AU2021101725A4 - The Computer Aided Diagnosis (CAD) System for the Detection of Alzheimer Disease Using MRI Real Images - Google Patents

Abstract

The Computer Aided Diagnosis (CAD) System for the Detection of Alzheimer Disease Using MRI Real Images Many instruments have been used to diagnose Alzheimer's disease, but due to the limitation of Magnetic Resonance Imaging (MRI) scanning equipment, it remains a financially expensive diagnostic system that detects disease with poor sensitivity and efficiency. This invention proposes a novel technique for forecasting AD called the CAD mechanism, which employs numerous algorithms. Due to the thermal activities of the hardware involved in the scanning device, the MRI images from the scanning device are an extremely noisy image. The 2D Adaptive Bilateral Filter (2D-ABF) algorithm is used in the image restoration process. Image enhancement strategies based on the Adaptive Histogram Adjustment (AHA) algorithm increase image efficiency in terms of brightness and contrast. The Adaptive Mean Shift Modified Expectation Maximization (AMS-MEM) algorithm is used to segment the Alzheimer's disease Region of Interest. The second order 2-Dimensional Gray Level Co-Occurrence Matrix is used to measure the different functions (2D-GLCM). The Deep Learning (DL) method is used to identify disease images and stages based on a range of characteristics. The Deep Convolutional Neural Network (DCNN) is a classification method used to identify diseases in order to make accurate diagnosis decisions. The experimental findings show that the newly developed approach is more accurate and efficient than the current system. 1 MRI input image 2D- Adaptive Bilateral Filter Enhancement by AHA Image Clustering MEM Setting Image Training Disease Threshold using 0 &Testing Classification AMA using DL Fig. 1

Description

MRI input image

2D- Adaptive Bilateral Filter

Enhancement by AHA

Image Clustering MEM

Setting Image Training Disease Threshold using 0 &Testing Classification AMA using DL

Fig. 1

TITLE OF THE INVENTION The Computer Aided Diagnosis (CAD) System for the Detection of Alzheimer Disease Using MRI Real Images

FIELD OF THE INVENTION

[001]. The present disclosure is generally related to a smart bag that is helpful to the parents and school children with respect to monitoring, safety, and assistance.

BACKGROUND OF THE INVENTION

[002]. The primary objective of this invention is to propose a computer aided diagnosis system for the detection of Alzheimer disease. The invented system is designed in such a way that it detects the Alzheimer disease using the MRI real images. The invented system uses Deep Convolutional Neural Network classification method to distinguish normal and irregular images. The 2D Adaptive Bilateral Filter algorithm is used in the image restoration process. Image enhancement strategies are used to increase image efficiency in terms of brightness and contrast. The invented system will be useful in the detection of Alzheimer disease with good accuracy.

SUMMARY OF THE INVENTION

[003]. Computer Aided Diagnosis (CAD) is a fascinating subject for Alzheimer's disease research (AD). Several architectures are based on a reasonably large training database. Small clinics, on the other hand, are often unable to procure adequate materials for intensive identity preparation. While knowledge sharing in science research is increasing, it is unclear if a structure built on a single database is well suited to other tools. The performance of the system increased by about 20% as compared to a system built on the original small dataset. The findings revealed that the devised approach is an innovative and efficient form of CAD in clinics, despite the fact that the data set for learning is small.

[004]. Traditional deep learning approaches and associated techniques are being used to reliably diagnose AD, and they have become a popular solution for AD diagnostics. However, gathering data from different modalities takes time and effort, and certain modalities can have radioactivity-related side effects. Systemic Magnetic Resonance Imaging is the subject of our invention (sMRI). Their goals are to: 1) boost accuracy to levels comparable to cutting-edge approaches; 2) resolve the issue of overcoming; and 3) investigate proven brain indicators of perceptible AD diagnostic functionality. The disease biological markers play a big role in the accuracy of an AD evaluation.

[005]. Hippocampal shrinkage is identified in the early stages of Alzheimer's disease, and it has been linked to memory loss. Systemic MRI analysis is a popular and effective technique for diagnosing Alzheimer's disease. The efficacy of software-assisted diagnostic models using brain MRI images is primarily determined by considerations such as the following: 1) nonlinear recognition based on image information, 2) brain tissue clustering A procedure for removing landmark based features that does not require nonlinear recognition and tissue clustering is devised to solve the limitation. For image filtering, a creative neighborhood filter that keeps the edge is recommended. The filter's main goal is to use a hybridde-noising feature to compensate for image quality while trying to keep the image's edge.

