Image processing method, image processing device, storage medium and equipment for early detection of Alzheimer's disease

Abstract

The invention provides an image processing method and device for early detection of Alzheimer's disease, a storage medium and equipment. The image processing method comprises the following steps: an image acquisition stage: acquiring a brain 1H magnetic resonance anatomical image of a tester before taking a labeled probe and acquiring a brain 2H magnetic resonance metabolism image set of the tester within a preset time period after taking the labeled probe; an image processing stage: performing magnetic resonance signal quantitative estimation according to the brain 2H magnetic resonance metabolism image set to obtain a metabolic flux characteristic quantitative image set, and registering the metabolic flux characteristic quantitative image set to the brain 1H magnetic resonance anatomical image to obtain a training sample; and a model training stage: training a preset diagnosis model by the training sample, wherein the diagnosis model is a feature pyramid deep learning model based on Context Transform Block. Metabolic flux characteristic changes of AD in the early stage can be captured, dynamic injection and multi-period continuous scanning of the molecular probe are marked, and long-term tracking is achieved.

Description

Technical Field

[0001] This application belongs to the technical field of image feature detection. Specifically, it relates to an image processing method, an image processing device, a storage medium, and a device for early detection of Alzheimer's disease. Background Art

[0002] Early diagnosis of Alzheimer's disease (AD) is a currently extremely challenging clinical problem. Traditional cognitive scales and structural images are characterized by functional cognitive impairment and brain structure atrophy, and can only detect patients with mid- and late-stage AD. Molecular markers such as Aβ and tau proteins can detect early abnormalities, but do not have the ability to predict disease progression. Recent studies have shown that abnormal brain energy substance metabolic pathways and metabolic fluxes are the decisive factors driving the progression of AD disease. However, there is a severe lack of means to reflect the metabolic activities of human brain energy substances in current clinical practice.

[0003] Currently, the methods for detecting AD in the human body are PET and magnetic resonance imaging molecular metabolic imaging, but they can only one-sidedly reflect brain metabolic activities and cannot accurately depict the metabolic pathways and metabolic characteristics of brain substances and energy: (1) PET uses radioactive probes to bind to specific receptors, transporters, or bioenzymes, and reflects the concentration of specific markers in each brain region through the uptake amount of the probes; (2) Magnetic resonance imaging can directly analyze the concentrations of various neuro-metabolites in brain regions through spectroscopy techniques.

[0004] Among them, PET probes do not directly participate in the brain molecular metabolism process and cannot detect the downstream metabolites of the probes. Therefore, they can only indirectly reflect brain molecular metabolic activities; the magnetic resonance imaging static spectroscopy imaging technology can also only describe the metabolic state of the brain at a certain moment, lacking specific reflection of specific metabolic pathways, and the traditional magnetic resonance spectroscopy imaging based on hydrogen protons (1H) has low metabolite quantification accuracy and poor stability, making it extremely difficult to be popularized and used clinically. Summary of the Invention

[0005] The technical problem solved by this application is: how to provide an image processing method that can be used for early detection and long-term dynamic tracking of Alzheimer's disease.

[0006] This application provides an image processing method for early detection of Alzheimer's disease, and the image processing method includes:

[0007] Image acquisition stage: acquiring the brain 1H magnetic resonance anatomical image of the test subject before ingesting the labeled probe and acquiring the brain 2H magnetic resonance metabolic image set of the test subject within a predetermined time period after ingesting the labeled probe;

[0008] Image processing stage: Based on the brain 2H magnetic resonance metabolic image set, perform quantitative estimation of magnetic resonance signals to obtain a quantitative image set of metabolic flow characteristics, and register the quantitative image set of metabolic flow characteristics to the brain 1H magnetic resonance anatomical image to obtain training samples;

[0009] Model training stage: Use the training samples to train a preset diagnostic model, and the diagnostic model is a feature pyramid deep learning model based on the Contextual Transformer Block.

