected to the micromanipulator and collected into a microcentrifuge tube.

[0049] The CTM separation process was carried out on the chip schematically illustrated in Figure 1 A. The inlet S1 is connected to the syringe pumps S2 and S3. The syringe pump (syringe pump 1 ) connected to the inlet S1 contains the cell mixture and the syringe pumps (syringe pump 2 and the syringe pump 3) connected to the inlet S2 and inlet S3 contain the culture medium. For the protocol, the filter structure is positioned directly above the image sensor. To pass the cell mixture through the filter, the valve V3 and the valve V2 are opened while all other valves remained closed, and the syringe pump 1 is activated (Process I, Figure 1 A). In the meantime, if a CTM is attached to the real-time imaged filter, it is detected by the developed image processing algorithm. Upon detection, the syringe pump 1 is stopped and only the valves V5 and V2 are opened, while the syringe pump 2 is activated (Process II, Figure 1 A). During this washing step, other cells outside the CTM are cleared from the channel. Once washing is completed, the syringe pump 2 is stopped, and only the valve V1 and valve V4 are opened while the syringe pump 3 is activated (Process III, Figure 1 A). The CTMs retained on the filter are collected in the reservoir at the end of this process. The separation processes (Processes l-lll) are repeated until 1 mL of cell mixture is used up. Once the separation is complete, the reservoir (5), where the CTMs are collected, is placed on top of the image sensor (11 ). Since the image sensor (11 ) is larger than the reservoir (5), the entire reservoir (5) could be imaged at once in real time. Using the developed deep learning algorithms, the positions of CTMs in the reservoir (5) are identified (Process IV). The localized CTMs can be manually collected with a micromanipulator without any mechanical impact on their integrity (Process V). For this process, CTMs are collected under negative pressure via a capillary mounted on the manipulator’s holder chamber. The CTMs, which are kept constant in the capillary by the reduced negative pressure, are collected into the microcentrifuge tube by the applied positive pressure. The process is repeated until all CTMs were collected in the tube. The total analysis time is approximately 1 hour per 1 mL of sample.

[0050] To model CTMs during the separation processes, the green fluorescent protein-labeled A549 lung cancer cell line was used to generate microemboli via the hanging droplet method. To model leukocytes, unstained U937 undifferentiated monocyte cell line was employed. Simultaneous monitoring of the separation processes was conducted using an integrated lensless holographic microscope. The microscope elements consists of (i) a light source (8) (400-700 nm, optimal 520 nm LED), (ii) a pinhole (9) (50-200 pm, optimal 100 pm), and an image sensor (11 ) (CMOS or CCD) (Figure 3). This imaging setup was integrated into the fabricated platform with spatial parameters of these components. The final CMOS-to-sample distance (z3) was 1 mm, while the sample-to-pinhole distance (z2) was 50 mm, and the LED-to-pinhole distance (z1 ) was 20 mm. Geometry aligned with microfluidic chip layout and imaging elements were designed using 3D CAD software (z3:1 - 3 mm, sample-to-pinhole distance (z2) can be 10-100 mm and LED-to-pinhole distance (z1 ) can be 5-30 mm). The platform was 3D-printed using 0.4 mm generic polylactic acid (generic PLA). Following assembly of the 3D-printed components, the built-in lens of the complementary metal-oxide semiconductor (CMOS) image sensor (4) was removed and placed under the channel.

[0051] Solenoid valves were used to automatically supply air pressure to the control channels of the microfluidic valves through tubes to fulfill the microfluidic valve function. 5 solenoid valves were connected to control 5 valves (2) located on the microfluidic chip (10). Each solenoid valve was controlled by the driver module via the microcontroller board (Figure 4A). Using a code written in Python programming language, the microcontroller card was controlled via a serial port connection via a computer and the valves were opened / closed. In order to automatically direct the CTMs captured in the filter structures to the on-chip reservoir (5), the flows in the microchannels were controlled by solenoid valves and syringe pumps that open and close microvalves that coordinately regulate the air flow (Figure 4B). These system components were programmed to operate hierarchically based on the presence / absence of CTMs at the filter, as defined in the developed protocol. On the microfluidic chip (10) there are 5 valves (2), 3 inlets to each of which a syringe pump (1 ) is connected and 2 outlets, one of which is a reservoir (5) (Figure 1 A).

