Neural network based automatic image segmentation system and its operation method

Abstract

The present invention provides an operation method of a neural network based automatic image segmentation system, which includes the following steps. A plurality of two-dimensional brain parenchyma images of the brain are extracted from a plurality of T2-weighted images of the brain through a two-dimensional neural network, and the plurality of two-dimensional brain parenchyma images of the brain constitute a three-dimensional brain parenchyma image of the brain; the three-dimensional brain parenchyma image of the brain is pre-processed to obtain a pre-processed three-dimensional brain parenchyma image; and at least one cerebral cavernous malformation image is segmented from the pre-processed three-dimensional brain parenchyma image through a three-dimensional convolutional neural network.

Description

[Technical Field]

[0001] This invention relates to a computer system and its operating method, and more particularly to an automatic image segmentation system based on neural networks and its operating method. [Previous Technology]

[0002] Cerebral cavernous malformation (CCM) is a disease characterized by abnormal cerebral blood vessel structure. CCM is characterized by recurrent episodes of minor hemorrhages and hemosiderin deposition, which can produce variable imaging features, making it difficult for traditional imaging techniques to define the true boundaries of the cerebral cavernous malformation.

[0003] It is evident that the existing detection methods described above still have inconveniences and shortcomings, and require further improvement. To address these issues, relevant fields have spared no effort in seeking solutions, but a suitable method has yet to be developed. Therefore, how to more accurately and automatically segment images of cavernous malformations in the brain is indeed one of the important research topics at present, and has become a goal that urgently needs improvement in related fields. [Summary of the Invention]

[0004] This invention proposes an automatic image segmentation system based on neural networks and its operation method, which improves the problems of prior art.

[0005] In some embodiments of the present invention, the neural network-based automatic image segmentation system proposed in the present invention includes a storage device and a processor. The storage device stores at least one instruction, and the processor is electrically connected to the storage device. The processor is used to access and execute at least one instruction to: extract multiple two-dimensional brain parenchyma images of the brain from multiple T2-weighted images of the brain through a two-dimensional neural network, wherein the multiple two-dimensional brain parenchyma images constitute a three-dimensional brain parenchyma image of the brain; preprocess the three-dimensional brain parenchyma image of the brain to obtain a preprocessed three-dimensional brain parenchyma image; and segment at least one image of a cavernous malformation of the brain from the preprocessed three-dimensional brain parenchyma image through a three-dimensional convolutional neural network.

[0006] In some embodiments of the present invention, the two-dimensional neural network is a masked region convolutional neural network, which removes signals associated with the skull and scalp to extract multiple two-dimensional brain parenchyma images.

[0007] In some embodiments of the present invention, the preprocessing performed by the processor includes: resampling a three-dimensional brain parenchyma image of the brain, wherein the resampling scales each voxel in the three-dimensional brain parenchyma image of the brain from its initial size to a predetermined size to obtain a resampled three-dimensional brain parenchyma image; and applying Z-score intensity normalization to normalize the grayscale intensity of each data point in the resampled three-dimensional brain parenchyma image to obtain a grayscale intensity normalized three-dimensional brain parenchyma image.

[0008] In some embodiments of the present invention, the preprocessing performed by the processor further includes: performing data increment on the grayscale intensity-normalized three-dimensional brain parenchyma image, the data increment generating an additional image based on the grayscale intensity-normalized three-dimensional brain parenchyma image after random rotation and random scaling, and the grayscale intensity-normalized three-dimensional brain parenchyma image and the additional image serving as the preprocessed three-dimensional brain parenchyma image.

[0009] In some embodiments of the present invention, the processor is configured to access and execute at least one instruction to: extract a first resolution image block from a preprocessed three-dimensional brain parenchyma image through a first channel of a three-dimensional convolutional neural network; extract a second resolution image block from the preprocessed three-dimensional brain parenchyma image through a second channel of a three-dimensional convolutional neural network, wherein the resolution of the second resolution image block is different from the resolution of the first resolution image block; and after the first resolution image block of the first channel and the second resolution image block of the second channel are convolved in the convolutional layer of the three-dimensional convolutional neural network, upsampling is performed through the second channel to facilitate information matching and classification processing in the fully connected layer of the three-dimensional convolutional neural network, thereby outputting a segmentation result of at least one image of a cavernous malformation of the brain in the classification layer of the three-dimensional convolutional neural network.