[0010] Optionally, the brain 1H magnetic resonance anatomical images include the brain 1H magnetic resonance anatomical images of Alzheimer's patients and non-Alzheimer's patients; the brain 2H magnetic resonance metabolic image set includes the brain 2H magnetic resonance metabolic image sets of Alzheimer's patients and non-Alzheimer's patients.

[0011] Optionally, the labeling probe is an exogenous 2H labeling probe, and the exogenous 2H labeling probe includes deuterated glucose, deuterated acetate, and deuterated pyruvate.

[0012] Optionally, the metabolites during the quantitative estimation of magnetic resonance signals based on the brain 2H magnetic resonance metabolic image set include water, glutamate, glutamate salt, glutamine, and lactate.

[0013] Optionally, the diagnostic model includes an encoder, a decoder, a skip connection module, and a feature combination module, and the encoder and the decoder have an attention CoT module.

[0014] Optionally, the image acquisition stage further includes:

[0015] Perform spatial normalization processing, data augmentation processing, and pixel value normalization processing on the brain 1H magnetic resonance anatomical image and the brain 2H magnetic resonance metabolic image set respectively in sequence.

[0016] This application also provides an image processing device for early detection of Alzheimer's disease, and the image processing device includes:

[0017] An image acquisition module, which is used to acquire the brain 1H magnetic resonance anatomical image of the test subject before ingesting the labeling probe and acquire the brain 2H magnetic resonance metabolic image set of the test subject within a predetermined period of time after ingesting the labeling probe;

[0018] An image processing module, which is used to perform quantitative estimation of magnetic resonance signals based on the brain 2H magnetic resonance metabolic image set to obtain a quantitative image set of metabolic flow characteristics, and register the quantitative image set of metabolic flow characteristics to the brain 1H magnetic resonance anatomical image to obtain training samples;

[0019] A model training module, which is used to train a preset diagnostic model with training samples, and the diagnostic model is a feature pyramid deep learning model based on Contextual Transformer Block.

[0020] The present application also provides a computer-readable storage medium storing an image processing program for early detection of Alzheimer's disease. When the image processing program for early detection of Alzheimer's disease is executed by a processor, the image processing method for early detection of Alzheimer's disease as described above is implemented.

[0021] The present application also provides a computer device, which includes a computer-readable storage medium, a processor, and an image processing program for early detection of Alzheimer's disease stored in the computer-readable storage medium. When the image processing program for early detection of Alzheimer's disease is executed by the processor, the image processing method for early detection of Alzheimer's disease as described above is implemented.

[0022] The image processing method, image processing device, storage medium, and device for early detection of Alzheimer's disease provided by the present application at least have the following technical effects:

[0023] By combining the use of deuterium-labeled molecular probes with high-resolution magnetic resonance 2H imaging technology, the changes in metabolic flow characteristics of AD at the early stage can be captured, thereby providing a more accurate early diagnosis, which makes up for the defect that the prior art cannot identify the disease in the early stage. At the same time, through the dynamic injection of deuterium-labeled molecular probes and multi-period continuous scanning, long-term tracking and time-series data analysis are realized. Description of the Drawings

[0024] Figure 1 It is a main step flowchart of an image processing method for early detection of Alzheimer's disease according to one or more embodiments;

[0025] Figure 2 It is a specific implementation flowchart of an image processing method for early detection of Alzheimer's disease according to one or more embodiments;

[0026] Figure 3 It is a schematic diagram of the biological metabolic basis for early detection according to one or more embodiments;

[0027] Figure 4 It is a sampling sequence diagram according to one or more embodiments;

[0028] Figure 5 It is a schematic diagram of the principle of a magnetic resonance signal quantitative estimation method according to one or more embodiments;

[0029] Figure 6 It is a framework diagram of a diagnostic model according to one or more embodiments;

[0030] Figure 7 It is a schematic block diagram of an image processing device for early detection of Alzheimer's disease according to one or more embodiments;

[0031] Figure 8 It is a schematic diagram of a computer device according to one or more embodiments. Detailed implementation manners