[0052] 5 microvalves in the system (V1 , V2, V3, V4, V5) in the system are defined as opened / closed based on the binary Boolean data principle using digital pins via a microcontroller board. The control of the valves is also achieved by using this binary logical order with the microprocessor, allowing the microvalves to close the inputs and outputs with the movements of the membrane valve structures inside the chip using liquid pressure. The syringe pumps (1 ) in the system can be controlled both manually and via serial ports using Python programming language. Through the serial port assigned to the three syringe pumps (S1 , S2, S3) in the system, the on / off positions, forward / backward movements, movement speeds, syringe type and characteristics, operating time and target volume of the syringe pumps are controlled using the input codes defined for the low-fluid syringe pump. In addition, the input values of the microprocessor that controls the microvalves are provided through the microvalve control inputs defined by the programming language via the serial port assigned to the microprocessor via the programming language. As a result, the inputs and controls of all microvalves, microprocessors and syringe pumps in the designed system are provided through the developed Python algorithm. The analytical steps of the algorithm, including cell inlet into the microfluidic chip (V3, S1 ), forward flushing in the upstream direction of the cell (V5, S2) and backward flushing in the opposite direction of the cell inlet (V1 , S3) are described in detail below. To facilitate these steps, the reservoir (V4) and waste outlets (V2) are also assigned to the relevant components. At the initial state of the system, all microvalves (V1 , V2, V3, V4 and V5) are closed, and all syringe pumps (S1 , S2 and S3) are idle. As the sample containing the cells is introduced into the microfluidic chip, the valve 2 and valve 3 are opened, and the syringe pump 1 begins injection. The system maintains this state until at least one CTM is detected stuck in the filter. Upon CTM detection, the syringe pump 1 stops injection and the valve 3 is closed. Once the flow inside the channel ceases (5 seconds), the valve 5 is opened and the syringe pump 2 initiates injection. The object of this process is to increase the purity of the separation by directing the non-CTM cells accumulated in the channel to the outlet channel. After washing, the syringe pump 2 is stopped and both the valve 5 and valve 2 are closed. For the reverse wash step, the valve 1 and valve 4 are opened, and the syringe pump 3 initiates injection. In this way, the reverse flow directs the CTMs trapped in the filter structures toward the reservoir (5). Upon completion of the wash, the syringe pump 3 stops the injection and the valve 1 and valve 4 are closed, returning the system to its initial state. This cycle is repeated until the entire sample is processed. To detect CTMs trapped in the filter structures on the microfluidic chip (10) used in the invention, an image processing algorithm was developed using a programming language. The detection of microemboli is mainly based on the detection of this immobility by an algorithm, in which CTMs, which are mobile in the area from the cell inlet channel to the filter structures, get stuck in the filters depending on their size when they arrive on the filter and finally remain immobile. The algorithm developed for the invention uses image processing tools developed through numpy, glob, serial, time and OpenCV libraries. First of all, the elliptical morphology of CTMs was determined by utilizing their morphology to define them as a solid body and this morphology is defined as a solid body. For the determination of elliptical structures with a primary radius greater than 4 pixels and a secondary radius greater than 4 pixels, the algorithm for determining the element structures of the OpenCV library was used. After the identification of elliptical CTM structures of certain sizes, the boundaries of these structures are determined by morphology classification algorithms developed by us using OpenCV and numeric Python libraries. These elliptical boundaries are filled in so that the pixel has a threshold between 175-255 after the pixel is defined in gray scale, which is defined as a solid object. Once objects are identified, the backgrounds of the image frames are removed and the image is redefined as object and space. The center points of the objects perceived as larger than 4 x 4 elliptical structure are determined by determining the closed object sizes exceeding the threshold value of 175- 255 in vertical and horizontal pixels, and by randomly selecting the pixel closest to the center for odd numbers, and the pixel closest to the center for even numbers. Bounding boxes are defined by drawing a square frame between both axes as the range of all pixels of the identified object that exceed the threshold value, and these bounding boxes are visualized as red boxes on the image. Depending on their displacement, the identified objects can be re-detected as the same object if their distance from the center is less than 5 pixels in each image frame. This allows the movement of objects to be detected. An object with a difference of more than 5 pixels between the center points in 4 consecutive frames in the video stream is defined as moving. If this difference is less than 5 pixels in 4 consecutive frames, the object is defined as stationary. If this definition of immobility is within the position of the filters on the horizontal and vertical axis, the relevant microvalves and syringe pumps are activated / deactivated via serial ports with the Python algorithm developed to provide forward flushing and backward flushing states. If the detected object remains stuck on the filter even after washing, the process is repeated. If there are no other objects trapped in the filter, the cell input continues until the next object on the filter is detected as motionless.

