Description:Complete Specification:

The following specification describes and ascertains the nature of this invention and the manner in which it is to be performed.
Field of the invention
[0001] This invention is related to a control unit for segmenting at least one red blood cell present in a sample in a device.

Background of the invention

[0002] Many medical image segmentation algorithms have been created and are in demand for reliable evaluation of RBC morphology and the diagnosis of different blood diseases. The central pallor area and the target cells must be segmented for diagnosis of different blood diseases. Due to diverse color, different types of RBC, size orientation of cells, segmenting the RBC cell boundary, central pallor and target cells is frequently difficult. Various image processing techniques are employed for segmentation of the RBC cell boundary, central pallor region, and target cell region which include image thresholding, edge detection, region-based segmentation, morphological operations, etc. There are various problems associated with traditional image processing techniques as RBC have different shapes, colors, sizes, and textures, which makes it difficult to segment them accurately. Furthermore, there are issues such as image noise and artifacts, clumped RBCs (basically overlapping cells), low contrast between RBCs and the background due to various reasons, a large amount of training data is required to obtain accurate image processing algorithms which will consume huge amount of time and is manually exhaustive, and different stains are used for smearing in different parts of the world.

[0003] A US20180211380 patent discloses a system for imaging biological samples and analyzing images of the biological samples is provided. The system can automatically analyze images of biological samples to classify cells of interest using machine learning techniques. Some implementations can diagnose diseases associated with specific cell types. Devices, methods, and computer program product for imaging and analyzing biological samples are also provided.

Brief description of the accompanying drawings
[0004] Figure 1 illustrates a control unit in a digital pathology device in accordance with an embodiment of the invention; and
[0005] Figure 2 illustrates a flowchart of a method of segmenting at least one red blood cell present in a sample in a device at least one in accordance with the present invention.

Detailed description of the embodiments
[0006] Figure 1 illustrates a control unit in a digital pathology device in accordance with an embodiment of the invention. The control unit 12 for segmenting at least one red blood cell 14 present in a sample 15 in a device 10 . The control unit 12 captures an image 16 of the sample 15 from an image capturing unit 13 and identifies a region of interest in the image 16. The control unit 12 divides the image 16 into multiple sub-images using an image processing technique and detects at least one of the red-blood cells 14 in each of the sub-image. The control unit 12 then normalizes the image pixel scale to a predefined scale for determining various portions of each of the detected cell 14. The control unit 12 segments at least one of the red blood cells 14 by using an intelligence module 18 and based on at least one extracted feature. The control unit 12 identifies different types of the red blood cells 14 based on the segmentation.

[0007] Further, the construction of the control unit 12 and the components of the control unit 12 is explained as follows. The control unit 12 is chosen from a group of control units comprising a microprocessor, a microcontroller, a digital circuit, an integrated chip, and the like. It is to be understood that the control unit 12 receives the data related to the blood sample 15 in any other form that is known to a person skilled in the art but is not restricted to the above-mentioned signal forms. The input form of the data is chosen from a group of signals comprising an image, a pulse signal, and the like.

[0008] According to one embodiment of the invention, the image 16 is passed through an image-processing unit 17 that uses any one of the image processing techniques known in the state of the art. The blood of a human being is smeared on the sample 15 for the analysis /diagnosis purpose. The sample 15 having the human body fluid/specimen like blood is placed in one of the analysis devices 10 that is known in the state of the art.

[0009] According to one embodiment of the invention, the blood sample 15 is placed in a digital pathology device 10. The digital pathology device 10 is used to focus the content of the blood sample 15 and to analyze the content of the blood sample 15 for diagnosing / analyzing purpose. According to one embodiment, the control unit 12 is made as an integral component of the digital focusing device 10. The device 10 captures the image 16 (which is two-dimensional) of the content of the blood sample 15 and processes it further to determine and represent the red blood cells 14 of the blood sample 15.

[0010] In another embodiment, the control unit 12 is an external source connected to the device 10, wherein the control unit 12 is a cloud repository that is connected to the device through a communication means. The control unit 12 displays the final result of the type and number of red blood cells 18 in the given blood sample 15 or the severity of the identified at least one medical/anomaly condition based on the retrieved at least one intelligence model 18, to the user directly using the communication means. The intelligence module 18 is chosen from a group of modules comprising an artificial intelligence module (AI) , a deep learning module (DL) , a machine learning module (ML) and the like. The intelligence module 18 can be of any other kind that uses multiple neural networks, and which is built and developed using multiple pre-trained data.