[0032] In order to make the objectives, technical solutions and advantages of the present application clearer and more understandable, the present application will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0033] Before describing the various embodiments of the present application in detail, first briefly describe the technical concept of the present application: In the current detection methods for Alzheimer's disease (AD), it is usually difficult to perform early detection and long-term dynamic tracking. For this reason, the image processing method for early detection of Alzheimer's disease provided by the present application first obtains the brain 1H magnetic resonance anatomical image and the brain 2H magnetic resonance metabolic image of the test subject before and after ingesting the labeled probe, and then processes these images to obtain training samples. The diagnostic model is trained using the training samples. The diagnostic model can perform dynamic prediction based on the magnetic resonance image of the person to be predicted. By combining the deuterium-labeled molecular probe with the high-resolution magnetic resonance 2H imaging technology, the metabolic flow feature changes in the early stage of AD can be captured, thereby providing a more accurate early diagnosis. This makes up for the defect that the prior art cannot identify diseases in the early stage. At the same time, through the dynamic injection of the deuterium-labeled molecular probe and multi-time continuous scanning, long-term tracking and time-series data analysis are realized. The following describes the specific principles of the image processing method and the image processing device for early detection of Alzheimer's disease of the present application with more embodiments.

[0034] Specifically, as Figure 1 and Figure 2 shown, the image processing method for early detection of Alzheimer's disease in the first embodiment includes the following steps:

[0035] Step S10, image acquisition stage: Obtain the brain 1H magnetic resonance anatomical image of the test subject before ingesting the labeled probe and obtain the set of brain 2H magnetic resonance metabolic images of the test subject within a predetermined time period after ingesting the labeled probe;

[0036] Step S20, image processing stage: Perform magnetic resonance signal quantitative estimation according to the set of brain 2H magnetic resonance metabolic images to obtain a set of metabolic flow feature quantitative images, and register the set of metabolic flow feature quantitative images to the brain 1H magnetic resonance anatomical image to obtain training samples;

[0037] Step S30, model training stage: Use training samples to train a preset diagnostic model, which is a feature pyramid deep learning model based on the Contextual Transformer Block.

[0038] In one or more embodiments, before Alzheimer's patients and non-Alzheimer's patients take the labeled probe, conventional imaging is performed using a magnetic resonance imaging device. For example, a magnetic resonance imaging device with the model of United Imaging uMR 5.0T human system and equipped with a 1H / 2H dual-core coil is used to collect the 1H magnetic resonance anatomical images of the brains of Alzheimer's patients and non-Alzheimer's patients using conventional sequences such as traditional T1-weighted and T2-weighted. Then, Alzheimer's patients and non-Alzheimer's patients take the labeled probe solution. Exemplarily, the labeled probe is an exogenous 2H-labeled probe, and the exogenous 2H-labeled probe includes deuterated glucose, deuterated acetate, and deuterated pyruvate. For example, the dosage of the labeled probe solution is 200 ml to 300 ml of deuterated glucose, dosage: 0.75 g / kg, maximum amount 60 g. Immediately after taking, 2H-MRSI imaging is continuously collected, and continuous scanning is performed for 0.5 - 1 hour to obtain a set of brain 2H magnetic resonance metabolic images within a continuous time period. Among them, deuterated acetate: Acetate is an important substrate for energy metabolism, and monitoring the metabolic flux can provide useful information for brain metabolic activities, and it is suitable for supplementing the detection of deuterated glucose, especially having significant advantages in understanding the changes in fatty acid metabolism. Deuterated pyruvate: Pyruvate, as a key metabolite in the glycolysis pathway, is sensitive in the early stages of neurodegenerative diseases such as Alzheimer's disease. Figure 3 It is the basic principle diagram for early detection of biological metabolism, Figure 4 It is the sampling sequence diagram.

