DE202022106114U1 - System based on deep learning to detect Alzheimer's disease - Google Patents

Abstract

A deep learning-based system for detecting Alzheimer's disease, the system comprising:
an acquisition unit configured to acquire magnetic resonance imaging (MRI) images of a patient's brain;
a feature extraction unit connected to the acquisition unit to extract features from the acquired MRI images of the brain;
a classification unit configured to compare and classify the features extracted from the MRI images of the patient's brain with the features of MRI images of a normal person;
a database unit connected to the classification unit for storing Alzheimer's disease data classified by the classification unit, allowing a user to review, download and collect data related to Alzheimer's disease to update; and
a data extension unit configured to increase the size and amount of data stored in the database unit by creating the data points from the existing data, wherein the data points are created from the existing data by making minor changes.

Description

FIELD OF THE INVENTION

The present invention relates generally to the field of a deep learning based system for detecting Alzheimer's disease.

BACKGROUND OF THE INVENTION

Alzheimer's disease (AD) is a neurological, degenerative brain disease that is fatal. Early detection of Alzheimer's disease can provide effective therapy and halt brain degeneration. Countless statistical and deep learning approaches have been tried by researchers to study Alzheimer's disease. The evaluation of magnetic resonance imaging (MRI) is a standard procedure in clinical research to diagnose Alzheimer's disease. Characterizing Alzheimer's disease has become increasingly difficult due to the convergence between MRI data from veterans who have the disease and MRI data from middle-aged adults in normal health. Enhanced human-level deep learning algorithms are being used in a variety of fields, including medical imaging.

Alzheimer's disease is a neurological, degenerative brain disease that is fatal. Early detection of Alzheimer's disease can provide effective therapy and halt brain degeneration. Countless statistical and deep learning approaches have been tried by researchers to study Alzheimer's disease. The evaluation of magnetic resonance imaging (MRI) is a standard procedure in clinical research to diagnose Alzheimer's disease. Characterizing Alzheimer's disease has become increasingly difficult due to the convergence between MRI data from veterans who have the disease and MRI data from middle-aged adults in normal health. Enhanced human-level deep learning algorithms are being used in a variety of fields, including medical imaging.

Alzheimer's disease is the most common form of dementia (AD). In developed countries, the probability of developing Alzheimer's is an alarming 30 percent after age 85 and about 5 percent after age 65. By 2050, it is predicted that 0.64 billion people in the world will develop Alzheimer's. Patients with Alzheimer's disease lose their memory, their senses and their ability to carry out everyday activities because the disease leads to the breakdown of brain cells. The first region of the brain to be affected by Alzheimer's disease is that responsible for language and memory. The result is that people with Alzheimer's experience memory loss, disorientation, and difficulty speaking, reading, and writing. Most of the time they are disoriented and even unable to recognize their family and friends. They find it difficult to do simple things like cleaning their teeth or hair. Alzheimer's patients become restless, aggressive, and more likely to run away from home because of all of these factors. Ultimately, Alzheimer's disease damages the part of the brain that controls breathing and heart function, leading to death.

The stages of Alzheimer's disease are very mild, mild and severe. Until a patient has reached the middle stage of the disease, the diagnosis of Alzheimer's disease (AD) is not definitive. For a complete medical assessment of Alzheimer's disease, the patient must undergo physical and neuro-radiological evaluation, complete the Mini-Mental State Examination (MMSE) and take a detailed medical history. Magnetic resonance imaging of the brain is a relatively new tool that doctors are using to detect Alzheimer's disease. The ventricles of the brain expand while the cerebral cortex and hippocampus shrink. The part of the brain responsible for episodic and spatial memory is called the hippocampus. It also acts as a link between our body and our mind. Deterioration of the hippocampus leads to the loss of cells and damage to synapses across synapses. As a result, regions of the brain related to remembering (short-term memory), thinking, and planning have become impaired.

Researchers have developed a variety of computer-aided diagnostic methods to aid in the accurate diagnosis of diseases. Rule-based expert systems and supervised models were developed between the 1970s and 1990s. Medical image data is used to extract feature vectors for training supervised algorithms. Extracting some properties requires human knowledge, which can take a lot of time, money, and effort. Since the development of deep learning models, we can now directly extract features from photos without the help of human specialists. Scientists are focusing their efforts on developing deep learning models for accurate disease diagnosis. MRI, microscope, CT, ultrasound, X-ray and Mam mography are just some of the medical image analysis applications where deep learning has made great strides. The use of deep learning models for organ and substructure segmentation, disease detection and classification led to remarkable discoveries in the fields of pathology, brain, lung, abdomen, heart, breast, bone and retina.

