Efficient Alzeihmer’s disease detection using Deep learning Technique

doi:10.21203/rs.3.rs-2575795/v1

Download PDF

Research Article

Efficient Alzeihmer’s disease detection using Deep learning Technique

https://doi.org/10.21203/rs.3.rs-2575795/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 17 May, 2023

Read the published version in Soft Computing →

You are reading this latest preprint version

The human brain serves as the primary control centre for the humanoid system. Computer vision plays a vital part in the field of human health, which helps to reduce the amount of human judgement that is required to produce accurate findings. Scans using computed tomography (CT), x-rays, and magnetic resonance imaging (MRI) are the most popular imaging technologies used in MRI, and they could also the greatest trustworthy and safe. The MRI can identify even the most minute of objects. In this paper Alzheimer’s disease detection in early-stage, based on MRI by using the deep learning technique U-Net and Efficient net which is a Convolutional neural network is implemented. Diagnosing of Alzheimer's disease (AD) accurately is an vital aspect in treating AD patients, eventually during the early disease stages. This is particularly true in the early disease stages of the disease, when awareness of risk, enables AD patients to takeup protective measures well before the occurrence of brain damage that cannot be reversed. Despite of the fact that computers have been utilised in a significant number of recent research to diagnose AD, the majority of machine detection approaches are restricted by congenital findings. Early-stage Alzheimer's disease (AD) can be identified, but early-stage AD cannot be predicted because prediction of the disease is successful only before the (AD) disease reveals itself. Deep Learning, often known as DL, has recently emerged as a popular method for the initial recognition of Alzheimer's disease (AD). In this article, we will give a quick overview of some of the key research that has been done on AD, and we will investigate how DL can assist researchers in the early phases of disease diagnosis.

Deep Learning

MRI

Alzheimer's disease

EfficientNet

SVM Classifier

Medical Image Processing

Alzheimer's disease (AD) is a form of brain disease-dementia, which influence a person's nervous system and ultimately nerve-cells in the brain die. The commencement of AD is typically slow and gradual, and the early symptoms may initially be mistaken for the natural progression of ageing or simple forgetfulness. The progression of the disease causes a decline in the patient's cognitive capacities, which includes the decision making of the patients and do daily chores. At this time, there is no treatment that may reverse the effects of the condition; all that can be done is adhere to a set of recommendations that may slow down the disease's progression. Because of this, a correct diagnosis will be an important aspect in order to enhance the patients' overall quality of life.

AD is a brain related ailment that slowly destroys a patients memory and also the ability like thinking, decision making and more worstly impacts the person completing from doing their basic tasks.The decline of intellectual faculties, as well as memory, judgement, and personality, to the extent that everyday functioning and quality of life are severely compromised as a result. AD Typically affects people over the age of 65 and can decrease brain function, which can eventually lead to dementia. There is a wide range of estimates, but most experts agree that more than 6 million people in the USA aged 65 and above could have AD. There are a significant number of people under 65 who also have the condition.

The prognosis of Alzheimer's disease AD & the methods of machine learning (ML) will be the focus of the work that will be done with the intention of determining the connection between the two fields. In order to accomplish this, both the areas will be individually researched, and an consorious perspective on the present situation will be obtained. This insight will make it possible to begin in-depth investigation, which, once the current issue has been comprehended, will make it possible to determine whether or not there is a potential solution that can be achieved through techniques of machine learning.

The ultimate purpose of this paper is offered as questions to be investigated in the followingsection. For accomplishing this primary objective, a variety of subsidiary inquiries have also been formulated. These supplementary questions will make it possible to structure the path that leads to the ultimate sound solution.

What kinds of machine learning strategies could be applied to make the prognosis of AD more accurate?
What kind of data are going to be essential for the successful training of the system?
Which architecture and parameter settings are going to produce the most accurate results, and how do you choose those settings?
To what degree of precision is it possible to get results?
Which framework(s) might be applicable for putting the chosen model into practise and putting it through its paces?