[0039] In other embodiments, when acquiring images, other imaging methods can also be used. For example: (1) Multimodal imaging: Combine PET imaging and magnetic resonance 2H imaging, and through joint analysis of multi-period data, further improve the accuracy and predictive ability of imaging. PET can provide labeled images at the molecular level, while magnetic resonance imaging provides structural and functional information. The combined application helps to achieve a more comprehensive disease assessment. (2) Photoacoustic Imaging: Use photoacoustic imaging technology to enhance the dynamic monitoring of physiological indicators such as blood oxygen metabolism and blood flow. Combining with deuterium metabolism imaging technology helps to improve the spatio-temporal resolution of imaging.

[0040] In other embodiments, the current 2H-MRSI imaging time is 0.5 - 1 hour. However, for clinical applications, the scanning time needs to be further shortened, especially in the case of dynamic monitoring. For this purpose, the following alternative solutions can be considered: Optimized sampling sequence: By improving the sampling strategy of magnetic resonance 2H imaging, such as introducing Compressed Sensing technology, the scanning time can be reduced, and the scanning speed and image quality can be improved. Multi-channel acceleration technology: Utilize multi-channel coils to accelerate the imaging process and combine it with dynamic time series technology to further shorten the scanning time and increase the availability of multi-period imaging data.

[0041] In one or more embodiments, the image acquisition stage further includes: sequentially performing spatial normalization processing, data enhancement processing, and pixel value normalization processing on the brain 1H magnetic resonance anatomical image and the brain 2H magnetic resonance metabolite image set respectively to improve the stability and generalization ability of the model.

[0042] Exemplarily, the method for quantitatively estimating magnetic resonance signals in this embodiment is as follows:

[0043] As Figure 5 shown, the method for quantitatively estimating magnetic resonance signals needs to ensure the accuracy of metabolite detection and metabolic flux quantification. Through exogenous 2 H-labeled probes, the molecular metabolic activity characteristics of each brain region are reflected by dynamically detecting the downstream metabolites of the probes. However, within the human safety range, the 2 H probe concentration is extremely low, resulting in 2 the detection of H-labeled metabolites being easily interfered by background noise, which is extremely challenging. To address this challenge, starting from physical models such as the 2 H magnetic resonance spectroscopy of the probe and brain metabolic kinetics, an artificial intelligence-based metabolic flux parameter estimation framework will be constructed to achieve accurate estimation of brain metabolic flux characteristics.

[0044] 2 After the 2 H-labeled probe is ingested into the body, the probe and its downstream metabolites can be detected on the magnetic resonance deuterium (

[0045]

[0046] H) spectrum, and its signal model can be expressed as: m where S is the overall spectrum signal intensity obtained by reconstruction, and A m (x, T) is the content of metabolite m at spatial position x and time T. M is the number of metabolites, and L m is the distribution characteristic of metabolites in the frequency domain. Directly solving the above equation will result in large fitting errors due to low signal-to-noise ratio and inaccurate estimation of the content of each metabolite. In the above model, since the probe and its metabolites are known, the distribution characteristic L of each metabolite in the frequency domainm The acquired spectral data can be used for early learning:

[0047]

[0048] where a n，m is the weighting coefficient, the number of spectral sampling points is 2Ns, the echo time is TE, and the sampling interval is Δf. The chemical shift of the metabolite is f m , and the main magnetic field offset in actual sampling is Δf m . The chemical shift f m of each metabolite can be obtained according to prior knowledge. According to the actually acquired spectral data and the above physical model, the distribution characteristics L m of each metabolite in the frequency domain are learned in advance.

[0049] In addition to the frequency domain characteristics of metabolites, glucose and its downstream metabolites also satisfy pharmacokinetic constraints, and this prior information will also be used for the construction of the network model. The downstream products include lactate, glutamate, glutamine, and water. The concentration changes of glucose and downstream products are affected by factors such as metabolic rate and clearance rate:

[0050]

[0051] where [Glc], [Lac], and [Glx] are the concentrations (mM) of glucose, lactate, and glutamate + glutamine, respectively; k g is the transfer rate of glucose between blood and brain tissue (min -1 ); is the concentration of glucose in plasma (mM); f is the enrichment degree of deuterium; v is the extracellular volume fraction outside the blood vessel; V max = V lac + V glx , that is, the consumption rate of glucose (V max ) is the sum of the synthesis rates of lactate and glutamate + glutamine (V lac and V glx ); k m is the constant of glucose absorption; k lac and k glx are the clearance rates of lactate and glutamate + glutamine, respectively.