[0010] In some embodiments of the present invention, the operation method of the image automatic segmentation system based on neural networks proposed in the present invention includes the following steps: (A) extracting multiple two-dimensional brain parenchyma images of the brain from multiple T2 weighted images of the brain through a two-dimensional neural network, wherein the multiple two-dimensional brain parenchyma images constitute a three-dimensional brain parenchyma image of the brain; (B) performing preprocessing on the three-dimensional brain parenchyma image of the brain to obtain a preprocessed three-dimensional brain parenchyma image; and (C) segmenting at least one image of a cavernous malformation of the brain from the preprocessed three-dimensional brain parenchyma image through a three-dimensional convolutional neural network.

[0011] In some embodiments of the present invention, the two-dimensional neural network is a masked region convolutional neural network, and step (A) includes: removing signals related to the skull and scalp through the masked region convolutional neural network to extract multiple two-dimensional brain parenchyma images of the brain.

[0012] In some embodiments of the present invention, step (B) includes: resampling a three-dimensional brain parenchyma image of the brain, wherein the resampling scales each voxel in the three-dimensional brain parenchyma image of the brain from its initial size to a predetermined size to obtain a resampled three-dimensional brain parenchyma image; and applying Z-score intensity normalization to normalize the grayscale intensity of each data point in the resampled three-dimensional brain parenchyma image to obtain a grayscale intensity normalized three-dimensional brain parenchyma image.

[0013] In some embodiments of the present invention, step (B) includes: performing data increment on the grayscale intensity-normalized three-dimensional brain parenchyma image, the data increment generating an additional image based on the grayscale intensity-normalized three-dimensional brain parenchyma image after random rotation and random scaling, and the grayscale intensity-normalized three-dimensional brain parenchyma image and the additional image serving as the preprocessed three-dimensional brain parenchyma image.

[0014] In some embodiments of the present invention, step (C) includes: extracting a first resolution image block from the preprocessed three-dimensional brain parenchyma image through the first channel of the three-dimensional convolutional neural network; extracting a second resolution image block from the preprocessed three-dimensional brain parenchyma image through the second channel of the three-dimensional convolutional neural network, wherein the resolution of the second resolution image block is different from the resolution of the first resolution image block; and after the first resolution image block of the first channel and the second resolution image block of the second channel are convolved in the convolutional layer of the three-dimensional convolutional neural network, upsampling is performed through the second channel to facilitate information matching and classification processing in the fully connected layer of the three-dimensional convolutional neural network, thereby outputting the segmentation result of at least one cerebral cavernous malformation image in the classification layer of the three-dimensional convolutional neural network.

[0015] In summary, the technical solution of the present invention has significant advantages and beneficial effects compared with the prior art. Through the neural network-based automatic image segmentation system and its operation method of the present invention, images of cerebral cavernous malformations can be segmented more accurately and automatically, facilitating the depiction process of cerebral cavernous malformations in Gamma Knife radiosurgery and related clinical studies.

[0016] The above description will be described in detail below with reference to embodiments, and the technical solution of the present invention will be further explained.

Implementation Method

[0017] To make the description of the present invention more detailed and complete, reference can be made to the accompanying drawings and the various embodiments described below, in which the same numbers represent the same or similar elements. On the other hand, well-known elements and steps are not described in the embodiments to avoid unnecessarily limiting the present invention.

[0018] Referring to Figure 1, the technical embodiment of the present invention is a neural network-based automatic image segmentation system 100, which can be applied to computers or widely used in related technical processes. This neural network-based automatic image segmentation system 100 represents a significant technological advancement and has broad industrial applicability. The specific implementation of the neural network-based automatic image segmentation system 100 will be described below with reference to Figure 1.

[0019] It should be understood that various embodiments of the neural network-based automatic image segmentation system 100 are described in conjunction with Figure 1. In the following description, for ease of explanation, numerous specific details are further provided to offer a comprehensive description of one or more embodiments. However, the technology can be implemented without these specific details. In other examples, known structures and apparatuses are shown in block diagram form for the effective description of these embodiments. The term "by way of example" as used herein means "as an example, instance, or illustration." Any embodiment described herein as "by way of example" is not to be construed as preferred or superior to other embodiments.

