A Comprehensive Fuzzy Ontology-Based Decision Support System for Alzheimer’s Disease Diagnosis

The World Health Organization (WHO) indicates that the proportion of the elderly will soon include nearly a quarter of the world population. Ensuring that health systems are prepared to deal with this phenomenal rate of aging and associated diseases generates many challenges. Among these challenges is facing Alzheimer’s Disease (AD) that may occur at some point in the elderly life and may harm societies. AD is considered a neurological, psychological, mental, and health setback. The Clinical Decision Support System (CDSS) can improve patient care and support many medical functions, such as diagnosing diseases that can reduce preventable harm. This research’s main objective is to design, implement, and evaluate the Alzheimer’s Disease Diagnosis Ontology (ADDO). It is a comprehensive semantic knowledge base toward the development of fuzzy ontology-based CDSS for AD diagnosis. ADDO can serve as a core component of CDSS, which provides representation, annotation, and access to aspects related to AD’s study and diagnosis. Toward the management of this objective, ADDO is based on the essentials of the Open Biomedical Ontology (OBO) and follows the Basic Formal Ontology (BFO) and the Ontology for General Medical Sciences (OGMS) principles. ADDO focuses on representing the patient characteristics, complications, drugs, diagnosis examination tests, and key aspects of their periodic visits in a standard way. The possibility of semantic interoperability is taken into account by integrating ADDO and a heterogeneous AD dataset. We used ADNI as a case study to mapping a set of real instances. To manage the medical domain’s uncertainty, ADDO is extended to fuzzy ontology to accommodate the medical linguistic variables and enhance diagnosis results’ efficiency. ADDO is constructed using Protégé 5.5.0 software and evaluated using the HermiT reasoner and SPARQL semantic queries. ADDO currently includes 7060 concepts, 99 properties, 46274 axioms, and 30708 annotations. As a result, ADDO is consistent and more reliable in managing most AD aspects than other existing AD ontologies.

Mild Cognitive Impairment (MCI) [4]. These patients have cognitive function problems greater than normal age-related changes related to language, organizing, planning, memory, and judgment. To attract attention, MCI is one of the main indicators that allow early detection of AD to improve patients' quality of life. The most noticeable sign of AD diagnosis is memory impairment, followed by cognitive test domains. Besides, psychological and behavioral disturbances (delusions, depression, sleep changes, paranoia, wandering, and anxiety). The cause of AD is incompletely defined, and no truly effective therapy exists.
In general, early disease diagnosis may allow intervention to identify the underlying disease processes and thus modify or halt disease progression. As for AD, it may represent a significant challenge for many reasons [5]. First, oftentimes, friends or family members notice signs of MCI even before the patient realizes they have a problem. Second, current diagnosis approaches as markers for brain atrophy is not always present at early diagnosis. Third, the accumulation of amyloid plaques deposition that appears in the brain probably begins 10-15 years before the first sign of clinical impairment appears, followed by intracellular neurofibrillary tangles. Fourth, the appearance of some signs related to the characteristics of MCI (i.e., losing things often, changes in social behavior, depression, neuroticism, etc.). It may have other causes, such as medication side effects and certain psychiatric disorders.
When a primary care physician suspects MCI or AD, the patient is shown to a group of specialized doctors for further evaluation [6]: Geriatricians are specialized in studying body changes with age to see if the patient's symptoms indicate a serious problem. Geriatric psychiatrists can assess memory and thinking problems. Neurologists perform brain scans of the patient and review the presence of abnormalities for the brain and central nervous system. Neuropsychologists specialize in performing tests of memory and thinking.
AD is a serious and complex disease. So an exact medical evaluation is needed to define complete clinical instructions for the patient to make the appropriate diagnosis [7]. It includes patient demographic data, patient history (disease history, medical history, family history of AD or dementia), complications, drugs, and diagnosis examination test (physical examination, gene, MRI image, laboratory results, behavioral, and cognitive tests) [8]. Neglecting one of these clinical guidelines may lead to an incorrect diagnosis. This leads to great pressure on doctors to deal with various diagnostic elements in light of dealing with a large number of patient cases. Besides, the chronic nature of AD requires long-term care systems to meet the needs of the patients. So, the Clinical Decision Support System (CDSS) development, besides current emerging computational intelligence methods, can bring new directions for pushing medical care to success, especially for these serious diseases as AD.