One solution describes how physical changes in the brain are becoming increasingly important for early diagnosis of Alzheimer's disease and can be used to support clinical assessments. The degenerative changes in the brain associated with Alzheimer's disease have been measured by researchers using imaging techniques. Using imaging data and clinical assessments, machine learning techniques were used to create classifiers for the Alzheimer's diagnosis. These studies found major structural differences between the brains of Alzheimer's patients and healthy individuals, particularly in the entorhinal cortex and hippocampal regions. Changes in brain tissue could be the reason for the different behavior of Alzheimer's patients. In addition, there is a link between the behavior of Alzheimer's patients and changes in the connectivity of brain tissue. Images from various imaging methods such as diffusion tensor imaging (DTI), position emission tomography (PET), single photon emission computed tomography (SPECT) and structural and functional magnetic resonance imaging (sMRI, fMRI) show the changes caused by AD brain cell degeneration. Neuroimaging methods have been used in several studies to identify Alzheimer's disease. For example, sMRI, fMRI, PET, SPECT and DTI have been used to diagnose and prognosticate Alzheimer's disease. To increase the accuracy of the diagnosis, data from other modalities were also combined.

In another solution, a conventional automated MRI-based AD diagnostic system is presented, which mainly consists of two components: the extraction of features/biomarkers from MRI data and a classifier based on these features/biomarkers. There are three main categories of feature extraction techniques: voxel-based methods, region-of-interest (ROI)-based methods, and patch-based methods. Voxel-based methods are independent of a structural hypothesis of the brain. For example, voxel-based morphometry assesses the density of local brain tissue (eg, white matter, gray matter, and cerebrospinal fluid). The voxel intensities are used as a classification feature in voxel-based methods. With voxel-based representations, the interpretation of the data is easy and intuitive, but they suffer from the problem of overfitting, as there are few (dozens or hundreds) participants with very high-dimensional (millions) features, posing a major obstacle to Alzheimer's diagnosis the basis of neuroimaging. Dimensionality reduction is required to obtain more compact and functional features. In addition, voxel-based techniques suffer from a lack of geographic knowledge.

In another solution, a region of interest (ROI) based technique is presented that extracts representative features from each of the physically or functionally defined brain areas. This work is based on specific hypotheses about disturbed brain areas. For example, gray matter volume, hippocampal volume, and cortical thickness have been used in several studies. Due to the low feature dimensionality and the coverage of the whole brain, ROI-based methods are often used. However, the derived features in ROI-based techniques are crude as they cannot capture minute or subtle changes associated with brain diseases. The structural or functional changes in the brain that occur as a result of a neurological disorder usually affect many areas of the brain. Because aberrant regions can be contained within a single ROI or span multiple ROIs, voxel-based or ROI-based techniques cannot effectively capture disease-related pathologies. In addition, determining a region of interest (ROI) requires specific human expertise. Patch-based methods subdivide the entire brain image into small patches and extract feature vectors from them. Patch extraction does not require ROI identification; therefore, fewer inputs from human experts are required than with ROI-based methods. Patch-based methods can capture small brain changes with much lower dimensionality than voxel-based approaches. Patch-based methods learn from the whole brain and better capture disease-related abnormalities, leading to more accurate diagnosis. However, selecting meaningful spots from MRI images and generating distinguishing features from these spots remains a challenge.

Another solution focuses on developing advanced machine learning models to diagnose Alzheimer's disease using MRI data. Among the widespread Approaches include Support Vector Machine (SVM), logistic regressors (e.g. Lasso and Elastic Net), sparse representation-based classification (SRC), random forest classifier, etc. Kloppel et al. For example, they used linear SVM to identify Alzheimer's patients from T1-weighted MRI scans. Aversen analyzed structural MRI data using dimensionality reduction and variational approaches. To detect Alzheimer's from MRI scans, they used both a binary SVM classifier and a multiclass classifier. To distinguish between Alzheimer's and healthy individuals, Vemuri et al. SVM to create three different classifiers based on MRI, demographic and genetic data. Gray created a multimodal classification model for Alzheimer's diagnosis from MRI and PET data using a random forest classifier. For the AD classification, Er et al. the grayscale co-occurrence matrix (GLCM) approach. Morra et al. compared the performance of hierarchical AdaBoost, SVM with manual features, and SVM with automatic features for AD detection. MRI data is often used to extract predetermined features for the development of these classifiers. However, due to the heterogeneity of classifier and features, training a classifier independently of feature extraction can result in poorer performance.

