A Comprehensive Method for Identification of Stroke using Deep Learning
Surya.S1, B. Yamini2, T. Rajendran3*, K.E. Narayanan41Department of Computer Science and Engineering, Vel Tech RangarajanDr.Sagunthala R&D Institute of Science and Technology, Chennai, Tamil Nadu, India.
2Department of Computer Science and Engineering, SRM Institute of Science & Technology, Kattankulathur, Tamil Nadu, India.
3Department of Computer Science and Engineering, Saveetha School of Engineering, Saveetha Institute of Medical and Technical Sciences, Chennai, Tamil Nadu, India.
4Department of Computer Science and Engineering, RajalakshmiEngineering College, Chennai, Tamil Nadu, India.
1suryacse05@gmail.com, 2yamini.subagani@gmail.com, *3tlrajen@gmail.com, 4narayanake@gmail.com. Article History: Received: 10 January 2021; Revised: 12 February 2021; Accepted: 27 March 2021; Published online: 16 April 2021
ABSTRACT : Stroke has been among the top ten causes of death in this country over the last few years. Stroke is caused
mainly by the blockage of insufficient blood supply across the brain. Using deep learning algorithms, within a short duration time can be able to identify the stroke for the patients. This paper discusses identifying the stroke from CT or MRI and EEG signals using deep learning methods. The main method for stroke diagnosis is CT-Scan. The use of CT-Scan is very limited and also fairly costly for developing countries. MRI provides accurate diagnosis of stroke, but it is both time-consuming and unsuitable for 24/7 monitoring. Another device is potential to diagnose stroke using an EEG signal. A literature survey found some papers that use various types of methods using deep learning algorithms to identify the stroke.
Keywords—Stroke, CT, MRI, EEG, Deep Learning Algorithm
I. INTRODUCTION
This Stroke is the significant cause as well as a major disability cause in the developing world, affecting one in six people, with an approximate three to six million stroke cases annually. Cerebral vascular accident (CVA) is the most serious and dangerous cerebrovascular disorder and one of the major contributors to global death increase. Survivors of strokes have a greater of sudden disorders, and stroke being the main cause of adult epilepsy. It is categorized as an ischemic stroke (blockage of insufficient blood supply) or a hemorrhagic stroke (Blood vessel break). 80-85 percent of all strokes occur in ischemic stroke. Different trials and samples are considered for schematic arrangements. Hypertension and plastic contours are merged with local factor activities. All the different samples to calculate the weighted removed for MIR treatments and CT scan performances. The listed levels and parties generated the defects for stroke activities[10][11].
Recently, detection of stroke using devices including such computed tomography (CT) and magnetic resonance imaging (MRI) can enable a skilled radiologist to determine if a patient has suffered a stroke or not, but if a particular radiologist makes an improper judgment, this can cause the patient to miss the best treatment time. Improving the consistency of the diagnostic image is also a vital way of helping the physician make a formal diagnosis or identification of the image. Imaging techniques developed the magnetic visualization controls and computed fluid transfer tomography are commonly derived for the brain strokes. MRI variations are calculated with the performances of brain strokes and schematic arrangements. Equal elements and wave frequencies are followed by thermal scanner and CT scanner. Low cost process of scanning techniques developed the common mistakes and structures[10][11]. Density formation creates the linear motions and limitations. All the equal structures and arrangements taken from the results.Hypo density stroke creates the initial signs and processing signals. Initial stroke developed the MRI vision and diagnosed the images. CT scan results are proved the better performances as compared to MRI scan. Isometric lenses derived the better structure results and creates the schematic strokes.DWI system creates the earliest symptoms of energy loss conditions and controls[12][13].
II. LITERATUREREVIEW A. Stroke Based Identification Using 1D CNN:
EndangPurnamaGiri et al.(2016) In this paper the deep learning methodology is 1D CNN has proved to be the best model for separating EEG stroke data from EEG data control. In this research, applied in early
B. Stroke Prediction Using EMCD(ELectronic Medical Claims Database):
Chen-Ying Hung et al.(2017) achieved 92 per cent accuracy across both the GBDT and DNN algorithms while DNN needs less training of data. This innovative approach in the development of automated systems to predict the occurrence of strokes potentially provides many advantages, including quality of performance, high precision and fast prediction reporting. However, because such predictive algorithms can have several functional points, their responsiveness and precision can be adjusted to match the clinical needs. C. Stroke Detection Using EEG Signals:
Arooj Ahmed Qureshi et al.(2018)In this paper, an ischemic stroke detection approach from wearable EEG devices and machine learning through the multi-domain study of the EEG brain signals. Using 40 safe and 40 patient results, we find that Multi Layered Perceptron (MLP) and Bootstrap (Extra-Tree and Decision-Tree) models can achieve 95 per cent test accuracy with a region under the 0.85 ROC curve.