Most of the research works use imaging techniques in where magnetic imaging such as MRI and CT scans of the human brain are commonly used to determine various brain related diseases or impairments which help in to identify its location in order to make judgements regarding subsequent treatment steps. Because of their portability and, more importantly, their ability to produce high-definition images of diseased tissues, these two types of scans continue to see widespread application. At the present time, there are a variety of additional approaches that can be taken to treat tumours. These approaches include chemotherapy, radiation therapy, and surgery, amongst others. The various parameters, such as the size, type, and grade of the tumour that is shown in the MR picture, are taken into consideration while deciding which treatment to use. In addition to this, it is responsible for determining whether or not the cancer has spread to other parts of the body.

It is extremely important for treatment operations to have an accurate diagnosis of the type of brain abnormality being treated in order to reduce the number of diagnostic errors that occur. When it comes to precision, computer-aided diagnostic (CAD) systems are frequently a patchwork solution. The primary objective of computer based automated solution is to generate a good output, in the form of an associate estimation, with the goals of assisting medical professionals in visual comprehension and reducing the amount of time spent viewing images. These developments make medical diagnosis more reliable and accurate; yet, segmenting an MR picture of the tumour and the space it occupies can be a very challenging task. The appearance of disease in particular locations of the less intensified brain image is an added difficulty which makes the computerised identification of brain diseases a challenging task.

To the present day, analyzing data from neuro imaging, such as that which is gained from MRI, positron emission tomography, Functional-MRI has mostly done by the specialized doctors and surgeons, such as radiologists and physicians, because it requires a potency of specialisation. Some of these imaging techniques include: AD, which is the extremely usual kind of dementia and often impacts people more than the age of 65, is defined as the continous degradation of cognitive & memory abilities. Timely therapy is essential, and in order to receive that treatment, early detection of AD and its different forms of stages are required. This allows for a slower progression of dementia (MCI). To reach to this goal, it is required to obtain a trustworthy diagnosis with the help of brain images. A powerful diagnostic system that is added by the analysis of neuro-imaging data will make it possible to take an approach that is both further helpful and trustworthy, and it could also enhance accuracy of diagnosis. On the other hand, given what we know now about how the brain works, this assumption is not valid.

In recent years, machine learning (ML) techniques that are able to take into account the intercorrelation between regions have emerged as a desirable and basic component of techniques related to computer aid. These techniques are also widely used for the automatic diagnosis and evaluation of neuropsychiatric disorders. In spite of the fact that a variety of machine-learning standards are applied to the automatic prediction of neurological disorders, the support vector machine (SVM)-based and deep learning (DL)-based diagnosis models continue to be two of the most important research directions. In this context, numerous in-depth reviews that are connected to medical imaging and the application of ML techniques have been published. Because they are unable to extract adaptive features, models based on SVM automated diagnosis models for neuropsychiatric illnesses typically rely on characteristics that have been hand-crafted instead. The functional-connectivity (FC) patterns, which show correlations between brain regions, are a common aspect of the models based on SVM used for diagnosis are now in use. Individual FC patterns are recovered for paired regions of the brain that have been segmented and labelled automatically according to their anatomical structure. Main Drawback of SVM is related to its bad accomplishment with respect to raw-data, as well as the fact that it requires the adept usage of design techniques in order to bringout edifying features, despite the fact that it is quite popular.

In an innovative and effective study, Sivaramakrishnan et al. (2013) [1] used a Fuzzy C-method grouping algorithm with histogram equalisation to pinpoint the exact site of a brain tumour in an imaging study. This became accomplished. The utilisation of primary factor assessment is carried out in order to lessen the extent of the wavelet coefficient, which ultimately results in the disintegration of images. The results of the forecast FCM cluster method could remove the tumour area on MR images in an accurate manner.

M. M. Sufyan et al. [2] has provided a detection method for brain tumour segmentation that makes use of an enhanced edge technique and relies mostly on the identification of Sobel features. The authors used thresholding operation using sobel technique. and it found various solutions by making use of a secure contour procedure. Following the conclusion of that procedure, cancer cells are isolated from the final image by analysing the intensity values of the pixels.

