Classification Of Ischemic Stroke Using Machine Learning Algorithms

International Journal of Computer Applications (0975 – 8887)Volume 149 – No.10, September 2016Classification of Ischemic Stroke using MachineLearning AlgorithmsSelma Yahiya AdamAdil YousifMohammed Bakri BashirFaculty of MathematicalSciencesUniversity of KhartoumFaculty of Computer ScienceUniversity of Science&TechnologyFaculty of Computer Science&TechnologyShendi UniversityABSTRACTStroke is the third leading cause of death following diseases ofheart and cancer. The majority of strokes are classified asischemic which have two types: thrombotic and embolic. Inthrombotic stroke, the blood clot (thrombus) forms in one ofthe arteries that supplies blood to brain. The embolic strokehappens when a blood clot that forms somewhere else in thebody (embolus) break loose and travels to the brain throughthe bloodstream. Hemorrhagic stroke is considered anothertype of brain stroke by some researchers as it happen when anartery in the brain leaks blood or ruptures. As a reason ofhemorrhagic stroke, the brain cells damages as result of thepressure from the leaked blood. There are many similaritiesbetween these types and it is difficult to classify the casesaccurately using medical procedures. Furthermore, there areno clear boundaries between these types. This paper reviewedand analyzed the current studies on classification of ischemicstroke. Furthermore, the study has developed a classificationmodel for ischemic stroke using decision tree algorithm and knearest neighbor. The classification model is based on adataset of 400 cases collected from different Sudanesehospitals. The results of the decision tree algorithm can beused by medical specialist to classify and diagnose ischemicstroke patients. Moreover, the study revealed that somefeatures can be used directly to determine the type of ischemicstroke. These results help the medical doctors in theclassification process of ischemic strokes. Furthermore, theresults found that most of the ischemic stroke cases in Sudanare thrombotic ischemic stroke.KeywordsIschemic Stroke, Machine Learning, Decision Tree, KNN1. INTRODUCTIONStroke is a blood clot or bleed in the brain which can makepermanent damage that has an effect on mobility, cognition,sight or communication. Stroke is considered as medicalurgent situation and can cause long-term neurologicaldamage, complications and often death [1, 2]. Stroke is thethird leading cause of death following diseases of heart andcancer. The majority of strokes are classified as ischemicwhich have two types, thrombotic and embolic. In thromboticstroke, the blood clot (thrombus) forms in one of the arteriesthat supplies blood to brain. An embolic stroke happens whena blood clot forms away from the patient brain usually in thepatient heart and travels through the patient bloodstream tolodge in narrower brain arteries. Hemorrhagic stroke isconsidered another type of brain stroke as it happen when anartery in the brain leaks blood or ruptures. As a reason ofhemorrhagic stroke, the brain cells damages as result of thepressure from the leaked blood. There are many similaritiesbetween these types and it is difficult to classify the casesaccurately using medical procedures. Furthermore, there areno clear boundaries between these types. This paper reviewedand analyzed the current studies on classification of ischemicstroke. Furthermore, the study has developed a classificationmodel for ischemic stroke using decision tree algorithm and knearest neighbor. The classification model is based on adataset of 400 cases collected from different Sudanesehospitals.One of the main reasons for clot is the fatty deposits that makearteries and lead to a reduced blood flow or other arteryconditions. One of the main techniques that is used todiagnose the clot is the brain computed tomography scan, orbrain CT scan, which is a test that uses x rays to take clear,detailed pictures of the patient brain[3, 4]. CT scan is mainlydone immediately after stroke is suspected. A bleeding in thebrain or damage to the brain can be seen using brain CT scan.Other brain conditions that cause patients symptoms can bediscovered using brain CT scan. Magnetic Resonance Imaging(MRI) is the second test that is used to examine brain strokes.MRI is based on magnets and radio waves that are used toproduce pictures of the organs and structures in the patient’sbody. Any changes in brain tissue and damage to brain cellsfrom a stroke can be discovered using MRI test. To diagnose astroke MRI, CT or both can be used[5].This paper contains seven sections. Section two reviewsmachine learning algorithms. Section three describes theprocess of classification of ischemic stroke using machinelearning algorithms. The dataset used in the study is describedin section four. Section five describes the researchmethodology. Section six illustrates and discusses theexperimentation results. We concluded in section seven.2. MACHINE LEARNINGThe main objective of machine learning methods is to developcomputer software that can adapt and learn from theirexperience. Machine learning is a subfield of artificialintelligence that evolved from the learning process of patternrecognition and computational learning theory[6]. The studyis based on two main types of machine learning algorithmsdecision tree and k nearest neighbor (KNN).2.1 Decision treeDecision trees are data structure that has a root node, branchesand leaf nodes. Each internal node represents a test on anattribute or feature of the data, each branch represents theoutcome of a test, and each leaf node holds a class label. Therood of a tree is located at topmost. “A decision tree is a classdiscriminator that recursively partitions the training set untileach partition consists entirely or dominantly of examplesfrom the same class” [7]. Each branches and root node of thetree contains a split point which is a test on one or moreattributes or features and decides how the data is divided andsplit [7]. The Agrawal and Srikant in [7] described an26