[0020] In practice, in some embodiments of the present invention, the neural network-based automatic image segmentation system 100 may be a server, a computer host, or other computer device. Regarding servers, many technologies have been developed or are under development to manage the operation of computer servers, generally providing accessibility, consistency, and efficiency. Remote management allows for the removal of input / output interfaces for the server (e.g., display screen, mouse, keyboard, etc.) and the need for network administrator entities to access each server. For example, large data centers containing many computer servers typically use various remote management tools to manage, configure, monitor, and troubleshoot server hardware and software.

[0021] It should be understood that the terms "about," "approximately," or "roughly" used herein are used to modify any quantity that may vary slightly, but such slight variations do not change its nature. Unless otherwise specified in the embodiments, the error range of the values ​​modified by "about," "approximately," or "roughly" is generally permissible within 20 percent, preferably within 10 percent, and more preferably within 5 percent.

[0022] In practice, in some embodiments of the present invention, the neural network-based automatic image segmentation system 100 may selectively establish a connection with the magnetic resonance imaging (MRI) device 190. It should be understood that, in the embodiments and claims, the description of "connection" can broadly refer to a component indirectly communicating with another component via wired and / or wireless means, or a component not physically connected to another component without the need for other components. For example, the neural network-based automatic image segmentation system 100 may indirectly communicate with the MRI device 190 via wired and / or wireless means, or the neural network-based automatic image segmentation system 100 may not physically connect to the MRI device 190 without the need for other components. Those skilled in the art should flexibly choose according to the needs at the time.

[0023] Figure 1 is a block diagram of a neural network-based automatic image segmentation system 100 according to an embodiment of the present invention. As shown in Figure 1, the neural network-based automatic image segmentation system 100 includes a storage device 110, a processor 120, a display 130, and a transmission device 150. For example, the storage device 110 may be a hard drive, flash memory, or other storage medium; the processor 120 may be a central processing unit; the display 130 may be a built-in display or an external screen; and the transmission device 150 may be a connector, a wired / wireless network device, or other transmission interface.

[0024] In terms of architecture, the neural network-based automatic image segmentation system 100 is electrically connected to the magnetic resonance imaging device 190, the storage device 110 is electrically connected to the processor 120, the processor 120 is electrically connected to the display 130, and the transmission device 150 is electrically connected to the processor 120. It should be understood that, in the embodiments and claims, the description involving "electrical connection" can generally refer to one element being indirectly electrically coupled to another element through other elements, or one element being directly electrically connected to another element without needing to go through other elements. For example, the storage device 110 may be a built-in storage device directly electrically connected to the processor 120, or the storage device 110 may be an external storage device indirectly connected to the processor 120 through a network device.

[0025] In practice, for example, the magnetic resonance imaging device 190 can measure T2-weighted images of the brain. Although only one magnetic resonance imaging device 190 is shown in Figure 1, this does not limit the invention. In practice, the magnetic resonance imaging device 190 can refer to one or more magnetic resonance imaging devices, and those skilled in this art should choose flexibly according to their needs.

[0026] In some embodiments of the present invention, the storage device 110 stores an electroencephalogram (EEG) and at least one instruction. The processor 120 accesses and executes the at least one instruction to: extract multiple two-dimensional brain parenchyma images from multiple T2-weighted images of the brain using a two-dimensional neural network, wherein the multiple two-dimensional brain parenchyma images constitute a three-dimensional brain parenchyma image; preprocess the three-dimensional brain parenchyma image to obtain a preprocessed three-dimensional brain parenchyma image; and segment at least one cavernous malformation (CVM) image from the preprocessed three-dimensional brain parenchyma image using a three-dimensional convolutional neural network. The display 130 can display the CVM image. The neural network-based automatic image segmentation system 100 more accurately and automatically segments the CVM image, facilitating the depiction process of CVMs in Gamma Knife radiosurgery and related clinical studies.

[0027] It should be understood that cavernous malformations of the brain are characterized by recurrent variable hemorrhages and hemosiderin deposition. Both of these conditions can make it difficult to define the true margins of the cavernous malformation in control experiments based solely on other images (e.g., electroencephalograms, positron emission tomography, T1 images, general T2 images, etc.). Compared to the aforementioned other images, in some embodiments of the present invention, the processor 120 uses T2-weighted images to facilitate the accurate definition of the true margins of the cavernous malformation by the three-dimensional convolutional neural network.