[006]. The new approach preserves the edges of the area by filtering it using image boundaries. The area between the object and the sensor influences the response of a variable in a pixel region during filtering. Furthermore, the measuring edge between it and the pixel does not result in the estimation of the gray point picture average, which reduces vibration and causes material blurring.

[0071. When the level of noise is minimal, the effect of material on image distortion can be greater than the impact of noise cancellation, and grouping the original image produces better outcomes than grouping the standard image. The results of machine representations suggest that the method overcomes the limitations of modern methods focused on vector image efficiency measurements. It's also narratively simple and fast to implement.

[008]. The MRI model is a powerful tool for brain imaging. While being collected, MRI brain images can be interfered with, lowering image quality and limiting diagnostic reliability and accuracy. The removal of noise from clinical photographs is an essential part of per-processing, and there are many methods for doing so. To eliminate noise in MR, this invention examines various de-noising approaches such as non-local steps, critical factor investigation, bilateral and temporally high adaptive non-local means (TANLM) filter. A structural relationship histogram is a grayscale-based analysis of the pixel brightness distribution degree that uses histogram similarity to boost flawed contrast in various objects.

[009]. During pre-processing of contrast enhancement, the approach produces significant results and facilitates subsequent CAD techniques, reducing detection time and increasing precision. Histogram equalization (HE) and selective eclipse of the adaptive histogram are two common methods for enhancing contrast images (CLAHE). HE creates an image pixel elements values histogram in the image based on the original exported MRI brain image, displaying each pixel element's feature. It also increases the contrast between the images by changing the original pixel amounts.

[0010]. In this invention the the Computer Aided Diagnosis (CAD) method is proposed that uses MRI real images which uses different algorithms to diagnose and identify

Alzheimer disease. Alzheimer's disease is a life-threatening disease, but the picture and tool for diagnosing it is prohibitively expensive. The Alzheimer's disease (AD) in MRI brain images is automatically segmented and classified. 2D Adaptive Bilateral Filter (2D ABF) and Adaptive Histogram Adjustment are used to properly preprocess the tool (AHA). Modified Expectation Maximization is used to segment the ROI (MEM). GLCM is used to extract different image characteristics, and the DCNN classification method is used to distinguish normal and irregular images. Deep Convolutional Neural Networks have a classification accuracy of over 98 percent.

DETAILED DESCRIPTION OF THE INVENTION

[0011]. Computer Aided Diagnosis (CAD) is a fascinating subject for Alzheimer's disease research (AD). Several architectures are based on a reasonably large training database. Small clinics, on the other hand, are often unable to procure adequate materials for intensive identity preparation. While knowledge sharing in science research is increasing, it is unclear if a structure built on a single database is well suited to other tools. The performance of the system increased by about 20% as compared to a system built on the original small dataset. The findings revealed that the devised approach is an innovative and efficient form of CAD in clinics, despite the fact that the data set for learning is small.

[0012]. Traditional deep learning approaches and associated techniques are being used to reliably diagnose AD, and they have become a popular solution for AD diagnostics. However, gathering data from different modalities takes time and effort, and certain modalities can have radioactivity-related side effects. Systemic Magnetic Resonance Imaging is the subject of our invention (sMRI). Their goals are to: 1) boost accuracy to levels comparable to cutting-edge approaches; 2) resolve the issue of overcoming; and 3) investigate proven brain indicators of perceptible AD diagnostic functionality. The disease biological markers play a big role in the accuracy of an AD evaluation.

[0013]. Hippocampal shrinkage is identified in the early stages of Alzheimer's disease, and it has been linked to memory loss. Systemic MRI analysis is a popular and effective technique for diagnosing Alzheimer's disease. The efficacy of software-assisted diagnostic models using brain MRI images is primarily determined by considerations such as the following: 1) nonlinear recognition based on image information, 2) brain tissue clustering A procedure for removing landmark based features that does not require nonlinear recognition and tissue clustering is devised to solve the limitation. For image filtering, a creative neighborhood filter that keeps the edge is recommended. The filter's main goal is to use a hybridde-noising feature to compensate for image quality while trying to keep the image's edge.