[0052] In one or more embodiments, a set of quantitative images of metabolic flux characteristics is registered to a brain 1H magnetic resonance anatomical image. Among them, the brain 1H magnetic resonance anatomical image serves as a localization image. After registration, the metabolite distribution in each brain region can be determined. Since the image set is acquired within a continuous time period, each image obtained after registration reflects the metabolite distribution in each brain region at different stages. The diagnostic model based on deep learning can be further used to learn the metabolite distribution rules of each image, so as to use this diagnostic model for early detection of Alzheimer's disease and the stage of the disease.

[0053] Exemplarily, the diagnostic model includes an encoder, a decoder, a skip connection module, and a feature combination module. The encoder and the decoder have an attention CoT module. In the encoder part, a series of feature maps with different resolutions are extracted, followed by a normalization and activation function and a max pooling layer; in the decoder part, each step includes upsampling the feature map, then halving the number of feature channels, concatenating with the corresponding feature map in the encoder, and finally merging the resulting feature map to generate a classification probability value.

[0054] Among them, as Figure 6 shown, a new attention CoT module is introduced in the encoder and decoder parts. This module makes full use of context information to guide the learning of the dynamic attention matrix, thereby enhancing the ability to represent image features. The attention CoT module integrates context information mining and self-attention learning into a unified architecture. It first uses k*k*k group convolutions on all adjacent key-values to contextually associate each key-value spatially. The learned key-values with context information are denoted as K1. Two consecutive 1*1*1 convolutions are also used in this module to obtain the relational matrix A. The two convolutions are denoted by W θ and W δ respectively. Before this, K1 needs to be concatenated with Q first to obtain the relational matrix A = [K1, Q]W θ W δ . The process formed is no longer an isolated Q-K pair, but a relational mapping that can obtain rich context feature information. Next, the relational matrix A is multiplied by V to obtain K2. K2 is the input dynamic context representation, which can capture the dynamic feature interaction between the inputs. Finally, K1 and K2 are fused through the attention mechanism to obtain the output result.

[0055] Among them, during training, training samples are input, feature extraction is performed through the CoT Block, then classification is carried out through the fully connected layer, and the binary cross-entropy loss function is used to calculate the error. The optimization method uses the Adam optimizer and combines the learning rate decay and gradient clipping strategies to prevent overfitting and gradient explosion. During the training process, the change of loss is continuously monitored until the model converges and the optimal weights are saved.

[0056] After the training is completed, the brain 1H magnetic resonance anatomical image of the person to be predicted before ingesting the labeled probe and the brain 2H magnetic resonance metabolic image set of the tester within a predetermined period after ingesting the labeled probe are obtained. After the above processing of the brain 1H magnetic resonance anatomical image and the brain 2H magnetic resonance metabolic image set, they are input into the diagnostic model to obtain the prediction result.

[0057] As Figure 7 shown, in one or more embodiments, the image processing device for early detection of Alzheimer's disease includes an image acquisition module 100, an image processing module 200, and a model training module 300. The image acquisition module 100 is used to obtain the brain 1H magnetic resonance anatomical image of the tester before ingesting the labeled probe and the brain 2H magnetic resonance metabolic image set of the tester within a predetermined period after ingesting the labeled probe. The image processing module 200 is used to perform quantitative estimation of magnetic resonance signals based on the brain 2H magnetic resonance metabolic image set, obtain a set of quantitative images of metabolic flow characteristics, and register the set of quantitative images of metabolic flow characteristics to the brain 1H magnetic resonance anatomical image to obtain training samples. The model training module 300 is used to train a preset diagnostic model using the training samples. The diagnostic model is a deep learning model of a feature pyramid based on the Contextual Transformer Block. The more detailed working processes of each model can refer to the related descriptions of the embodiments of the image processing method for early detection of Alzheimer's disease, which will not be elaborated here.