One of the most critical operations in the medical domain is data sharing. But without the possibility of semantic interoperability among enormous medical data resources, sharing data in a meaningful way is unrealistic. It requires establishing a standardized structure for terms and relationships and describes their processes. The way to provide standardization is through building ontologies. Recently, there are more trends towards developing ontology-based CDSS models for highly specialized domains such as medical. Where ontologies development [9] is the critical step for modeling experts' knowledge. It plays a major role in knowledge management by enhancing the intelligence, efficiency and preserving the semantic interoperability of the CDSS [10].
Despite the success of ontologies in maintaining semantic relationships between their concepts, CDSS based on a crisp ontology cannot deal effectively with uncertainty or vague knowledge. This type of knowledge may not appropriate to deal with the difficult nature of the medical domain. To overcome these problems, Bobillo and Straccia [11] comprised fuzzy logic with an ontology using current languages and resources. Fuzzy ontology is an effective technology that significantly increases CDSS decision-making accuracy in real-world application domains.
Developing good ontology from scratch takes a lot of significant time and effort of ontology's authors [12]. So, ontology reuse allows for saving effort, time and guarantees a consistent representation of a specific domain. The starting point for developing a standard ontology is using Basic Formal Ontology (BFO) and Ontology for General Medical Sciences (OGMS) [13] as upper-level ontologies then reuse existing medical ontologies or terminologies, such as Systematized Nomenclature of Medicine Clinical Terms (SNOMED CT) [14], which is a multilingual vocabulary used for coding of clinical content (more than 300,000 medical concepts represented by an individual number). It is divided into several hierarchies as clinical findings, body structure, and biological product. It increases patient care efficiency by supporting the exchange of clinical terms, providing consistent representation, eliminating confusion, and improving data analysis.
BFO [15] developed by Barry Smith. Its goal is to make ontologies in a specific domain like biomedical have equivalent views on fundamental concepts, such as universals and particulars, time and space, and substances and qualities. It adopts a realistic approach into two disjoint categories of continuant (independent and dependent continuants, attributes, and locations) and occurrent (processes and temporal regions). OGMS [16] symbolizes the clinical entities in the disease and medical practice domain. It is useful in interpreting and recognition of the disease to improve healthcare.
Electronic Health Records (EHR) play a great role in monitoring and managing chronic diseases [17]. CDSS needs several data inputs to be processed. When CDSS integrates with the EHR, it gives CDSS an advantage to utilizing health management in real-time. One of the AD data sources available is Alzheimer's Disease Neuroimaging Initiative (ADNI) [18]. It is a longitudinal study designed to collect clinical, genetic, imaging, and biomarkers of MCI. The contributions of this research are as follows: 1) Provide the structure of the proposed ontology-based fuzzy CDSS model and discuss its sub-modules. 2) Develop a standard ontology as a knowledge base for AD patients (ADDO). 3) Ensure the semantic interoperability by integration between ADDO and ADNI to enable patient mapping. 4) Implement and validate ADDO for accuracy using the Protégé 5.5.0 tool. 5) Improve the way that crisp ADDO is used to represent imprecise and linguistic information by extending ADDO into the fuzzy ontology. 6) Compare ADDO with existing ontology regarding the completeness reusability.
The rest of this paper is arranged in seven sections. Section 2 presents a review of related work that highlights the existing AD ontology. Section 3 introduces our approach as the ontology-based fuzzy CDSS, which supports physicians in their AD diagnosis decision-making. Section 4 includes the construction mechanism and ontology evaluation of the crisp ADDO, which increases the AD knowledge resources' satisfaction. Section 5 introduces the ADDO fuzzy extension. Section 6 demonstrates the fuzzy ADDO experiment and results. Finally, the conclusion and future work are presented in Section 7.

II. RELATED WORK
For generating accurate and acceptable CDSS, the knowledge base is the essential component that can semantically represent disease and structure knowledge. Handling this challenge requires three main issues. First, provide semantic consistency issues that require a detailed analysis of patients. Second, build a complete clinical knowledge base model in the lack of a comprehensive clinical approach. Third, support the integration with the EHR system, which allows for effective management in real-time.