A variety of clustering algorithms, including K-means, improvised K-means, C-means, and improvised C-means algorithms, were introduced by Sathya et al. (2011) [3] in their study. In their study, they described an experimental analysis for large data sets that were composed of individual pictures. They employed a large number of parametric tests in order to investigate the identified ramifications.

It has been suggested by B. Devkota and colleagues [4] that a computer-aided detection (CAD) method can be utilised to identify aberrant tissues through the application of morphological procedures. The operations such as opening and closing morphologically are the preferred methods of segmentation out of the many different segmentation approaches that are currently available because they require less processing time while providing the highest level of effectiveness in removing tumour areas with the fewest possible errors.

K. Sudharani et al. [5] applied a K-nearest neighbour method to the MR images in order to locate the hysterically fully-fledged component and keep it confined within the aberrant tissues. The method that is being offered for the task is a slow one, but it results in beautiful effects. The sample training step is directly responsible for the accuracy.

In the process of segmentation, Jaskirat Kaur et al. (2012) [6] created a limited cluster approaches and carried out an evaluation of different styles for those strategies. Kaur proposed a method to evaluate various clustering strategies by comparing them to the degree to which they remained consistent in challenging bids. In addition to this, they outlined the various tests for evaluation of performance using accuracy, specificity and sensitivity.

J.T. Kwok et al. [7] presented photo fusion method using wavelets fusion to readily realize item with respect to focus length. This was accomplished since numerous vision-related processing jobs are able to be done more easily when all aspects within the photographs are brilliant. In their research, Kwoke et al. studied numerous datasets, and the findings demonstrate that the provided work is more accurate since it does not suffer from evenness at various phases of computations.

Dalia Mahmoud et al. [9] developed tumor detection method on binary image using Artificial Neural Networks (ANN).The authors developed automated reconizition system for MRI that uses ANN.

Automated three-dimensional segmentation of brain MRI data was presented by L. Marroquin and colleagues [10]. When compared to using a single multiplicative magnificence, the utilisation of many parametric models will result in a reduction in the influence felt by a grandeur's intensities. Brain atlas has been commissioned to locate a non-rigid conversion in order to map the typical brain. In the following step, the brain is separated from the rest of the tissues in the body using this transformation. After computing the prior probability and locating the automatic initialization, the MPM-MAP algorithm is then utilised to determine the optimal brain segmentation. The study came to a number of important conclusions, one of which was that the MPM-MAP approach is more error-resistant than the EM algorithm when it comes to calculating the posterior marginal. The MPM-MAP algorithm, which is used for optimal segmentation, requires just the solution of linear systems; as a result, it has a high level of computing efficiency.

Deep Learning (DL) standards, enable a system to utilize raw information as an input, which enables the system to automatically uncover extremely discriminatory features within the provided training data set12. The essential philosophical groundwork upon which DL is built is an end-to-end learning design concept. End-to-end learning offers more advantages and optimize concurrently of all stages of the processing pipeline, which may result in improved overall functionality. In this study, a hierarcy for brain MRI analysis is put out for consideration [11–13]. There are a total of four levels in the hierarchy, starting at level 1 (none) (full). The majority of the already published research makes use of Level 1 or Level 2, whose results are very dependent on the specific software being used and, at times, even setting up hyper parameters and the elimination of noise. Because of such dependencies,only a fraction of main dataset is used for performance evaluation. As a result, apparent outliers were excluded, which made it difficult to compare the two groups' performances in an objective manner. An further advantage of end-to-end learning is the possibility of producing an excellent visual explanation of the reasoning behind the categorization conclusion that the CNN [19] arrived to. This explanation assists the physician in comprehending the actions of the CNNs and locating novel bio-markers. At Level 2, the description is confined to the segmented area, which may prevent the discovery of new bio-markers in deleted region. This is because the segmented component is the only part that is being explained.