[0053] Deep learning-based algorithms have been developed for the detection of CTMs in the invention. For this, classification algorithms are used to separate the cells from each other. Classification is one of the important parameters for the detection model. A dataset of 100 images of CTM and U937 cell lines in PDMS reservoir structures on a holographic microscope was used for classification. The training and validation dataset is randomly divided into 80 and 20 images respectively. For this purpose, the data in the images are marked with rectangular bounding boxes as CTM and U937 cell classes using marking tools. These markings allow the machine to identify the CTM or U937 cells in the respective rectangular bounding boxes during data training. After the marking of all images was completed, the classes of the relevant cells were exported from the marking tool in YOLOv5 format.

[0054] In the preferred embodiment of the invention, the YOLOv5m architecture is used for the dataset consisting of training and test images of CTM and U937 cells. The training process was carried out on the open source Google Colaboratory (Colab) by using an external CPU of the Colab system and running it with the Python programming language. The whole process was operated with an external host. When starting the training of the model, necessary parameters such as image size, batch size, epoch, YOLO model were defined. These parameters were set to 416 for image size, 16 for batch size, 230 for epoch and medium for the YOLO model. YOLOv5m was chosen for its fast and detailed image processing capability, on-the-fly processing of images and strong prediction capabilities. The model is defined to be trained 500 times to find the epoch of maximum fitness. In the training steps, the image is now divided into grids of blocks, with each block containing probability values for the position and class of an object. Boundary boxes, which are combinations and intersections of these blocks, are used to detect objects in the image. The grid box with the highest similarity was obtained by assuming the intersection of all predicted boxes with an intersection on the junction. In this way only the boxes with the highest probability of detection were retained. During the training process, training performance was evaluated and various metrics such as recall, precision, mAP @0.5 (mean precision value) and object loss were obtained with YOLOv5 on the Tensorboard machine learning tool. According to the highest calculated mAP@0.5 value, the epoch with maximum compliance was measured as 397. The training resulted in a training weights file corresponding to epoch 397 with maximum fit. The training weights were used to detect objects through YOLO's detection algorithm.

[0055] Model microemboli were filtered in a 1 mL syringe at a concentration of 200 CTM / mL at flow rates of 0.5-4 mL / h to determine the filtration efficiency of the microfluidic chip (10). When the solution in the syringe was finished, the microemboli stuck in the microfluidic filter and the microemboli collected through the filter into the microcentrifuge tube connected to the outlet were counted by fluorescence microscopy. The efficiency of the microfluidic filter was calculated as the ratio of the number of separated CTMs to the total number of CTMs processed in the microfluidic channel (Figure 5A). For 0.5, 1 , 2 and 4 mL / h flow rates, the efficiency of the microfluidic filter was calculated as 47.5 ± 6.1%, 42.85 ± 1.14%, 25.44 ± 6.4% and 23.6 ± 5.3%, respectively. As shown in Figure 5A, increasing the flow rate led to a reduction in filter sepatation efficiency. This is because the microemboli trapped in the filter are continuously exposed to the flow and after a while they deform and break free from the filter structures. Target-sized microemboli (>25 pm) are recovered faster through the filter at flow rates higher than 2 mL / h, but they tend to lose integrity. The CTMs in the cell solution are processed at a flow rate of 1 mL / h so that they can be rapidly separated with over 40% efficiency while maintaining their integrity. Similarly, the efficiency of leukocyte separation in the filter was tested at different flow rates (0.3-2 mL / h). U937 cells were used to model leukocytes. For a concentration of 105cells / mL, the passage of U937 cells through the microfluidic filter is shown in Figure 5B. Separation efficiencies of <0.9% were achieved even at low rates, indicating that most delivered U937 cells can be removed from the chip without getting stuck in the filter. The average size of the U937 cells was 14.6 ± 5 pm, whereas the size of those trapped in the filter cells was 21.9 ± 0.9 pm. This suggests that the size of the cells trapped in the filter is larger than the filter widths. Furthermore, the separation efficiency was investigated at a concentration of 106cells / mL U937 at a flow rate of 1 mL / h. A low separation efficiency (0.3 ± 0.3%) was again achieved at this cell concentration. This shows that high cell concentration does not change the separation efficiency. For a flow rate of 1 mL / h, the designed filter assembly was found to be suitable for both keeping CTM structures containing different numbers of cells on the filter without degradation for a certain period of time (>10 seconds) and passing leukocytes at a high rate (>99%).

[0056] Industrial Applicability of the Invention

[0057] The invention relates to a lensless holographic microscopy-integrated microfluidic device for the real-time detection and isolation of circulating tumor microemboli (CTMs), which are extremely rare in blood, and the operating method of said device and is industrially applicable.

[0058] The invention is not limited to the above descriptions and a person skilled in the art may readily implement alternative embodiments of the invention. These should be considered within the scope of the protection requested by the claims of the invention.