Another solution discloses a deep learning model that has recently gained notoriety for its ability to extract feature representations from input data. Deep learning networks progressively capture abstract feature representations from input data using a layered, hierarchical structure. Deep learning architectures learn simple, low-level features from data and then generate complicated high-level features in a hierarchy. Visual object identification, human action recognition, natural language processing, object tracking, image recovery, diagnostics, segmentation tasks, audio data classification, and brain-computer interaction are just a few of the areas where deep- Learning methods have achieved revolutionary achievements. Deep learning models, especially convolutional neural networks (CNNs), have shown great achievements in the field of medical imaging in recent years, especially in the areas of segmentation, detection, registration, and classification. Deep learning methods can reveal the latent or hidden representation in neuroimaging data and efficiently capture disease-related abnormalities. As a result, researchers have recently started using deep learning models to diagnose Alzheimer's and other brain diseases.

Another solution is a method based on transfer learning to diagnose Alzheimer's using MRI images. The hypothesis is that adopting a robust and proven natural imagery architecture and applying transfer learning with intelligent selection of training data can not only improve the accuracy of a model, but also reduce dependence on a large training set. The hypothesis was validated with detailed experiments on the ADNI benchmark dataset, which used MRI scans from 50 subjects from each category of AD, MCI, and NC (150 subjects total) to obtain accuracy data. They have also investigated whether layer-by-layer transfer learning affects their application by stepwise freezing groups of layers in the architecture, and present detailed results of layer-by-layer transfer learning and its relation to training data size. Finally, they present comparative results with six other state-of-the-art methods, where the method they propose significantly outperforms the others and has a 4 percent and 7 percent higher accuracy over the prior art for AD vs. MCI, respectively and provides MCI vs. NC classification problems. They also report 3-way classification results that reach the state of the art and prove the robustness of the proposed method.

In another solution, a deep learning method using an 8-layer 3D convolutional network (3D Con-vNet) for automatic feature learning and classification of Alzheimer's disease was tested. The results have shown that the proposed method is simple and effective for the ADNI data set containing 1198 MRI brain scans from 199 AD patients and 141 NC subjects. Using the random data partitioning strategy, the proposed scheme has achieved an accuracy of 98.74% in the test dataset, with 100% for AD detection at a false alarm rate of 2.4%. The results of the empirical study also showed that the choice of hyperparameters, pre-processing, partitioning of the data by subjects, and the size of the data set can have an impact on the final performance of the deep learning classifier. Finally, comparisons with 7 state-of-the-art methods have further supported the robustness of the proposed 3D ConvNet schema and architecture. Further studies will be conducted on large datasets using subject-separated data partitioning tests.

Existing technologies do not offer a solution to the problem of detecting Alzheimer's disease in its early stages. Therefore, there is an urgent need to solve this problem by developing a deep learning-based Alzheimer's detection system.

SUMMARY OF THE INVENTION

The present invention discloses a deep learning based system for detecting Alzheimer's disease. The deep learning based system includes: an acquisition unit configured to acquire magnetic resonance imaging (MRI) images of a patient's brain; a feature extraction unit connected to the acquisition unit to extract features from the acquired MRI images of the brain; a classification unit configured to compare and classify the features extracted from the MRI images of the patient's brain with the features of MRI images of a normal person; a database unit connected to the classification unit to store data of Alzheimer's disease classified by the classification unit, allowing a user to review collected data related to Alzheimer's disease, download and update; and a data augmentation unit configured to augment the size and amount of data stored in the database unit by creating the data points from the existing data, wherein the data points are created from the existing data by making minor changes.

An aim of the present invention is to develop a system for detecting Alzheimer's disease based on deep learning.

Another object of the present invention is to develop a low-cost, deep learning-based system for detecting Alzheimer's disease.