[0028] Regarding the specific mechanism of the above-mentioned two-dimensional neural network, in some embodiments of the present invention, the two-dimensional neural network is a masked region convolutional neural network. The masked region convolutional neural network removes signals related to the skull and scalp to extract multiple two-dimensional brain parenchyma images of the brain, so as to improve the accuracy of the three-dimensional convolutional neural network in segmenting the images of cavernous vascular malformations of the brain.

[0029] Regarding the specific mechanism of the above preprocessing, in some embodiments of the present invention, the preprocessing performed by the processor 120 includes: resampling the three-dimensional brain parenchyma image of the brain, wherein the resampling scales each voxel in the three-dimensional brain parenchyma image of the brain from its initial size (e.g., 0.508 x 0.508 x 3.01 mm³) to a predetermined size (e.g., 1 x 1 x 1 mm³) to obtain a resampled three-dimensional brain parenchyma image, thereby improving the accuracy of the three-dimensional convolutional neural network in segmenting the image of the cavernous malformation of the brain; and applying Z-score intensity normalization to normalize the grayscale intensity of each data point in the resampled three-dimensional brain parenchyma image to obtain a grayscale intensity normalized three-dimensional brain parenchyma image, thereby improving the accuracy of the three-dimensional convolutional neural network in segmenting the image of the cavernous malformation of the brain.

[0030] Furthermore, regarding the specific mechanism of the above preprocessing, in some embodiments of the present invention, the preprocessing performed by the processor 130 further includes: performing data increment on the grayscale intensity-normalized three-dimensional brain parenchyma image to enhance the segmentation ability of the three-dimensional convolutional neural network for the cavernous malformation image of the brain and prevent overfitting. The data increment generates an additional image based on the grayscale intensity-normalized three-dimensional brain parenchyma image, which is randomly rotated (e.g., randomly rotated -45 to 45 degrees along the X, Y, and Z axes) and randomly scaled (e.g., 0.9 to 1.1 times). The grayscale intensity-normalized three-dimensional brain parenchyma image and the additional image serve as the preprocessed three-dimensional brain parenchyma image.

[0031] Regarding the specific mechanism of the above-mentioned three-dimensional convolutional neural network, in some embodiments of the present invention, the processor 120 is used to access and execute at least one instruction to: extract a first-resolution image patch from the preprocessed three-dimensional brain parenchyma image through a first channel of the three-dimensional convolutional neural network (e.g., the DeepMedic model); and extract a second-resolution image patch from the preprocessed three-dimensional brain parenchyma image through a second channel of the three-dimensional convolutional neural network. The resolution of the second-resolution image patch (e.g., low resolution) is different from the resolution of the first-resolution image patch (e.g., standard resolution). In practice, for example, the first channel extracts a standard-resolution and normal-scale image patch (to capture detailed features), and the second channel extracts a low-resolution and large-scale image patch (to capture spatial location information). Using individual channels to extract image patches of multiple resolutions can enhance the ability of the three-dimensional convolutional neural network to learn and recognize images of different sizes of cavernous vascular malformations in different brain locations.

[0032] Continuing above, the processor 120 is configured to access and execute at least one instruction to: after the first resolution image block of the first channel and the second resolution image block of the second channel are convolved in the convolutional layer of the three-dimensional convolutional neural network, upsampling is performed through the second channel (e.g., interpolation or other methods) to facilitate information matching and classification processing in the fully connected layer of the three-dimensional convolutional neural network, and then outputting the segmentation result of at least one image of a cavernous malformation of the brain in the classification layer of the three-dimensional convolutional neural network.

[0033] To further illustrate the operation method of the neural network-based automatic image segmentation system 100 described above, please refer to Figures 1 and 2. Figure 2 is a flowchart of the operation method 200 of the neural network-based automatic image segmentation system 100 according to an embodiment of the present invention. As shown in Figure 2, the operation method 200 includes steps S201 to S203 (it should be understood that, unless otherwise specified, the order of the steps mentioned in this embodiment can be adjusted according to actual needs, and they can even be executed simultaneously or partially simultaneously).

[0034] The operating method 200 may take the form of a computer program product on a non-transitory computer-readable recording medium having a plurality of computer-readable instructions contained in the medium. Suitable recording media may include any of the following: non-volatile memory, such as read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electronically erasable programmable read-only memory (EEPROM); volatile memory, such as static memory (SRAM), dynamic memory (DRAM), double data rate random access memory (DDR-RAM); optical storage devices, such as read-only optical disc (CD-ROM), read-only bit multi-function audio-visual disc (DVD-ROM); magnetic storage devices, such as hard disk drives, floppy disk drives.