[0011] Figure 2 illustrates a flowchart of a method for segmenting at least one red blood cell presents in a sample 15 in a device in accordance with the present invention. In step S1, an image 16 of the sample 15 from an image capturing unit 13 of a control unit 12 is captured and in step S2, a region of interest in the image 16 is identified. In step S3,the image 16 is divided into multiple sub-images using an image processing technique and at least one of the red-blood cells 14 in each of the sub-image is detected. In step S4, the image pixel scale is normalized to a predefined scale for determining various portions of each of the detected cell. In step S5,at least one of the red blood cells 14 is segmented by using an intelligence module 18 and based on at least one extracted feature. In step S6, the different types of the red blood cells 14 are identified based on the segmentation.

[0012] The method of segmenting the at least one red blood cell 14 present in the sample 15 is explained in detail. The red blood cell 14 is a cell type found in blood cell which is made in bone marrow of the human body. It is clinically known by the term Erythrocytes and holds a circular shape with normal diameter of 7 µm. Each of the red blood cell comprises three regions a central pallor region, a target blob /a target cell region, and a cell boundary. The center pallor region represents the amount of hemoglobin bound to the cell. Lower the amount of hemoglobin, higher is the area of center pallor region and vice versa.

[0013] The target cell or target blob is a hemoglobin disc present in the center of the cell, surrounded by a pallor region, and an outside rim of hemoglobin close to the cell membrane that represents the cell as a bull's-eye or shooting star look. The cell boundary is the outer region that holds the circular shape in case of a healthy red blood cell. The image capturing unit 13 present in the device 10 captures the image 16 of the sample 15 of the blood that is smeared and placed in the device 10. The captured image 16 is processed using the image processing unit for removing multiple artefacts and the unwanted portions and identifying a region of interest where the control unit 12 needs to be focused upon.

[0014] The captured image 16 is made to pass through a preprocessing technique called normalization technique. The preprocessing technique that involves normalizing an image’s pixel value to a common scale for removing the impact of variable brightness and contrast levels. According to one embodiment of the invention, the e RGB (Red, Blue, Green) values range between (0, 255) are normalized to a pixel value that is rescaled between (0, 1). The control unit uses four distinct RGB values for these respective regions: (0,0,0) i.e., black, signifies the background; (0,255,0) i.e., green, signifies the cell membrane/cell boundary; (0,0,255) i.e., blue, signifies the central pallor region and (255,0,0) i.e., red, signifies the target cell/target blob.

[0015] The normalized image 16 is then processed via the intelligence module, wherein according to one embodiment of the invention, the intelligence module 18 uses a U-net architecture module. U-Net is deep learning model widely used for semantic segmentation of medical images. It is made up of an expanded path and a contracting path. The contracting route adheres to the normal convolutional network design, which aids in learning the characteristics of the image and creating a feature map. To create the final segmentation output mask, the expanding route incorporates up sampling of the feature map at each stage. The control unit 10 disclosed in the present invention, builds, and develops the intelligence module 18 (for instance an AI module) using the pre-trained data (data of the multiple blood sample images comprising different shapes of cell boundaries, different sizes of central pallor regions and target blobs/target cells). The developed intelligence module 18 uses the pre-loaded U-net architecture in segmenting the real-time blood cells that are detected in the sample which is loaded in the device.

[0016] The U-Net architecture is made up of a contracting path (on the left side) and an expansive path (on the right side). The contracting path is designed in the manner of a convolutional network. It is composed of two 3x3 convolutions (unpadded convolutions) applied repeatedly, each followed by a rectified linear unit (ReLU) and a 2x2 max pooling operation with stride 2 for down sampling. The control unit double the number of feature channels with each down sampling step. Every step in the expansive path begins with an up sampling of the feature map, followed by a 2x2 convolution ("up-convolution") that cuts the number of feature channels in half, a concatenation with the correspondingly cropped feature map from the contracting path, and two 3x3 convolutions, each followed by a ReLU. Because of the loss of border pixels in each convolution, cropping is required.

[0017] A 1x1 convolution is used at the final layer to map each 16-component feature vector to the desired number of classes. The network has a total of 23 convolutional layers. To enable seamless tiling of the output segmentation map, the input tile size must be chosen so that all 2x2 max-pooling operations are applied to a layer with an even x- and y-size. The U-Net takes a normalized image as input and gives the corresponding mask containing the four classes. The intelligence module is trained by the above disclosed U- net architecture methodology , wherein the sub images undergoes a semantic segmentation. The segmentation of the at least one of the red blood cells involves detecting a cell boundary, a pallor area, and a target blob region.