[0014]. The new approach preserves the edges of the area by filtering it using image boundaries. The area between the object and the sensor influences the response of a variable in a pixel region during filtering. Furthermore, the measuring edge between it and the pixel does not result in the estimation of the gray point picture average, which reduces vibration and causes material blurring. When the level of noise is minimal, the effect of material on image distortion can be greater than the impact of noise cancellation, and grouping the original image produces better outcomes than grouping the standard image. The results of machine representations suggest that the method overcomes the limitations of modern methods focused on vector image efficiency measurements. It's also narratively simple and fast to implement.

[0015]. The MRI model is a powerful tool for brain imaging. While being collected, MRI brain images can be interfered with, lowering image quality and limiting diagnostic reliability and accuracy. The removal of noise from clinical photographs is an essential part of per-processing, and there are many methods for doing so. To eliminate noise in MR, this invention examines various de-noising approaches such as non-local steps, critical factor investigation, bilateral and temporally high adaptive non-local means (TANLM) filter.

[0016]. A structural relationship histogram is a grayscale-based analysis of the pixel brightness distribution degree that uses histogram similarity to boost flawed contrast in various objects. During pre-processing of contrast enhancement, the approach produces significant results and facilitates subsequent CAD techniques, reducing detection time and increasing precision. Histogram equalization (HE) and selective eclipse of the adaptive histogram are two common methods for enhancing contrast images (CLAHE). HE creates an image pixel elements values histogram in the image based on the original exported MRI brain image, displaying each pixel element's feature. It also increases the contrast between the images by changing the original pixel amounts.

[0017]. The dynamic model focuses on individual objects to represent pixel similarity functions for optimization and controls the strength of each individual object in an evolutionary manner.

[0018]. As a result, incorporating medical data segmentation remains a daunting challenge that has piqued the interest of a few researchers in the last year. The area of interest is differentiated using MRI. The idea is to approach this problem as a classification challenge, with the aim of distinguishing between normal and abnormal elements on an MRI image based on a variety of characteristics such as strength and form. More precisely, they recommend Support Vector Machine (SVM), which is a well known and well-motivated classification strategy.

[0019]. The deep neural network is a deep learning approach that has been shown to be suitable for various image classifications. For optimal outcomes, the CNN precisely monitors various image recognition functions. Clinical image databases, on the other hand, are difficult to come by because they require a great deal of technological expertise to mark. The thesis looks at how the CNN based algorithm can be used to diagnose pneumonia in a chest X-ray dataset.

[0020]. Kernel support vector machine classification with local free rotating and path characteristics, learning transfer from the ground up on two CNN models: probabilistic neural Group i.e. VGG16 and Inception V3, and module network training are among them. Information raise is a type of data pre-processing that can be used in any of the three processes. The results of the experiments show that increasing knowledge is normally an effective way to improve the performance of all three algorithms. When compared to an aid support vector with robust, isolated specific characteristics and capsule network dependent quick and Rotating Binary, transfer learning is also a more powerful method of recognition on a restricted database (RB).

[0021]. For transfer learning to be effective, specific applications must be retrained on a current target collection of data. The second critical aspect is that the system's difficulty is sufficient for the set of data measures. The statistical approach for a variety of neurological disorders varies based on computerized and accurate systemic categorization and distinction. The DL-based recognition and segmentation systems have recently attracted research attention due to their self-learning capabilities over large sets of data. To categorize the collection of data for the MRI brain image, a convolutional neural network (CNN) must use preprocessed feature maps in the Curvelet system.

[0022]. Curvelets have a better feature vector, and the extracted functions are far more precise than traditional wavelet transformations due to their multi-directional capabilities. Following that, the methods of distinction used to invention anatomical structures and the location of brain tumors are discussed, followed by the CNN quality. As opposed to wavelet transformation and recognition using traditional classification techniques like SVM and Probabilistic Neural Networks, feature extraction in the Curvelet domain and CNN have a higher level of precision (PNN).