In order to further clarify the advantages and features of the present invention, a more detailed description of the invention will be given by reference to specific embodiments thereof illustrated in the accompanying figures. It is understood that these figures represent only typical embodiments of the invention and therefore should not be considered as limiting its scope. The invention will be described and illustrated with additional specificity and detail with the accompanying figures.

character list

• 1 ein Blockdiagramm eines auf Deep Learning basierenden Systems zur Erkennung der Alzheimer-Krankheit zeigt; und
• 2 eine tabellarische Darstellung zeigt, in der die in einer CNN-Einheit (Convolutional Neutral Network) verwendeten Parameter definiert sind.

These and other features, aspects and advantages of the present invention will be better understood when the following detailed description is read with reference to the accompanying figures, in which like characters represent like parts throughout the figures, wherein:

• 1 Figure 12 shows a block diagram of a deep learning based system for detecting Alzheimer's disease; and
• 2 Figure 12 shows a tabular representation defining the parameters used in a CNN (Convolutional Neutral Network) entity.

Those skilled in the art will understand that the elements in the figures are presented for simplicity and are not necessarily drawn to scale. For example, the flow charts illustrate the method in terms of the most important steps that help to better understand aspects of the present invention. Furthermore, one or more components of the device may be represented in the figures by conventional symbols, and the figures only show the specific details relevant to an understanding of the embodiments of the present invention, omitting the figures with details overload, which will be readily apparent to those skilled in the art familiar with the present specification.

DETAILED DESCRIPTION

For the purposes of promoting an understanding of the invention, reference will now be made to the embodiment illustrated in the figures and specific language will be used to describe the same. It should be understood, however, that no limitation on the scope of the invention is intended, and such alterations and further modifications to the illustrated system and such further applications of the principles of the invention set forth therein are contemplated as would occur to those skilled in the art invention would normally come to mind.

Those skilled in the art will understand that the foregoing general description and the following detailed description are exemplary and explanatory of the invention and are not to be taken as limiting.

When this specification refers to "an aspect,""anotheraspect," or the like, it means that a particular feature, structure, or characteristic described in connection with the embodiment is at least at least one embodiment of the present invention. Therefore, the phrases "in one embodiment,""in another embodiment," and similar phrases throughout this specification may or may not all refer to the same embodiment.

The terms "comprises," "including," or other variations thereof are intended to cover non-exclusive inclusion, such that a method or method that includes a list of steps includes not only those steps, but may also include other steps that are not expressly stated or pertaining to any such process or method. Likewise, any device or subsystem or element or structure or component preceded by "comprises...a" does not, without further limitation, exclude the existence of other devices or other subsystem or other element or other structure or other component or additional device or additional subsystems or additional elements or additional structures or additional components.

Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one skilled in the art to which this invention pertains. The system and examples herein are intended to be illustrative only and not limiting.

Embodiments of the present invention are described in detail below with reference to the attached figures.

1 Figure 12 shows a block diagram of a deep learning based system (100) for detecting Alzheimer's disease. The deep learning based system (100) comprises: an acquisition unit (102) configured to acquire magnetic resonance imaging (MRI) images of a patient's brain; a feature extraction unit (104) connected to the acquisition unit (102) for extracting features from the acquired MRI images of the brain; a classification unit (106) configured to compare and classify the features extracted from the MRI images of the patient's brain with the features of MRI images of a normal person; a database unit (108) connected to the classification unit (106) for storing Alzheimer's disease data classified by the classification unit, allowing a user to review collected data relating to Alzheimer's disease obtain, review, download and update; and a data augmentation unit (110) configured to augment the size and amount of data stored in the database unit (108) by generating the data points from the existing data, wherein the data points are generated from the existing data by making minor changes.

In another embodiment, the MRI images of the brain acquired by the acquisition unit (102) serve as input for the detection of Alzheimer's disease.

In another embodiment, the feature extraction unit (104) is configured to extract features from the MRI images using a CNN unit (112).

In another embodiment, the CNN entity (112) comprises a convolution layer, a pooling layer, a fully connected layer, and a classifier, where the convolution layer is configured to extract features from the captured images.

In another embodiment, the pooling layer is configured to reduce the number of features in the convolution layer.

In another embodiment, the fully linked layer is configured to reduce the extracted features and convert the extracted features into a column feature map that condenses the parameter to an ideal value.

In another embodiment, the classifier of the CNN unit is configured to detect Alzheimer's disease.