Claims

CLAIMS1. A semi-automated device based on filtration and image processing for the real-time detection and isolation of circulating tumor microemboli (CTMs), which are rare in blood, comprising:• at least three syringe pumps (1 ) for injecting solutions,• at least five valves (2) for controlling the flow direction of the solutions,• at least one outlet channel (3),• at least two microfilters (4) for the removal of leukocytes,• at least one reservoir (5) on the microfluidic chip (10) in which the CTMs filtered according to their size in microfilters (4) in the microfluidic chip (10) are instantaneously observed under a lensless holographic microscope system and transferred to prevent the CTMs from degrading their structure under flow,• at least one light source (8) for illuminating the sample for imaging,• at least one pinhole (9) for spatial filtering of the light,• at least one microfluidic chip (10) on which filtering and image processing is performed to isolate CTMs from leukocytes according to their size and morphology and at least one image sensor (11 ) for capturing holographic images of the sample.

2. A device according to claim 1 , characterized in that said device is manufactured from generic polylactic acid (generic PLA).

3. A device according to claim 1 , characterized in that said microfluidic chip (10) is manufactured from PDMS material.

4. A device according to claim 1 , characterized in that said microfilter (4) has an oval geometry and a final inter-filter spacing of 10-20 pm.

5. A device according to claim 4, characterized in that said microfilter (4) has an oval geometry and a final inter-filter spacing of 18 pm.

6. A device according to claim 1 , characterized in that said light source (8) is an LED light of 400-700 nm.

7. A device according to claim 6, characterized in that said light source (8) is an LED light of 520 nm.

8. A device according to claim 2, characterized in that said pinhole (9) has a diameter of 50-250 pm.

9. A device according to claim 1 , characterized in that said pinhole (9) has a diameter of 100 pm.

10. A device according to claim 1 , characterized in that said image sensor (11 ) is a CMOS or CCD sensor.

11. A device according to claim 10, characterized in that the CMOS-sample distance (z3) is 1-3 mm, the sample-pinhole distance (z2) is 10-100 mm, and the LED- pinhole distance (z1 ) is 5-30 mm.

12. A device according to claim 11 , characterized in that z3 is 1 mm, z2 is 50 mm, and z1 is 20 mm.

13. A device according to claim 1 , characterized in that said microfluidic chip (10) comprises three layers of PDMS and a glass surface.

14. A device according to claim 13, characterized in that the upper PDMS layer comprises control channels for controlling the valve (5), the middle layer and two bottom layers comprise fluidic channels, the bottom layer comprises passages that connect fluidic channels in different layers, and a microfilter (4) for isolating CTMs.

15. An operating method of the device according to claim 14, comprising the following process steps of: i. spatially filtering the light by a pinhole (8) placed in front of the light source (8) in the lensless holographic microscope, ii. automatically controlling the syringe pump (1 ) and the valve (2) structure via the microcontroller using image processing algorithms, iii. switching all syringe pumps (S1 , S2 and S3) and valve (2) structures (V1 , V2, V3, V4 and V5) to the closed state in the system ready position, iv. injecting (process I) the sample containing the cells into the system with S1 by switching the structures V2 and V3 on the microfluidic chip (10) to the opened state, v. real-time processing the microfluidic filters by image processing algorithms through a lensless holographic microscope, vi. automatically stopping the injection of S1 and switching the structure V3 to the closed state by inserting at least one CTM into the microfilter (4) structures on the microfluidic chip (10), vii. injecting the solution via S2 by automatically switching the structure V5 to the opened state for the purification (process II) of CTMs captured on the microfilter (4) within the microfluidic channel from leukocytes, viii. automatically stopping S2 and switching V5 and V2 to the closed state at the end of process II, ix. injecting the solution with S3 by automatically switching the structures V1 and V4 to the opened state in order to direct the CTMs captured in themicrofilter (4) structures on the microfluidic chip (10) to the reservoir (process HI), x. switching the system to the initial state by automatically stopping S3 at the end of process III and switching the structures V1 and V4 to the closed state, xi. repeating this cycle in the process steps (i-x) until the cell-containing sample is exhausted, xii. manually placing the reservoir (5) region on the microfluidic chip (10) onto the image sensor (1 1 ) after the cell solution sample is exhausted, xiii. identifying CTMs by deep learning algorithms while observing the reservoir (5) in real time with a lensless holographic microscope (process IV).

16. A method according to claim 15, characterized in that said syringe pump 1 (S1 ) comprises the cell mixture, and the syringe pump 2 (S2) and syringe pump 3 (S3) contain the culture medium solutions.

Images