[0035] In some embodiments of the present invention, in step S201, multiple two-dimensional brain parenchyma images of the brain are extracted from multiple T2-weighted images of the brain through a two-dimensional neural network, and the multiple two-dimensional brain parenchyma images of the brain constitute a three-dimensional brain parenchyma image of the brain; in step S202, the three-dimensional brain parenchyma image of the brain is preprocessed to obtain a preprocessed three-dimensional brain parenchyma image; in step S203, at least one image of a cavernous malformation of the brain is segmented from the preprocessed three-dimensional brain parenchyma image through a three-dimensional convolutional neural network.

[0036] In some embodiments of the present invention, the two-dimensional neural network is a masked region convolutional neural network, and step S201 includes: removing signals related to the skull and scalp through the masked region convolutional neural network to extract multiple two-dimensional brain parenchyma images of the brain.

[0037] In some embodiments of the present invention, step S202 includes: resampling a three-dimensional brain parenchyma image of the brain, wherein the resampling scales each voxel in the three-dimensional brain parenchyma image of the brain from its initial size to a predetermined size to obtain a resampled three-dimensional brain parenchyma image; and applying Z-score intensity normalization to normalize the grayscale intensity of each data point in the resampled three-dimensional brain parenchyma image to obtain a grayscale intensity normalized three-dimensional brain parenchyma image.

[0038] In some embodiments of the present invention, step S202 includes: performing data increment on the grayscale intensity-normalized three-dimensional brain parenchyma image, the data increment generating an additional image based on the grayscale intensity-normalized three-dimensional brain parenchyma image after random rotation and random scaling, the grayscale intensity-normalized three-dimensional brain parenchyma image and the additional image serving as the preprocessed three-dimensional brain parenchyma image.

[0039] In some embodiments of the present invention, step S203 includes: extracting a first resolution image block from the preprocessed three-dimensional brain parenchyma image through the first channel of the three-dimensional convolutional neural network; extracting a second resolution image block from the preprocessed three-dimensional brain parenchyma image through the second channel of the three-dimensional convolutional neural network, wherein the resolution of the second resolution image block is different from the resolution of the first resolution image block; and after the first resolution image block of the first channel and the second resolution image block of the second channel are convolved in the convolutional layer of the three-dimensional convolutional neural network, upsampling is performed through the second channel to facilitate information matching and classification processing in the fully connected layer of the three-dimensional convolutional neural network, thereby outputting the segmentation result of at least one cerebral cavernous malformation image in the classification layer of the three-dimensional convolutional neural network.

[0040] In summary, the technical solution of the present invention has significant advantages and beneficial effects compared with the prior art. Through the neural network-based automatic image segmentation system 100 and its operation method 200 of the present invention, images of cerebral cavernous malformations can be segmented more accurately and automatically, facilitating the depiction process of cerebral cavernous malformations in Gamma Knife radiosurgery and related clinical studies.

[0041] Although the present invention has been disclosed above by way of embodiments, it is not intended to limit the present invention. Any person skilled in the art can make various modifications and refinements without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be determined by the appended claims. [Simplified Explanation of the Diagram]

[0042] To make the above and other objects, features, advantages and embodiments of the present invention more apparent and understandable, the accompanying drawings are described as follows: Figure 1 is a block diagram of an automatic image segmentation system based on a neural network according to an embodiment of the present invention; and Figure 2 is a flowchart of an operation method of an automatic image segmentation system based on a neural network according to an embodiment of the present invention.

Claims

1. A neural network-based automatic image segmentation system, comprising: a storage device storing at least one instruction; and a processor electrically connected to the storage device, wherein the processor is configured to access and execute the at least one instruction to: extract multiple two-dimensional brain parenchyma images of a brain from multiple T2-weighted images of a brain using a two-dimensional neural network, the multiple two-dimensional brain parenchyma images constituting a three-dimensional brain parenchyma image of the brain; perform a preprocessing on the three-dimensional brain parenchyma image of the brain to obtain a preprocessed three-dimensional brain parenchyma image; and segment at least one image of a cavernous malformation of the brain from the preprocessed three-dimensional brain parenchyma image using a three-dimensional convolutional neural network, wherein the processor is further configured to access and execute the at least one instruction to: A first resolution image patch is extracted from the preprocessed 3D brain parenchyma image through a first channel of the 3D convolutional neural network, and a second resolution image patch is extracted from the preprocessed 3D brain parenchyma image through a second channel of the 3D convolutional neural network. The resolution of the second resolution image patch is different from that of the first resolution image patch. After the first resolution image patch of the first channel and the second resolution image patch of the second channel are convolved in the convolutional layer of the 3D convolutional neural network, upsampling is performed through the second channel to facilitate information matching and classification in the fully connected layer of the 3D convolutional neural network. The segmentation result of the at least one cavernous malformation image is then output from the classification layer of the 3D convolutional neural network.