[0023]. Material And Methods: Pre-processing, segmentation of the area of interest, attribute extraction, and disease classification are the four stages of the invented procedure. During the training phase, the training MRI Brain image is fed into the network as an input and is subjected to all of the preceding steps. The extracted features are combined with the class labels to train a classifier using a classification model. The algorithm has now been transformed into a classification algorithm. During the research point, a test brain MRI image is information into the structure, and the image goes through all of the above stages. Using the details it gathered during the training phase, the trained classifier will now identify the characteristics of the test object and assign a class label to it, such as 'affected disease' or 'unaffected disease'.

[0024]. The invention incorporates a new predictive method for segmenting the entire brain into image sequences and calculating its density to diagnose illness using magnetic resonance imaging (MRI). The related MRI brain scans were obtained from the Alzheimer's Neuro-imaging Initiative's website (ADNI). The entire MRI brain Ti weighted MRI was considered at 1.5 T level in participants, as stated in the specification. The invented automatic clustering approach is based on image numerical anatomy, and our "head concept" algorithm is used to limit the MRI brain pixel size.

[0025]. MRI brain image Pre-processing: The overall architecture of the invented technique is depicted in Figure 1. Two methods called image reconstruction and image enhancement are used to complete the MRI image preprocessing technique. Using the 2D Adaptive Bilateral Filter (2D-ABF) algorithm, the image restoration is completed. Unwanted disturbances such as speckle noise, binary noise, and random noise are removed using the 2D-ABF.

[0026]. The 2D ABF is used to process noises in MRI images without compromising the original information quality. The 2D Adaptive Histogram Adjustment (2D AHA) algorithm can be used to create an image quality enhancement technique. The 2D AHA is used to improve image quality by increasing visual contrast and brightness. For MRI brain image segmentation, the gray image is the most useful format. If it selects an MRI color image, it must be transformed to a grayscale image. On gray files, the MEM algorithm can cluster odd pixels.

[0027]. Image Restoration using 2D Adaptive Bilateral Filter (2D-ABF): De-noising has long been a research focus, but there is still room for improvement, particularly in picture de-noising. The simple temporal filtering of a distorted image can be efficient when high frequency distortion is to be removed from the distorted image. The most significant issue is the difficulty of the computation used to perform the convolution. Since noise is spread over all wavelengths, wavelength-based de-noising methods use low-pass filtering to eliminate most high-frequency components in an effort to de-noise the signal.

[0028]. However, it is rarely effective because it disrupts all noise and other high frequency image characteristics, resulting in an image that is overly smooth denoted. The aim of image de-noising is to remove noise while maintaining as much visibility as possible of important image features like boundaries. Vector filters, which consist of transforming the image with a steady vector to obtain a linear mixture of neighborhood values, are widely used to reduce noise in the presence of exponential noise.

[0029]. As a result, it can provide a fuzzy and smooth picture with poor feature approximation and noise suppression. The forward and reverse Fourier transformations of a Gaussian function are true Gaussian structures, so Gaussian-based filters are particularly important. Alternatively, as the frequency domain filter shrinks, the temporal domain filter expands and modulates the low frequencies, resulting in increased smoothing and blurring. Traditional linear filters, such as Gaussian filters, were widely used to de-noise images. The 2D Adaptive Bilateral Filter is used in the proposed picture de-noising method. A discrepancy between the original image and the de-noised version implies that the algorithm has erased noise, which is referred to as machine noise. In principle, the system's vibrations would feel like a stimulus.

[0030]. Even if even high-quality photographs have some noise, it makes sense to try out some de-noising techniques in this way, rather than following the old trick of "using noise and then suppressing it.

[0031]. Image Enhancement: 2D Adaptive Histogram Adjustment (or 2D-AHA for short) is a machine-based image processing technique for boosting or increasing image strength. AHA is suitable for photographs that are both normal and therapeutic. Rather than applying a map or conversion to the whole image, it is done separately on sub-images in this method. AHA applies a system for correctly running and merging the sub-image. To combat the over-amplification of noise, an image quality control approach was incorporated into the image histogram correction methodology. This differs from standard histogram correction in that it improves the accuracy of a given portion of the MRI image by using several histograms, all of which are proportional to a single area of the image, to reallocate image quality metrics like brightness and contrast measurement. Instead of standard histogram equalization, AHA enhances the contrast of an image, which improves transparency but tends to intensify distortion.