Proposed Method

The computation task of neural network starts after the information that has been obtained through the processing and examination of the input data. In the future, it will be the responsibility of the neural network to learn from such data in order to generate predictions based on new data inputed. In order to accomplish this goal, we will work to develop an MLP. In this section, one of the first tasks that will be completed is going to be the partitioning of the dataset into a few different subsets. A training set, a validation set, and a testing sets will make up these sets, respectively. In general, the training set will be used to train the network, the validation set will be used to select the hyperparameters that are the most appropriate for the model, and the testing set will be used to test the results. The phase of the work devoted to the selection of the most appropriate features will be an important one. This is because it will provide correct results for future projections. According to what was indicated on the non-functional requirements document, the current module's implementation will be coded in Python and will have support from the Keras library.

Figure 1 : Proposed method for Alzheimer Disease Classification

After the data has been separated into its respective sets, now we move to next step, which is the development of the structure of the neural network. A Sequential-model, which can be thought of as a series of stacked layers, will serve as the foundation for this one. As has been stated previously, the requirement concerning the MLP design will serve as the primary focus of the selection process for this particular model. As a result of this, multiple layers will be added, each of which will have their activation-functions, number of neurons, and input-data specified. The model will be compiled after the structure has been constructed successfully. Parameters used for optimization [14, 15] for training the network and performance evaluation metrics such as loss function is specified well in advance befiore compilation starts.

The training of the network [16–18] will take place once the compilation of the model has been completed. Training data must be provided as an input in order to accomplish this goal. In order to conduct an investigation into the development of the training, validation data may also be supplied. It will be required to define the batch size, which is the number of records to utilise in each iteration, as well as the number of epochs, which are both going to be used as parameters. It is important to perform preliminary tuning on all of the parameters that were discussed in the phases of compiling and training, such as finding the optimal activation function, optimization technique[, or batch size. By carrying it out, correct findings will be produced.

At this stage, the neural network could be trained, and the next step is to evaluate its behaviour using fresh input data. In order to accomplish this objective, the following module will be responsible for processing each of the metrics that were defined in the compilation of the model earlier. The steps that need to be completed for this module are broken down and summarised in Fig. 1. It is clear from looking at this picture that the data frame values from the previous module will be used as the required input data for this module, and that the response of the model will be the output provided by this module. The response will be investigated further and handled accordingly.

Figure 2: Design of the implementation of the neural network

Table 1

Alzheimer MRI Preprocessed Dataset
Images	Size of the Image	Total MRiImages	Classes of Images
Preprocessed MRI	128 x 128 pixels	6400	4

Efficientnet Architecture

Although accuracy is the criterion that is most frequently used for evaluating models, Additionally, the evaluation of the model using all of distinct measures will make it possible to obtain good results.

The confusion matrix of the system is become the first metric could be displayed by this section of the system. This matrix will make it possible to conduct an analysis of the many sorts of accurate and inaccurate predictions produced by the model with regard to all of its categories. The evaluation of false-positives and false-negatives [19] is made possible by the tool. A false

positive occurs when a result is wrongly labelled as positive when it turns out that the result is actually negative. If the result is positive but wrongly labelled as negative, it is called as false negative. In light of the breadth of this undertaking, the examination of the matrix will be fraught with significance. Due to the fact that there are numerous potential ethical and moral debates that can take place over the significance of false positives or negatives in the diagnosis of Alzheimer's disease. Obtaining Curves such as a Receiver operating characteristic (ROC) for designed model will be a great help in summarizing between the true and the false positive rates for a predictive model that makes use of varying probability thresholds. This is something that will be really beneficial.

The final result of the accuracyof the model can be derived as a direct consequence of the model metrics that are defined throughout its computation. This is going to be a static outcome, which is going to match with the end result of the entire training. However, it will be important to analyse the progression of the model as it is being trained during the entire process. Using a graph to assess this progression is the most straightforward approach to do so. The progression of the error throughout the training periods will be depicted on this graph. On this basis, it will be able to determine if, for instance, the model is likely to learn continuously as the number of times used to learn is increased, or whether, on the other hand, the learning has achieved its maximum potential. As a result, the graphical analysis of the results for data used for training and validation are shown below, and accomplishing this will be the purpose of this module. By plotting the graphical data set used, it is possible to evaluate the model's ability to generalise.