[0018] Each of the sub-image undergoes a post processing methodology. In which a sliding mechanism is deployed. This mechanism works by sliding a fixed-size window or patch over the image in a systematic manner. The patch is fed into the segmentation module 19 at each position, which predicts a segmentation mask for that patch. The process is then repeated for each subsequent patch present in each of the sub-images, with the window sliding in a predetermined direction with a predetermined stride, until the entire image 14 has been segmented. This mechanism predicts the segmentation mask within the boundaries of sub images or masks. After sliding mechanism each sub image is passed on to the predicted module 20. In each of the sub-images ,the control unit 12 detects different red-blood cells based on the center pallor region, the cell boundary, the target blob/target cell region. This detection comprises the usage of multiple extracted features from the main image during the pre-processing state.

[0019] The features comprises the color, texture, a cellular geometry , a morphology, color, texture, a cellular arrangement, a cellular size/area, a pallor size/area, an energy value, difference pallor, rectangular difference pallor, axis percentage difference pallor, feret major difference pallor, feret minor difference pallor, Chroma decider pallor, circular area difference pallor, red center mean, blue center mean, green center mean, target blob, compactness pallor, form factor pallor, area, area pallor, red mean pallor, green mean pallor, blue mean pallor, compactness pallor, rectangularity pallor, elongation pallor, circularity pallor, aspect ratio pallor, Red Centre Mean, First Order Moment, Second Order Moment, Variance, Homogeneity, Nuclei Circularity, Hemoglobinized pallor texture and the like. However, the extracted features is not restricted to the above disclosed features but can be any other features that provides the information of at least one cell of the human body fluid (which is a blood sample according to one embodiment of the present invention).

[0020] Based on any of the above disclosed features, the center pallor area, target blob region and the cell boundaries are efficiently identified which helps in segmenting the distinct types of the red blood cells 14. The output images that are obtained are connected to form the whole image prediction mask/new main image which contains the overlapping regions as well.

[0021] With the above-mentioned methodology, the framework was able to accurately predict the mask/image 16 with an average intersection of union (IoU) of 85% between the ground truth mask /image 16 (Which was manually annotated) and predicted mask/image 16. The usage of U-Net architecture and its ability to handle the problems faced using traditional image processing techniques such as smearing with different stains is made more efficient. The U-Net architecture was successfully able to resolve these issues and produce reliable results. With the efficient analysis of RBC morphology and classification of different types of RBC’s 14 is made more accurate and efficient for diagnosis of the medical conditions of the patients.

[0022] It should be understood that embodiments explained in the description above are only illustrative and do not limit the scope of this invention. Many such embodiments and other modifications and changes in the embodiment explained in the description are envisaged. The scope of the invention is only limited by the scope of the claims.
, Claims:We Claim:

1. A control unit (12) for segmenting at least one red blood cell (14) present in a sample (15) in a device (10), said control unit (12) adapted to :
- capture an image (16) of said sample (15) from an image capturing unit (13) and identify a region of interest in said image (16) ;
characterized in that :
- normalize said image pixel scale to a predefined scale for determining various portions of said cells (14);
- divide said image (16) into multiple sub-images using an image processing technique and detecting at least one said red-blood cell in each of said sub-image;
- segment at least one said red blood cell by using an intelligence module (18) and based on at least one extracted feature;
- identify different types of said red blood cells (14) based on said segmentation.

2. The control unit (12) as claimed in claim 1, wherein intelligence module (18) is trained by a U- net architecture methodology , wherein said sub images undergoes a semantic segmentation.

3. The control unit (12) as claimed in claim 1, wherein segmentation of said at least one said red blood cell involves detecting a cell boundary, a pallor area, and a target blob portion.

4. The control unit (12) as claimed in claim 1, wherein said control unit (12) identifies a target cell when a hemoglobin disc in the center of said cell surrounded by a pallor area is detected.

5. The control unit (12) as claimed in claim 1, wherein each of said sub image is fed into said intelligence module (18) in a predefined position via a sliding mechanism in said control unit (12).

6. The control unit (12) as claimed in claim 1, wherein said control unit (12) adapted to pass each of said sub image through a segmentation module (19) at each position and to predict a segmentation mask for each of said sub-image.

7. The control unit(12) as claimed in claim 1, wherein each of said sub image is made to pass through a predicted module (20) and all said sub images are combined to form a new main image .

8. A method for segmenting at least one red blood cell (14) present in a sample (15) in a device (10), said method comprising :
- capturing an image (16) of said sample (15) from an image capturing unit (13) of a control unit (12) and identifying a region of interest in said image (16) ;
characterized in that :
- normalizing said image pixel scale to a predefined scale for determining various portions of said cells(14);
- dividing said image (16) into multiple sub-images using an image processing technique and detecting at least one said red-blood cell (14) in each of said sub-image;
- segmenting at least one said red blood cell (14) by using an intelligence module (18) and based on at least one extracted feature;
- identifying different types of said red blood cells (14) based on said segmentation.