[0032]. Proposed System: Step 1. Input (MRI image) is interpreted or perused. Step 2. Change the MRI image into MRI at Gray point. Step 3. To maximize the image quality, add or conduct the Histogram Equalization technique to the reference image. Step 4. Take the improved-image histogram. Step 5. Consider or execute the Local Equalization Histogram (LHE). Apply method for improving image contrast on the input image (MRI image). Step 6. Step 4 has to be repeated. Step7. Enable or execute Adaptive Histogram (AHA) Adjustment. Apply method for improving viewpoint intensity on the reference image (MRI viewpoint). Step 8. Do again step 4. Step9. To improve image quality, apply or execute the quality dependent adaptive histogram equalization technique on the input image. Step 10. Repeat step 4.

[0033]. Alzheimer segmentation : To begin, the MRI information spread is approximated using the Expectation-Maximization principle. The Bayesian Knowledge Criteria decide the category number. To group pixels in images into the closest segment, the Highest Probability method is used. The accuracy of the solution adopted is independent of the initial estimation and can be applied to data clustering without supervision. The K-means formula was used to configure the MRI's Gaussian Mixture Attributes. The mixed model form numbers are determined using the Bayesian criterion and the expectation maximizationprocess.

[0034]. This process incorporates the computation of parameters and the selection of models into a single stage, resulting in fully unsupervised clustering. In the second part, the Gaussian Mixture scheme is briefly applied, and a solution is proposed based on the 'Expected Total Bayesian Information Criterion' methodology. In the third part, a complete clustering scheme will be given. The fourth feature is analysis; real MRIs are used in the clustering process, and the results show that the technique is useful in this invention. Original clustering was achieved using the K-means algorithm. The measurement process can be defined as follows:

[0035]. Step 1: Starting By the number of groups k, use the initial segmentation centers.

[0036]. Step 2: The pixel specifics are organized into the column with the smallest pixel size in the centre.

[0037]. Step 3: Get started. After a classification method, calculate the mean of the pixel values within the class.

[0038]. Step 4: The current classification is the result of clustering the K-means, unless the words with the clusters have the same value.

[0039]. Step 5. Using the current stage, edit the definition of the clusters and then continue to step 2.

[0040]. Expectations that have been changed The maximization algorithm heavily relies on gradient-based activation of clustering of pixels in different regions of the image. A common approach for determining the final option is to calculate Peak Probability, but this adds a lot of work to the estimation. The average K procedure is used to configure the mixture parameters in this invention. The difficulty of measuring class level is the reason for using Gaussian mixture. Too many methods assume that the number is directly defined, that is, that clustering determines the amount of each category of image.

[0041]. Clearly, this is a controlled or semi-supervised operation. In order to solve this problem, a selection criterion for the model Bayesian Information Criteria (BIC) is used. The first phase in the Clustering Methodology is called class approximation. To estimate the numbers of the mixing variables and Gaussian components, the 'Expected Limit with Bayesian Criterion' method is used. The second stage is pixel sorting. Each pixel is assigned to a certain class using the Maximum Likelihood theorem.

[0042]. Threshold using Adaptive Mean Shift: The threshold technique is useful for segmenting images and detecting patterns. It's a technique for dividing an image into different areas. Selecting acceptable gray scale values as thresholds and categorizing the picture into more than one field is the most straightforward approach. Automated threshold strategies are divided into two categories: global and local. Global methods use constant pixel values for the whole image, while local strategies use dynamic thresholds. Though both of these methods optimize a criterion attribute based on information obtained from histogram construction or spatial distribution.

[0043]. The AMS threshold method, which is focused on distinguish analysis to evaluate the highest class separable and is used to do histogram-based image threshold effectively, is one of the most well-known and widely understood adaptive threshold techniques. By considering all pixels with a group value, the AMS threshold guarantees successful clustering. The output threshold in standard Fuzzy C Means is dependent on the input threshold, which is set at random, so an appropriate initial threshold must be sought in order to overcome the complexities. The gray color values were selected as the Fuzziness component. The threshold algorithm is used to assess abnormal pixel differentiation. The hardening mechanism comes before the fuzzy c.

[0044]. Feature Extraction: The feature report based on a skewed picture is easily labeled as normal or abnormal for feature extraction during this phase. Function extraction is a method of classifying characteristics to represent the raw image to a portion that needs to be extracted from the MRI image. The amount of data needed for consistently identifying large database the characteristics used as inputs to define and location in the class defined is reduced with feature extraction. The goal of feature extraction is to reduce the amount of actual information by evaluating the beneficial property, or attributes, that distinguish one sample from another. For indexing and identification of related images, taking into account numerical attribute vectors is beneficial.