The results obtained shows classification of data set into four classes

Figure 8: Graph between count and NonDemented, MildDemented, ModerateDemented, Very Mild Demented representing train data

Table 2

Alzeihmers disease accuracy and othe parametrs
	Precision	Recall	F1-Score	Support
0	0.96	1.00	0.98	90
1	1.00	0.80	0.89	10
2	0.98	0.99	0.99	306
3	0.98	0.94	0.96	103
Accuracy			0.98	509
Macro Average	0.98	0.93	0.95	509
Weighted Average	0.98	0.98	0.98	509

Using a Convolution Neural Network, we suggested a computerised method for the prediction of AD. The input MR pictures are then transformed into grayscale images after being read . During preprocessing step,the noises originally present in the image have been removed from these images by the use of an adaptive-bilateral filtering method.The picture that has been denoised is then subjected to binary thresholding, and then the Convolution Neural Network segmentation technique is utilised, both of which contribute to the process of pinpointing the location of the desease in MRI images. The model that was provided had 84% an accuracy, also delivers with good results without errors, and significantly reduces the amount of processing effort required. In addition to our research work, when we incorporate clinical data related to patients habits, life style, past health history and cognitive examinations,the accuracy may increase to more than 90%. This has been accomplished. Furthermore, it has been proved that acceptable results can be produced even if this type of data has been declared by the medical community may not be enough to confidently diagnose the disease. This is something that has been shown to be the case. Throughout the entirety of the paper, it has been demonstrated that this is doable with extensive data cleansing, as well as the selection and construction of the appropriate model.

The implementation of a deep neural network which is built on the selecting correct input data as well as the appropriate design of the parameters and architecture of the model. This will result in a precise discrimination between patients who have Alzheimer's disease and those who do not have the condition, with an accuracy of 82.61%.

We decided to focus on this issue due to the fact that there is now no treatment or cure for Alzheimer's disease. Recent research has shown that the amount of white matter (WM) and grey matter (GM) present in the human brain is a crucial factor in the development of AD. It is possible that the atropy of grey matter density and cortical thinning next to the cortex of the human brain could be considered one of the most critical features of Alzheimer's disease early diagnosis.

Compliance with Ethical Standards:

Conflict of Interest: B V D S Sekhar declares that he/she has no conflict of interest, Alok Kumar Jagadev declares that he/she has no conflict of interest.

Ethical approval: This article does not contain any studies with human participants or animals performed by any of the authors.