D. Stroke Detection Using Unsupervised Feature Perception:
Tyan, Yeu-Sheng et al.(2014) This paper proposes a method of computer aided tool controls and checking parameter levels. All the stroke identifications and scan techniques generated the coding concepts. CT scan performed areas are generated the better conceptual stages and ceramal structures. All the equivalent conditions and errors are identified successfully. It identified the radiology techniques and strategies to follow the suction rates and times. CT scan images are rectified the performances of thermal behavior structures and algorithms. Detecting images identified the stroke areas and increasing radiology percentages. All the percentages of increased levels to maintain the stroke behaviors and signals. All the waveforms are carried the potential signals and structures. Lowest frequency structures and shapes decided the imaging structures.
III. Methodology
Electroencephalographic (EEG) signals act as an essential information source for brain activity. Abnormal brain processing events may be identified using EEG signals. Most cerebral signals found in the scalp fall within the range of 1 − 20 Hz with EEGs. Waveforms are divided into frequency bands known as delta, theta (theta), alpha (α), and beta (β) to reflect much of the EEG used in clinical practice. Ischemic stroke is largely caused by changes in Vascular Blood circulation and can be identified by variations in EEG signal pattern. Prominent modifications include delta reduction (lowest frequency band) or high frequency band involvement (beta and alpha). Additionally, the power density proportion between the two hemisphere bands increases as a stroke affects one hemisphere. Use MRI scans in combination with meta-data such as history of patients, medical records and most notably MRI scans, the best results for Ischemic stroke detection so far are obtained. Magnetic Resonance Imaging (MRI) offers reliable stroke detection results, but it is a comprehensive tool, takes many hours to produce a summary report and is available for a limited period. MRI is often used in circumstances where there is no time pressure to provide treatment, usually as follow-up imaging. Magnetic resonance imaging is typicallypreferred as the tool for structural research in the brain, because it offers high soft tissue contrast images and strong spatial resolution, which poses no known medical risks. In contrast, Electroencephalography (EEG) provides a continuous, real-time, non-invasive brain function test derived the ischemic strokes. Due to variability in the blood vessels cerebral blood flow. This has been shown to be effective in identifying other brain-related behaviors such as Rapid Eye Movement (REM), Sleep and Wake Phase and other seizures[12][13].
Table.1CT Scan and MRI ScanComparison results
CT Scan MRI Scan
Uses Radiation No Radiation Donut Shape Tanning bed Shape Typically lasts 5 min Can last 30 min or more Best for seeing organs
and bony details
Fig 1. CT Scan Image
Fig 2. MRI Scan Image
Fig 3.Types of EEG Signal
A.Signs of Stroke:
Sudden face, arm or leg numbness or stiffness, particularly on one side of the body; confusion, trouble communicating or difficulty hearing speech; disturbed vision in either or both eyes; disturbed walking, dizziness, loss of coordination or lack of control and serious headache with no obvious cause.
B. Deep Learning Algorithms:
Deep learning algorithms referredthe multi-layer task of generated the neural networks.Thermal behavior properties taken from the specific input ranges (usually more than five). The most popular technique in machine learning methods are extracts from images a complex hierarchy of features as conflicting to the manual extraction of features in conventional deep learning algorithms because of their ability to learn themselves.MR scan training will be quantities the data, they achieve remarkable performance, and generalizability. The continuos rotating moment processed the MIR power has allowed state-of - the-art deep learning algorithms to evolve. This provided training with lots of images with deep learning techniques, and offered image variations reliability. Many trials and formations are created the MIR scan segments and controls. All the machining applications to generate the deep learning controls. CNN techniques worked in the field of micro scanning
Convolutional Neural Network derived the features of neural networks.It generated the code for run in CNN input stages and signal passed geometric detail, visual cortex field works which the individual cortical neurons belong to the Scanning field and does not transferred the CNN weak signals. CNN nodes are connected and do not represent all-to-all connections in a geometrical structure. Nodes are arranged in the image processing input layer to create spectrum ranges for scanning elements and shapes. All the equivalent structures and images (kernel) are derived. In the kernel signals transferred the image to MIR outputs. In CNN algorithms should not include computational methodology, the first step to remove human intervention in system selection or design.