[0045]. The texture mechanism provides information on image intensity characteristics such as homogeneity, softness, angularity, and contrasting at the diffusion level. Based on the probability and allocation of the pixel level strength, statistical texture features such as mean, kurtosis, variance, image energy, standard deviation, skewness, entropy, and smoothness are computed.

[0046]. For human sensory perception from deep learning, texture analysis efficiently separates natural from abnormal products. It also makes a distinction between normal and malignant tissues that may not be visible to the naked eye. By choosing appropriate early diagnostic comparative characteristics, it increases accuracy. Data from the histogram of pixel intensity was derived from the first-order statistical textural information associated during the initial stage, and grey-level frequencies were measured at arbitrary image positions. It doesn't notice any co-occurrences or associations between pixels.

[0047]. The second-order textural evaluation features were then obtained using the probability of pixel values at arbitrary ranges and across entire image configurations. A appropriate set of attributes must be specified since the accuracy of a classification system is based primarily on the proper selection of the function. In this classification method, a feature matrix 2D-GLCM is used, which is a statistical technique that makes use of the pixel temporal relation. The GLCM is used to determine the attributes should be generated based on the cluster of a pixel.

[0048]. The invention is carried out by directing the distribution of 2D-GLCM features in such a way that a collection of rotation-invariant statistics is proposed. By taking the average and distribution of each mode of function over the four shapes that are used, the scalar invariant attributes of rotation can be obtained from vectors of co-occurrence. Gray-level variation info, which is directly linked to 2D-GLCM, is another form of texture classification. The continuum of co-occurring elements at a given offset is described over an image by a matrix of co-occurrence, also known as a distribution of co occurrence. It defines the structural relationship between distance and angle over a similar-scale picture sub-region. A gray-level image is used to construct the 2D-GLCM. The 2D-GLCM measures how much a gray-level pixel value appears to adjacent pixels horizontally, vertically, or diagonally. The 2D-GLCM matrix can be a well-known statistical method for extracting texture information from second-order images. The 2D GLCM vector is one of the most common and efficient forms of texture evaluation characteristics.

[0049]. For an area defined by a user set frame, 2D-GLCM is the vector of all quantities for all gray level couples. Rather than the original gray-level pixel numbers, the absolute differences between couples of gray-lines or mean 2D-GLCM lines are used to evaluate attributes in this method. In comparison to the GLCM case, this feature makes the numbers a bit more accurate for variations in illumination. The frequency of the gray level is extracted as a vector from the above images. For the purposes of this invention, classification can be described as the ability to recognize a group of categories to which a picture belongs, whether normal or affected by Alzheimer's disease. Deep Learning can be used to successfully accomplish recognition decision-making tasks (DL).

[0050]. Alzheimer disease classification Deep Learning (DL): The CNN technique is used to build the DCNN. The several stacked CNN model is referred to as DCNN. The use of several layer stacks would improve recognition performance. There are two forms of Alzheimer's disease: benign and malignant. Successful and reliable cancer diagnosis and treatment leads to a higher quality of life and a longer life span in all cases. One of the most functional and efficient methods is to use DNN. Using images from body MRI images, a DCNN was used to detect an illness.

[0051]. CNN uses the proposed approach to identify and describe the condition based on brain scans. The main difference between the neural network's major channels and the traditional neural network is that they can dynamically and globally extract the attribute from each pixel. This networks are made up of neurons and can be learned with the help of weights and biases. The deep learning algorithm is used for recovery features. The used algorithm was the clustering algorithm, which was applied to the data collection, and then the images were applied to the DCNN.

[0052]. The suggested solution was found to be accurate, according to the results. The purpose of acquiring the property before applying it in the DCNN is to assess fatty masses as disorder in some images, or the disorder is wrongly treated as fat in some images, which will result in increased drug errors. Better network stability and accuracy are achieved by extracting the parameter first and then applying the DCNN. Deep learning is one of the most current and realistic methods for computers to understand. To phrase it another way, studying is referred to as deep-seated architecture. Those implementations are, in effect, the same old nerve systems that have evolved into DNN.