Sivaramakrishnan And Dr.M.Karnan “A Novel Based Approach For Extraction Of Brain Tumor In MRI Images Using Soft Computing Techniques,” International Journal Of Advanced Research In Computer And Communication Engineering, Vol. 2,Issue 4, April 2013.
Asra Aslam, Ekram Khan, M.M. Sufyan Beg, Improved Edge Detection Algorithm for Brain Tumor Segmentation, Procedia Computer Science, Volume 58,2015, Pp 430- 437, ISSN 1877- 0509.
B.Sathya and R.Manavalan, Image Segmentation by Clustering Methods: Performance Analysis, International Journal of Computer Applications (0975 – 8887) Volume 29– No.11, September 2011.
Devkota, B. & Alsadoon, Abeer & Prasad, P.W.C. & Singh, A.K. & Elchouemi, A.. (2018). Image Segmentation for Early Stage Brain Tumor Detection using the a Mathematical Morphological Reconstruction. Procedia Computer Science. 125. 115- 123. 10.1016/j.procs.2017.12.017.
K. Sudharani, T. C. Sarma and K. Satya Rasad, "Intelligent Brain Tumor lesion classification and identification from MRI images using k-NN technique," 2015 International Conference on Control, Instrumentation, Communication and Computational Technologies (ICCICCT),
Kumaracoil, 2015, pp. 777-780.
Kaur, Jaskirat & Agrawal, Sunil & Renu, Vig. (2012). A Comparative Analysis of Thresholding and Edge Detection Segmentation Techniques. International Journal ofComputerApplications.vol. 39.pp. 29-34. 10.5120/4898-7432.
Li, Shutao, JT-Y. Kwok, IW-H. Tsang and Yaonan Wang. "Fusing images with different focuses using support vector machines." IEEE Transactions on neural networks 15, no. 6 (2004): 1555-1561.
M. Kumar and K. K. Mehta, "A Texture based Tumor detection and automatic Segmentation using Seeded Region Growing Method," International Journal of Computer Technology and Applications, ISSN: 2229-6093, Vol. 2, Issue 4, PP. 855-859August 2011.
Mahmoud, Dalia & Mohamed, Eltaher. (2012). Brain Tumor Detection UsingArtificial Neural Networks. Journalof Science and Technology. 13. 31-39.
Marroquin J.L., Vemuri B.C., Botello S., Calderon F. (2002) An Accurate and EfficientBayesian Method for Automatic Segmentation of Brain MRI. In:A., SparrG., Nielsen M., Johansen P. (eds) Computer Vision — ECCV 2002. ECCV 2002.Lecture Notes in Computer Science, vol2353. Springer, Berlin, Heidelberg.
B.V.D.S. Sekhar, Pvgd Prasad Reddy, Gps Varma (2015), “Performance Of Secure And Robust Watermarking Using Evolutionary Computing Technique” JGIM, Vol 25, Issue 4, Article 5. October 2017, Pp 61-79.
B.V.D.S. Sekhar, Pvgd Prasad Reddy, Gps Varma (2015) ” Novel Technique Of Image Denoising Using Adaptive Haar Wavelet Transformation “Irecos , 2015, Vol 10, No 10, Pp 1012-1017.
B.V.D.S.Sekhar, Et Al (2018), “Image Denoising Using Wavelet Transform Based Flower Pollination Algorithm “, Vol 862, Advs In Intelligent Syst., Computing (Aisc), Springer.
B.V.D.S.Sekhar, Et Al (2019), “Image Denoising Using Novel Social Grouping Optimization Algorithm With Transform Domain Technique “, International Journal Of Natural Computing Research, Volume 8,Issue 4,Pp28-40.
Udayaraju P, Jeyanthi P. (2022) “Early Diagnosis of Age-Related Macular Degeneration (ARMD) Using Deep Learning”, Smart Innovation, Systems and Technologies, Vol.289, pp. 657–663, 2022.
B.V.D.S. Sekhar Et Al, (2022)” Artificial neural network-based secured communication strategy for vehicular ad hoc network”, Soft Computing, Springer, Vol: 27, Issue 1, ISSN:1432-7643,1433-7479, pp 297-309
B.V.D.S. Sekhar Et Al, (2022)” Novel Technique of Threshold Distance-Based Vehicle Tracking System for Woman Safety.”, Intelligent System Design, Lecture Notes in Networks and Systems 494, INDIA 2022,
B.V.D.S. Sekhar Et Al , (2023)” Sustainable and reliable healthcare automation and digitization using deep learning technologies”, Journal of Scientific & Industrial Research , Vol 23, Pp:226-231.

Download PDF

Journal Publication

published 17 May, 2023

Read the published version in Soft Computing →

Reviewers agreed at journal
19 Feb, 2023
Editor assigned by journal
15 Feb, 2023
First submitted to journal
10 Feb, 2023

You are reading this latest preprint version

Efficient Alzeihmer’s disease detection using Deep learning Technique

Status:

Journal Publication

Version 1

Abstract

Figures

Introduction

Literature Survey

Existing Method

Proposed Method

Results

Conclusion

Declarations

References

Status:

Journal Publication

Version 1