Fig 4.Deep Learning Algorithms
C. Stroke Detection Flow Chart:
Stroke Detection Flow Chart is divided into some processing stages like pre-processing, extraction of brain tissue, significant area from extraction, and highlight area for strokes
Fig 5.Flow Chart for stroke detection
IV.COMPARATIVEANALYSISFORSTROKEDETECTIONUSINGDEEPLEARNING METHODS
S.N o
Authors Type of Input/Data
Methods Recognition Result
1 Chiun-li-Chin (2017)
CT or MRI image CNN >90%
2 Ming Sian,Lee (2012)
CT or MRI image Mathematic Morphology 85%
3 Chen-Ying-Hung (2017) EMCs DNN GBDT LR SVM 87.3 86.8 86.6 83.9 4 EndangPurnama (2016)
EEG and EOG 1D CNN F-Score(0.861) and
Precision(0.870)
5 Arouj Ahmed Qureshi(2018)
EEG MLP
And Boot Strap Models (Extra-true&
Decision-Tree)
95%
TABLE-IVCOMPARATIVE ANALYSIS FOR STROKE DETECTION USING DEEP LEARNING METHODS
V.ABBREVIATIONSANDACRONYMS
CT or MRI SCAN
RESULT IMAGE PREPROCESSING
BRAIN TISSUE EXTRACTION
MEANINGFUL AREA EXTRACTION
CVA Cerebral Vascular Accident CNN Convolutional Neural Network FLAIR Fluid-attenuated inversion recovery DWI Diffusion-weighted Imaging
REM Rapid Eye Movement
MLP Multi Layered Perceptron
VI.CONCLUSION
Stroke is the blood clot in the brain of blood vessels that can induce further brain injury, organ damage and sometimes even death. This paper discusses various conventional methods for the identification of stroke detection using CT scan or MRI Scan and signals. The feature extraction and classification methods which were used in many conventional stroke detection methods are discussed in this paper. This method can help radiologists accurately identify stroke regions solved the current affairs with local structures.Error levels and transferred rates are calculated. The experimental results of the conventional methods were also discussed in this paper.
REFERENCES
1. Giri, EndangPurnama, Mohamad Ivan Fanany, AniatiMurniArymurthy, and SastraKusumaWijaya. "Ischemic stroke identification based on EEG and EOG using ID convolutional neural network and batch normalization." In 2016 International Conference on Advanced Computer Science and Information Systems (ICACSIS), pp. 484-491.IEEE, 2016.
2. Hung, Chen-Ying, Wei-Chen Chen, Po-Tsun Lai, Ching-Heng Lin, and Chi-Chun Lee. "Comparing deep neural networks and other machine learning algorithms for stroke prediction in a large-scale population-based electronic medical claims database."In 2017 39th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), pp. 3110-3113.IEEE, 2017.
3. Qureshi, Arooj Ahmed, Canxiu Zhang, RongZheng, and Ahmed Elmeligi. "Ischemic stroke detection using EEG signals." In Proceedings of the 28th Annual International Conference on Computer Science and Software Engineering, pp. 301-308. IBM Corp., 2018.
4. Tyan, Yeu-Sheng, Ming-Chi Wu, Chiun-Li Chin, Yu-Liang Kuo, Ming-Sian Lee, and Hao-Yan Chang. "Ischemic stroke detection system with a computer-aided diagnostic ability using an unsupervised feature perception enhancement method."International journal of biomedical imaging 2014 (2014).
5. Praveen, G. B., Anita Agrawal, PonrajSundaram, and Sanjay Sardesai. "Ischemic stroke lesion segmentation using stacked sparse autoencoder."Computers in biology and medicine 99 (2018): 38-52. 6. Lee, Eun-Jae, Yong-Hwan Kim, Namkug Kim, and Dong-Wha Kang. "Deep into the brain: artificial
intelligence in stroke imaging." Journal of stroke 19, no. 3 (2017): 277.
7. Zhang, Rongzhao, Lei Zhao, Wutao Lou, Jill M. Abrigo, Vincent CT Mok, Winnie CW Chu, Defeng Wang, and Lin Shi. "Automatic segmentation of acute ischemic stroke from DWI using 3-D fully convolutional DenseNets."IEEE transactions on medical imaging 37, no. 9 (2018): 2149-2160
8. DouradoJr, Carlos MJM, SuanePires P. da Silva, Raul Victor M. da Nóbrega, Antonio Carlos da S. Barros, Pedro P. ReboucasFilho, and Victor Hugo C. de Albuquerque. "Deep learning IoT system for online stroke detection in skull computed tomography images." Computer Networks 152 (2019): 25-39.
9. Dr.T.Rajendran, O.Vignesh. Optimized Global Routing Techniques for Efficient VLSI Physical Layout Design . International Journal of Advanced Science and Technology, 29(06), (2020)., 2570 - 2577.
10. Pravin Kshirsagar, Nagaraj Balakrishnan & Arpit Deepak Yadav (2020) Modelling of optimised neural network for classification and prediction of benchmark datasets, Computer Methods in Biomechanics and Biomedical Engineering: Imaging & Visualization, 8:4, 426-435, DOI: 10.1080/21681163.2019.1711457. 11. Pravin Kshirsagar, Sudhir Akojwar & Nidhi Bajaj(2020),”A hybridised neural network and optimisation
algorithms for prediction and classification of neurological disorders”, International Journal of Biomedical Engineering and Technology Volume 28, Issue 4 ,DOI: 10.1504/IJBET.2018.095981
12. P. Kshirsagar and S. Akojwar, “Optimization of BPNN parameters using PSO for EEG signals,” ICCASP/ICMMD-2016. Advances in Intelligent Systems Research. Vol. 137, Pp. 385-394. © 2017- The authors. Published by Atlantis Press
13. P. Kshirsagar and S. Akojwar, "Classification & Detection of Neurological Disorders using ICA & AR as Feature Extractor", Int. J. Ser. Eng. Sci. IJSES, vol. 1, no. 1, Jan. 2015.