Before developing ADDO, we surveyed the ontology repositories such as BioPortal [19] and recent researches for available ontologies related to AD. There are several existing ontologies of AD with different end goals categorized as storing and recovering AD information, supporting standardization, and suggesting the diagnosis of AD. Regarding to AD knowledge modeling, Semantic Web in Neuromedicine (SWAN) was developed by Ciccarese et al. [20] as a standard project related to significant practical research in AD. It focused on storing existing AD knowledge allowing for building publications, digital reference repository, and the semantic web. However, SWAN is not available on any online ontology repositories. For supporting AD research purposes, National Institute on Aging (NIA) published CADRO [21] ontology based on three categories. The first category is the molecular and physiological processes. The second category is the assessment, disease monitoring, and diagnosis. The last category is research and clinical interventions. However, CADRO was not operated in the realworld. For the purpose of supporting SemBiP semantic portal, it is a digital library specialized in reviewing AD scientific papers, Dramé et al. [22] built the OntoAD ontology based on reusing the available terminological resources. It is a bilingual (English-French) ontology for modeling AD knowledge. However, OntoAD has not been validated by using a real application. Henry et al. [23] tried to interest the pathophysiology of AD that can improve understanding of the AD and generate new hypotheses. They converted AlzPathway, a disease map that detailed AD pathophysiology, into a consistent ADMO ontology.
Jensen et al. [24] developed Neurological Disease (ND) Ontology-based on BFO. It specialized in representing the key aspects of neurodegenerative diseases, including symptoms, signs, diagnosis, interventions, and evaluations in the context of clinical practice. Later, Cox et al. [25] extended ND ontology and developed NeuroPsychological Testing Ontology (NPT) that includes new classes and annotation of a large group of neuropsychological tests. They provided more representation of the cognitive functions, but NPT suffered from complexity. Since it had more classes out the field in which this ontology is focused. Batrancourt et al. [26] used DOLCE as the primary basis and developed OntoNeu-roLOG ontology aimed to spot on the brain and its cognitive functions. It is considered one of the complete ontologies. However, its biggest disadvantage could be a bad impact on reasoning time.
Regarding ontologies as diagnosis support, it is used to aid physicians in the early disease diagnosis. It often includes facts or given axioms, ontology with semantic reasoner to infer possible diagnosis. One of the important knowledge in these ontologies related to the tests carried out to patients as the neurological, neuropsychological, radiological, genetic, and metabolomic tests.
Malhotra et al. [27] tried to cover the different aspects of the AD, including non-clinical, clinical, risk factors, diagnosis, and treatments and developed Alzheimer's Disease Ontology (ADO) based on BFO. ADO is a semantic, standardized, biological representation, and allowing retrieval and inference of the information. However, ADO has low utilization of existing ontologies and gained more attention to disease mechanisms. Sanchez et al. [28] extended MIND based on SNOMEDCT and SWAN ontologies to support AD research. For the detection of AD, MIND used semantic reasoning over 350 patients. It was challenging to be reused because it had a brief description. Besides, it was only limited to the diagnosis concept and ignored the different cognitive processes. Zhang et al. [29] proposed ontology-driven decision support for MCI diagnosis by detection of the thickness of the cortical cortex through magnetic. However, This ontology ignored necessary tests, such as neuropsychological tests, and Just focus on an MRI imaging approach.
Zekri et al. [30] proposed AlzFuzzyOnto as fuzzy AD diagnosis ontology. It used MIND as a basic ontology and tried to overcome imprecision and uncertainty in some significant terms and ontology concepts using the fuzzy logic. Alz-FuzzyOnto was developed using classes, which are tests, tests value, patient, doctor, diagnosis, enrollment, and follow-up. However, the validation of this ontology had not been carried out. To facilitate remote monitoring of cognitive impairment patients, Ivascu et al. [31] developed a multiagent ontology. This system requires automated devices/sensors and data privacy improvements.
According to the real-world scenario and ability to reuse, ADO and ADMO are the available ontologies developed according to BFO. ADMO spotlight the biological description by representing the complexity of AD pathophysiology. ADO cared more about the disease mechanism and ignored patient data. Finally, ADO is the most suitable ontology to be reused in developing our ontology ADDO. It contains classes like gene, brain region, diagnosis procedure classes that might be useful for importing into ADDO.