[0058] The image processing method and image processing device for early detection of Alzheimer's disease provided in this embodiment have the following advantages compared with the prior art:

[0059] 1. Higher early diagnosis ability

[0060] Although the prior art (PET and traditional MRI) can be used to detect the metabolic activities of Alzheimer's disease (AD), they usually can only detect mid- to late-stage metabolic abnormalities and cannot accurately capture the early stage of the disease. In this embodiment, by combining the deuterium-labeled molecular probe with the high-resolution magnetic resonance 2H imaging technology, the change of metabolic flow characteristics in the early stage of AD can be captured, thus providing a more accurate early diagnosis, which makes up for the defect that the prior art cannot identify the disease in the early stage.

[0061] 2. Dynamic monitoring and prediction of disease progression

[0062] Existing PET scans and traditional MRI cannot perform long-term dynamic monitoring and cannot effectively predict the progression of the disease. In contrast, this embodiment achieves long-term tracking and time series data analysis through dynamic injection of deuterium-labeled molecular probes and continuous scanning over multiple time periods. This not only helps track early changes in the disease, but also effectively predicts the progression stage of AD and provides a more comprehensive disease assessment.

[0063] 3. No radioactive probes required

[0064] Existing PET scans use radioactive probes, which can show brain metabolic activity, but have potential hazards of radioactive substances and may limit their frequency of clinical application. The deuterium-labeled molecular probes used in this embodiment are completely non-radioactive, can be completely metabolized, and have the safety of long-term monitoring, which allows the technology to be frequently checked without causing radioactive hazards.

[0065] 4. Improved imaging resolution and accuracy

[0066] Existing deuterium imaging methods may not be able to accurately depict the details of brain metabolic pathways and metabolic flows due to imaging resolution limitations when reflecting brain metabolic activities. In contrast, this embodiment combines high-resolution magnetic resonance 2H imaging technology to accurately detect metabolic flow characteristics in various regions of the brain, especially at high resolution, it can capture subtle metabolic differences in different brain regions, providing more powerful support for early diagnosis and disease prediction.

[0067] 5. Quantitative analysis of multiple metabolites

[0068] Traditional magnetic resonance spectroscopy imaging technology lacks accurate reflection of metabolic pathways. This embodiment provides richer and more multidimensional data support for early diagnosis of AD by quantitatively analyzing multiple metabolites (such as water, glutamate, glutamine, lactate, etc.) and combining deep learning to process and compare the metabolic characteristics of different brain regions.

[0069] In one or more embodiments, a computer-readable storage medium stores an image processing program for early detection of Alzheimer's disease, and when the image processing program for early detection of Alzheimer's disease is executed by a processor, the image processing method for early detection of Alzheimer's disease of the above embodiment is implemented.

[0070] In one or more embodiments, a computer device includes a computer-readable storage medium, a processor, and an image processing program for early detection of Alzheimer's disease stored in the computer-readable storage medium. When the image processing program for early detection of Alzheimer's disease is executed by the processor, the above-described image processing method for early detection of Alzheimer's disease is implemented. At the hardware level, as Figure 8 shown, the computer device includes a processor 12, an internal bus 13, a network interface 14, and a computer-readable storage medium 11. The processor 12 reads the corresponding computer program from the computer-readable storage medium and then runs it, forming a request processing device at the logical level. Of course, in addition to the software implementation, one or more embodiments of this specification do not exclude other implementation manners, such as logic devices or a combination of software and hardware, etc. That is to say, the execution subject of the following processing flow is not limited to each logical unit, and can also be hardware or a logic device.