[0053]. These networks are data-driven, and role formation happens automatically and without human intervention, which is why they are both reliable and efficient in a variety of fields. In essence, this is a deep learning of a set of nerve-related techniques that dynamically learn features from our own feedback results. The CNN is a kind of DNN whose architecture is inspired by the vision cortex of cats. The CNN is structured in a bureaucratic manner and has several layers. CNN uses input, output, convolution, pooling, standardization, and Completely Connected layers among other strategies. The number of layers used, the size, the number of images used, and the type of kernel feature used differentiate CNN. The parameters are chosen using CNNs, based experimental outcomes, and trial-and-error methods. In other words, each CNN is made up of many layers, the most important of which are the Convolutional layer and the Sub-sampling layer, which are added in the following points:

[0054]. 1) Convolution layer: Standard parameters, such as metrics, have specified values that are exactly the same as other image parts. It means that acquired characteristics from one part of the image will be applied to other parts as well, and that all portions of the image will have similar characteristics. The Convolutional layer functions are used to distinguish images after the application's scheme.

[0055]. 2) Sub-sampling layer: This layer's functions are used to reduce the image input's complexity. It gets a location matrix from the DCNN layer at the top. Mean pooling or peak pooling refers to the aggregation or Sub-Sampling operation.

[0056]. In a few cases, certain areas of fat in the photographs are misidentified as illness, or the practitioner does not know the disease; the most accurate diagnosis is entirely dependent on the physician's ability. The DCNN was used via brain scanning to detect sickness with this innovation. There are extra margins on the photographs obtained from the imaging centers. To prevent the photographs from making noise, both margins have been trimmed. To recover the object extraction of the images and increase the network's stability, some of the main reasons for using and combining the extracting features strategy with the DCNN are to recover the object extraction of the images. According to the findings of the DCNN on the initial photos, a new technique combining cluster analysis for feature decomposition and DCNN is proposed in this research to improve network precision.

[0057]. We used a transition learning technique in DCNN for training on topics that are close to the same problem that should be overcome. A training model on classification may use several stacks of layers in the training model.

[0058]. The photos were originally uploaded to CNN without any kind of feature extraction. The DCNN architect was used to identify and categorize photos, which included five Convolutional layers and three sub-sampling layers, standardization layers, standardization layers, Fully Connected layers, and finally classification layers. With 4096 neurons, the layers are totally linked. There are two types of patients in this layer: patients with brain injury and normal patients.

[0059]. The DCNN architecture is used to identify Alzheimer's disease in this invention. Experiments show that our novel DCNN architecture achieves maximum productivity and outperforms the majority of cutting-edge systems. The outcome clearly demonstrates the efficacy and advantages of the proposed two-stage transfer learning approach, as well as the potential for learning information from MRI results.

[0060]. On MRI real images, the Computer Aided Diagnosis (CAD) method is proposed, which uses different algorithms to diagnose and identify Alzheimer disease. Alzheimer's disease is a life-threatening disease, but the picture and tool for diagnosing it is prohibitively expensive. The Alzheimer's disease (AD) in MRI brain images is automatically segmented and classified. 2D Adaptive Bilateral Filter (2D ABF) and Adaptive Histogram Adjustment are used to properly preprocess the tool (AHA). Modified Expectation Maximization is used to segment the ROI (MEM). GLCM is used to extract different image characteristics, and the DCNN classification method is used to distinguish normal and irregular images. Deep Convolutional Neural Networks have considerable classification accuracy.

Claims (6)

1. CLAIMS: We Claim: 1. This invention is related to the computer aided diagnosis system for the detection of Alzheimer disease.
2. 2. This invented system will detect the Alzheimer disease using the MRI real images.
3. 3. As claimed in 1 and 2, the invented system uses Deep Convolutional Neural Network classification method to distinguish normal and irregular images.
4. 4. The 2D Adaptive Bilateral Filter algorithm is used in the image restoration process.
5. 5. Image enhancement strategies are used to increase image efficiency in terms of brightness and contrast.
6. 6. The invented system will be useful in the detection of Alzheimer disease with good accuracy.

   MRI input image

   2D- Adaptive Bilateral Filter 2021101725

   Enhancement by AHA

   Image Clustering MEM

   Setting Image Training Disease Threshold using &Testing Classification AMA using DL

   Fig. 1