In another embodiment, the most typical technique for describing the product design, architecture, modules, interfaces, and information for a schema to meet specific requirements is schema design. The use of schemas should be obvious because the application of schemas is a hypothesis for product development. We store the disease-related information in the cloud. The information boundaries affecting Alzheimer's disease are used to find out if the person has Alzheimer's. The expectation model is then transferred to the cloud so that specialists around the world can access the predictions. this is in .1 shown. MRI images of the brain are taken and feature extraction is performed. CNN and Inception V3 models are applied to the obtained data set. Then the model trained with the data set. The model with the best accuracy is selected for testing. Then the performance evaluation takes place. Then it is predicted whether the person has Alzheimer's disease or is healthy.

In another embodiment, two deep convolutional neural networks with slightly different settings form the network we propose. We carefully considered the architecture we chose and the design of the proposed system. All models follow the same architectural framework consisting of four primary operations including convolution, pooling, rectified linear unit, and stack normalization.

In another embodiment, the Alzheimer's MRI image data set from the OASIS repository was used in this study. The repository provides access to this dataset. The collection includes MRI brain scans taken from healthy people and people with Alzheimer's disease. Only MRI images are used for classification in this report. There is no need to manually separate the data set as it already contains the train set and the test set. The OASIS dataset for this study was created by Dr. Randy Buckner of Harvard University's Howard Hughes Medical Institute (HHMI), the Washington University School of Medicine's Neuroinformatic Research Group (NRG), and the Biomedical Informatics Research Network (BIRN). There are 416 patients aged 18 to 96 years, each with three or four T1-weighted sMRI images. The dataset includes hundreds of people with mild to severe Alzheimer's disease, all over the age of 60.

In another embodiment, a data augmentation unit capable of artificially increasing the amount of data by creating new data points from existing data is used. Minor changes can be made to the data to increase the data set, or machine learning models can be used to create new data points in the latent space of the original data. Below are the steps of a general data augmentation process: The data augmentation pipeline receives the input data; the data growth pipeline is characterized by the order in which the various growths are performed; the image is fed into the pipeline and processed at each stage with a certain probability; after processing the image, a human expert randomly examines the magnified results and provides input; the augmented data can be used for the training process after human verification. To minimize overfitting when training neural networks, the data set is artificially enlarged by class-preserving perturbations of individual data. The reproducible perturbations allow the development of additional patterns without affecting the semantic meaning of the image. Since it is difficult to manually find more tagged photos in the medical field due to the limited availability of expert knowledge, data augmentation is a reliable method to increase the size of the dataset. For our project, we developed a cropping-based enhancement strategy for each image. The dimension of the slice is set to match the dimension of the proposed Deep CNN classifier.

2 Figure 12 shows a tabular representation of the parameters used in the CNN unit (112). The properties from the MRI imaging are extracted using the CNN unit (112), a type of deep learning technology. To achieve the greatest effect on recognition, the CNN unit (112) takes advantage of the human visual system to memorize and learn the edges and features of images. Three distinctive features of the CNN unit (112) that significantly reduce computational complexity and thus simplify the network are local connection, parameter sharing, and invariant representation. The basic components of the CNN are described as a convolutional layer, pooling layer, fully connected layer and classifier. To create feature maps, the convolutional layer was used to extract image features. The pooling layer was used to reduce the number of features in the convolution layer. After optimally minimizing the feature maps, we reduced them and converted them into a columnar feature map that condensed the parameter to an ideal set. Thereafter, the classifiers were used to identify Alzheimer's.

In another embodiment, the differences between 2D and 3D CNNs are constructed. When classifying MRI scans by 2D and 3D CNNs (for 3D CNNs), slices or volumetric sections are classified using 2D and 3D convolution kernels, respectively. In 2D CNNs, the convolution kernels achieve the slice height and width measurements required for classification by acquiring a whole volume of MRI predictions slice by slice. On the other hand, 2D CNNs are limited to using single layers as inputs and are unable to extract context from connected layers to deliver. This means that voxel data from linked slices can be used to predict the results of MRI scans. 3D CNNs solve this problem by classifying volumetric patches from scans with 3D convolution kernels. Although using information between layers can increase performance, it comes with higher computational costs because these CNNs use many parameters.

In another embodiment, the four basic actions are in are listed, which are performed by several layers in each of the different convolutional neural networks. The layers of the model adhere to a specific type of connectivity pattern called dense connectivity. When training with a limited dataset, the dense connections provide a regulatory effect that reduces overfitting of the network. We connect each layer to every other layer and keep these layers very tight (e.g. 12 filters per layer). So we refer to the strata as dense strata and the combined strata as dense blocks. Since all dense layers are connected, the i-th layer receives the feature maps from all previous layers. As a result, the mesh has a large collection of feature maps, to which each layer has added a tiny number. Each layer has access to both the original input and the gradients of the loss function during the training process. This makes the gradient flow of the mesh more powerful and improves the flow of information.