[0071] Exemplarily, a computer-readable storage medium includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data. Examples of computer-readable storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory, or other memory technologies, compact disc read-only memory (CD-ROM), digital versatile disc (DVD), or other optical storage, magnetic cassette tapes, disk storage, quantum memory, graphene-based storage media, or other magnetic storage devices, or any other non-transmission media that can be used to store information accessible by a computing device.

[0072] The specific embodiments of the present application have been described in detail above. Although some embodiments have been shown and described, those skilled in the art should understand that these embodiments can be modified and perfected without departing from the principles and spirit of the present application as defined by the claims and their equivalents, and these modifications and improvements should also be within the protection scope of the present application.

Claims

1. An image processing method for early detection of Alzheimer's disease, characterized in that: The image processing method comprises: Image acquisition stage: acquiring a 1H MRI anatomical image of the brain of the test person before taking the labeled probe, and acquiring a set of 2H MRI metabolic images of the brain of the test person within a predetermined time period after taking the labeled probe; Image processing stage: quantitative estimation of magnetic resonance signals is performed based on the brain 2H magnetic resonance metabolic image set to obtain a metabolic flow feature quantitative image set, which is then registered to the brain 1H magnetic resonance anatomical image to obtain training samples; Model training phase: Use training samples to train the preset diagnostic model, which is a feature pyramid deep learning model based on Contextual Transformer Block.

2. The image processing method for early detection of Alzheimer's disease according to claim 1, characterized in that: The brain 1H magnetic resonance anatomical images include brain 1H magnetic resonance anatomical images of patients with Alzheimer's disease and patients without Alzheimer's disease; the brain 2H magnetic resonance metabolic image set includes brain 2H magnetic resonance metabolic image set of patients with Alzheimer's disease and patients without Alzheimer's disease.

3. The image processing method for early detection of Alzheimer's disease according to claim 1, characterized in that: The labeling probe is an exogenous 2H labeling probe, and the exogenous 2H labeling probe includes deuterated glucose, deuterated acetate, and deuterated pyruvic acid.

4. The image processing method for early detection of Alzheimer's disease according to claim 1, characterized in that: The metabolites when the MRI signal is quantitatively estimated based on the brain 2H MRI metabolic image set include water, glutamate, glutamate, glutamine, and lactate.

5. The image processing method for early detection of Alzheimer's disease according to claim 1, characterized in that: The diagnostic model includes an encoder, a decoder, a skip connection module and a feature combination module, and the encoder and the decoder have an attention CoT module.

6. The image processing method for early detection of Alzheimer's disease according to claim 1, characterized in that: The image acquisition stage also includes: The brain 1H MRI anatomical images and the brain 2H MRI metabolic images were subjected to spatial normalization, data enhancement and pixel value normalization respectively.

7. An image processing device for early detection of Alzheimer's disease, characterized in that: The image processing device comprises: An image acquisition module, the image acquisition module is used to acquire a 1H magnetic resonance anatomical image of the brain of the test person before taking the labeled probe and to acquire a set of 2H magnetic resonance metabolic images of the brain of the test person within a predetermined time period after taking the labeled probe; An image processing module, wherein the image processing module is used to perform quantitative estimation of magnetic resonance signals based on a brain 2H magnetic resonance metabolic image set, obtain a metabolic flow feature quantitative image set, align the metabolic flow feature quantitative image set to a brain 1H magnetic resonance anatomical image, and obtain a training sample; A model training module is used to train a preset diagnostic model using training samples, and the diagnostic model is a feature pyramid deep learning model based on Contextual Transformer Block.

8. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an image processing program for early detection of Alzheimer's disease, and when the image processing program for early detection of Alzheimer's disease is executed by a processor, the image processing method for early detection of Alzheimer's disease according to any one of claims 1 to 6 is implemented.

9. A computer device, characterized in that: The computer device includes a computer-readable storage medium, a processor, and an image processing program for early detection of Alzheimer's disease stored in the computer-readable storage medium. When the image processing program for early detection of Alzheimer's disease is executed by the processor, the image processing method for early detection of Alzheimer's disease according to any one of claims 1 to 6 is implemented.

Images