International Journal of Computer Applications (0975 – 8887)Volume 149 – No.10, September 2016example that demonstrate how decision trees works as shownin Figure 1.the way the muscles in his mouth and throat move, making itdifficult for him to talk clearly, swallow or eat. Many strokesymptom and complications have been considered in thisresearch to classify the ischemic stroke accurately asdescribed in the following subsections.4. THE DATASETFig 1: Decision Trees for High and Low credit risk classesClassification[7]In Figure 1 we can see that (Age 25) and (Salary 50K) aretwo split points that divide the customers fields into high andlow credit risk classes. As in this example the decision treecan be employed to predict future applicants by classifyingthem into the high or low risk classes.2.2 K nearest neighbor (KNN)The k-Nearest Neighbors algorithm is a non-parametrictechnique utilized in classification and regression. When usingThe k-Nearest Neighbors algorithm for classification andregression, the input consists of the k closest trainingexamples in the feature space[8].The first step and contribution of this study is generating adata set for ischemic stroke disease. To our best knowledge,the dataset generated by this study is the first dataset forischemic disease in Sudan. The data set was generated usingstandard methods for generating benchmarked datasets asdescribed in [10-12]. The dataset items were collected fromseveral hospitals and medical centers in Sudan. The hospitalreport includes the patient number, age, sex, CT, MRIdiagnoses, and other variables for all patients hospitalized inthe hospitals participated in the study. The hospitalsparticipated in the study include Alzetona hospital, Antliahospital and Alneelen center. The data used in the datasetinclude the data of patient of cases from 2013 to 2015.The dataset contains 400 patients; their age is mainly between50 and 88 years. A few cases in the age of 33 years and mostof them are male.Table 1: The features and attributes considered in thestudy as suggested by specialists in ischemic stroke.3. CLASSIFICATION OF ISCHEMICSTROKE USING A12A13A14A15The majority of strokes are classified as ischemic which havetwo types, thrombotic and embolic. In thrombotic stroke, theblood clot (thrombus) forms in one of the arteries that suppliesblood to brain. An embolic stroke happens when a blood clotforms away from the patient brain usually in the patient heartand travels through the patient bloodstream to lodge innarrower brain arteries. Hemorrhagic stroke is consideredanother type of brain stroke as it happen when an artery in thebrain leaks blood or ruptures. As a reason of hemorrhagicstroke, the brain cells damages as result of the pressure fromthe leaked blood[9].There are many similarities between these types and it isdifficult to classify the cases accurately using medicalprocedures. Furthermore, there are no clear boundariesbetween these types.This study employed machine learning algorithms to classifythe ischemic strokes. Stroke symptoms that have been used asfeatures for machine learning’s classification process aredescribed in the following paragraphs. Brain hemorrhagefollowing an ischemic stroke is a severe difficulty oftreatment; yet, its pathology is poorly understood. Usingbrain imaging for classification may help to better understandand avoid causal factors. Temporary or permanent disabilitiesmay sometimes occur as results of strokes based on how longthe brain lacks blood flow and sometimes which part wasaffected. Other complications such as paralysis or loss ofmuscle movement may also occur.The patient may become paralyzed on one side of his body orexamine some difficulties in controlling some muscles, forinstance those on one side of the patient face or one of hisarms. The patient can use physical therapy which can help inreturning to activities affected by stroke paralysis, such aswalking, eating and dressing. Furthermore, difficulties intalking or swallowing are considered stroke complications aswell. A stroke patient may encounter to have less control overThe data that the feature containsThe patient numberthe age of patientthe sex of patientif patient have irritabilityif patient have convulsionsif patient have left-side weaknessif patient have right-side weaknesspatient have mouth deviationif patient have difficulty in speakingif patient have unable to walkif patient have headacheif patient have difficulty in seeingthe result of CT as abovethe result of mri as abovethe three classes {thrombotic,hemorrhagic embolic}Table 2: A sample of the datasetA A21A3A A A4 5 6A A7 8A A9 10A11175femaleF F TF TT F284maleT T FT TT TT367maleF F TF FF TT455maleT F TF FF TT27