In another embodiment, transfer learning is disclosed, a deep learning technique that reuses a previously trained model. The figure shows the basic concept of transfer learning. The trick is to "transfer" the learned information to a relatively simple task (in this case categorizing MS from HC) with a small amount of data, using a fairly difficult and successfully pre-trained model, like ImageNet, which consists of many source data was trained. “These three properties are essential for a successful transfer: The target model may only need to retrain the last layers of the pre-trained model, since we assume that the last layers handle the complex identification tasks. The first layers of the pre-trained model can be viewed as feature extractors that help extract low-level features such as edges, tints, spots, shading, and textures. The last layers of the pre-trained model can take over the complex identification tasks.

In another embodiment, two types of CNN models were trained for 10 epochs. The performance of these models was then evaluated against the validation group. The differences between the 2D-CNN were presented using specificity, sensitivity, accuracy, receiver operating characteristic curves (ROCs) and area under curves (AUCs) to determine which method provides the most reliable and efficient results. The best working point of the ROC curve is used to report all accuracy, sensitivity and specificity results for the binary classification. Accuracy, positive predictive value (PPV) or precision, sensitivity or recall, and the harmonic mean of precision and sensitivity are the four metrics used for quantitative assessment and comparison (f1 score). We refer to TP, TN, FP and FN as true positive, true negative, false positive and false negative. In the proposed system, the CNN unit is able to provide good accuracy when training on the MRI data set. Therefore, a CNN is still used for testing.

The figures and the preceding description give examples of embodiments. Those skilled in the art will understand that one or more of the elements described may well be combined into a single functional element. Alternatively, certain elements can be broken down into multiple functional elements. Elements from one embodiment may be added to another embodiment. For example, the order of the processes described herein may be changed and is not limited to the manner described herein. Additionally, the actions of a flowchart need not be performed in the order shown; Also, not all actions have to be carried out. Also, the actions that are not dependent on other actions can be performed in parallel with the other actions. The scope of the embodiments is in no way limited by these specific examples. Numerous variations are possible, regardless of whether they are explicitly mentioned in the description or not, e.g. B. Differences in structure, dimensions and use of materials. The scope of the embodiments is at least as broad as indicated in the following claims.

Advantages, other benefits, and solutions to problems have been described above with respect to particular embodiments. However, the advantages, benefits, problem solutions and components that can cause an advantage, benefit or solution to occur or become more pronounced are not to be considered critical, necessary or essential To understand feature or component of any or all claims.

Reference List

100

A deep learning-based system (100) for detecting Alzheimer's disease

102

registration unit

104

Feature extraction unit

106

classification unit

108

database unit

110

Data expansion unit

Claims (7)

A deep learning-based system for detecting Alzheimer's disease, the system comprising: an acquisition unit configured to acquire magnetic resonance imaging (MRI) images of a patient's brain; a feature extraction unit connected to the acquisition unit to extract features from the acquired MRI images of the brain; a classification unit configured to compare and classify the features extracted from the MRI images of the patient's brain with the features of MRI images of a normal person; a database unit connected to the classification unit for storing Alzheimer's disease data classified by the classification unit, allowing a user to review, download and collect data related to Alzheimer's disease to update; and a data extension unit configured to increase the size and amount of data stored in the database unit by creating the data points from the existing data, wherein the data points are created from the existing data by making minor changes. system after claim 1 , where the MRI images of the brain acquired by the acquisition unit serve as input for the detection of Alzheimer's disease. system after claim 1 , wherein the feature extraction unit is configured to extract features from the MRI images using a Convolutional Neutral Network (CNN) unit. system after claim 3 wherein the CNN entity comprises a convolution layer, a pooling layer, a fully connected layer and a classifier, the convolution layer being configured to extract features from the captured images. system after claim 4 , where the pooling layer is configured to reduce the number of features in the convolution layer. system after claim 4 , where the fully linked layer is configured to reduce the extracted features and convert the extracted features into a column feature map that condenses the parameter to an ideal value. system after claim 4 , where the CNN unit's classifier is configured to detect Alzheimer's disease.

Images