2. The neural network-based automatic image segmentation system as described in claim 1, wherein the two-dimensional neural network is a masked region convolutional neural network that removes signals associated with the skull and scalp to extract the plurality of two-dimensional brain parenchyma images of the brain.

3. The neural network-based automatic image segmentation system as described in claim 1, wherein the preprocessing performed by the processor comprises: resampling the three-dimensional brain parenchyma image of the brain, the resampling scaling each voxel in the three-dimensional brain parenchyma image of the brain from its initial size to a predetermined size to obtain a resampled three-dimensional brain parenchyma image; and applying Z-score intensity normalization to normalize the grayscale intensity of each data point in the resampled three-dimensional brain parenchyma image to obtain a grayscale intensity normalized three-dimensional brain parenchyma image.

4. The neural network-based automatic image segmentation system as described in claim 3, wherein the preprocessing performed by the processor further comprises: performing data increment on the grayscale intensity-normalized three-dimensional brain parenchyma image, the data increment generating an additional image based on the grayscale intensity-normalized three-dimensional brain parenchyma image through random rotation and random scaling, the grayscale intensity-normalized three-dimensional brain parenchyma image and the additional image serving as the preprocessed three-dimensional brain parenchyma image.

5. A method for operating an automatic image segmentation system based on neural networks, the method comprising the following steps: (A) extracting multiple two-dimensional brain parenchyma images of a brain from multiple T2-weighted images of a brain using a two-dimensional neural network, wherein the multiple two-dimensional brain parenchyma images constitute a three-dimensional brain parenchyma image of the brain; (B) performing a preprocessing on the three-dimensional brain parenchyma image of the brain to obtain a preprocessed three-dimensional brain parenchyma image; and (C) segmenting at least one image of a cavernous malformation of the brain from the preprocessed three-dimensional brain parenchyma image using a three-dimensional convolutional neural network, wherein step (C) further comprises: extracting a first resolution image block from the preprocessed three-dimensional brain parenchyma image using a first channel of the three-dimensional convolutional neural network, and extracting a second resolution image block from the preprocessed three-dimensional brain parenchyma image using a second channel of the three-dimensional convolutional neural network, wherein the resolution of the second resolution image block is different from the resolution of the first resolution image block; and, After the first resolution image block of the first channel and the second resolution image block of the second channel are convolved in the convolutional layer of the three-dimensional convolutional neural network, upsampling is performed through the second channel to facilitate information matching and classification in the fully connected layer of the three-dimensional convolutional neural network, and then the segmentation result of the at least one cavernous malformation image of the brain is output in the classification layer of the three-dimensional convolutional neural network.

6. The method of operation as described in claim 5, wherein the two-dimensional neural network is a masked region convolutional neural network, and step (A) includes: removing signals related to the skull and scalp through the masked region convolutional neural network to extract the plurality of two-dimensional brain parenchyma images of the brain.

7. The method of operation as described in claim 5, wherein step (B) comprises: resampling the three-dimensional brain parenchyma image of the brain, the resampling scaling each voxel in the three-dimensional brain parenchyma image of the brain from its initial size to a predetermined size to obtain a resampled three-dimensional brain parenchyma image; and applying Z-score intensity normalization to normalize the grayscale intensity of each data point in the resampled three-dimensional brain parenchyma image to obtain a grayscale intensity normalized three-dimensional brain parenchyma image.

8. The method of operation as described in claim 7, wherein step (B) comprises: performing data increment on the grayscale intensity-normalized three-dimensional brain parenchyma image, the data increment generating an additional image based on the grayscale intensity-normalized three-dimensional brain parenchyma image by random rotation and random scaling, the grayscale intensity-normalized three-dimensional brain parenchyma image and the additional image serving as the preprocessed three-dimensional brain parenchyma image.

Images