AD is a dangerous disease that causes death to a large proportion of the elderly in the US, as it is ranked third after heart disease and cancer [32]. Concerning the current AD ontology, there are limited studies that have been presented toward semantically intelligent knowledge. They do not not commensurate with the severity of this disease. Most of these ontologies terms, relations, and axioms are not clear, and many of them are not publicly available, such as CADRO, MIND, and AlzFuzzyOnto [33]. Other studies ignored the analysis of patients' characteristics like ADO. None of them provide complete clinical knowledge of AD. Most of them are crisp ontology that cannot deal effectively with uncertainty and relationships in real-medical knowledge. Their support for integration with the EHR system was weak. To avoid the aforementioned limitations, we proposed ADDO ontology to be a good core of our CDSS for AD diagnosis. Its a standard, comprehensive, effective in dealing with real-medical knowledge. It provides a complete clinical knowledge of AD. Besides, it supports interoperability with heterogeneous data.

III. THE PROPOSED FUZZY ONTOLOGY-BASED CDSS MODEL
In this study, an ontology-based fuzzy CDSS is presented to warn patients who are at high risk of having AD. It is able to classify NC, MCI, and AD patients based on comprehensive clinical characteristics. Fig. 2 presents the architecture of the established framework. It is divided into three main components. First, the ADDO construction mechanism aims to provide AD-related knowledge using ontological knowledge representation to support physicians' diagnosis decisionmaking. So, ADDO is the core component of an intelligent CDSS. According to the real-medical scenario, diagnosis tests have different linguistic values that were not possible using the crisp ontologies. Second, the ADDO is fuzzified based on the FuzzyOWL2 plug-in to enhance ADDO ontology for representing uncertain concepts and relationships in real-medical knowledge, thus facilitating decision-making efficiency results. Third, the rule base construction handles the generation of SWRL rules. An accurate rule base improves the accuracy and the interpretability of the CDSS.
Rules are extracted based on learning data mining techniques, such as Bayesian networks, neural networks, and fuzzy decision trees, with the domain expert's guide. A Fuzzy Decision Tree (FDT) is very suitable for the induction of simple medical decisions. It is applied in the medical decision by examining ambiguously and incompletely clinical data. We selected the FDT to keep interpretability, investigate medical error, robustness, and applicability of resulting rules. The rule set is trained by ADNI data of NC, MCI, and AD patients inducted by the FDT algorithm.
Full integrated ontology and rule-based reasoning include the following components. First, the database provides semantic interoperability for the mapping of patient's profile data, demographic data, the longitudinal progression of the patients visit, gene, MRI image features, symptoms, medical history, medical disease history, medication intake, diagnosis examination test including physical examination, chemical biomarker (laboratory results, behavioral, and cognitive tests), and complications. Second, the knowledge base comprises the fuzzy ADDO and rule-based model using SWRL. Third, fuzzification transforms the patient data as test values into degrees of the match with linguistic values in fuzzy ADDO ontology. Fourth, fuzzy inference uses some fuzzy if-then rules to map a given input to an output. Fifth, defuzzification transforms fuzzy results of inference to crisp output values. Sixth, the query engine handles physician's queries. Seventh, the reasoning engine checks the consistency and deduces diagnosis decisions.

IV. METHODOLOGY FOR ADDO DEVELOPMENT
CDSS is essential to improve healthcare practice and reduce preventable medical errors by offering the right, efficient, and effective decisions at the right time. CDSS, which is provided with consistent ontology and integrated with patient databases, are the main aspects of proactively. The most critical CDSS component is its knowledge base. This knowledge base can be represented and structured using ontologies. In which ontology plays a key role in semantic reasoning and making inferences about diagnoses using the collected patient data. Concerning semantic intelligence, ontologies can add more power to CDSS [34]. For example, to mimic doctors' reasoning, CDSS cannot avoid semantic relationships between some patient's features, such as diseases, symptoms, and drugs. CDSS with ontologies can handle this challenge because it can offer several advantages as they can be reused in similar domains, easy maintenance, support knowledge sharing, semantic integration with EHR, and other reasoning techniques. As a CDSS support, ontologies are designed for a practical purpose to be integrated as a submodule working as a knowledge database into a larger system. Building coherent, interoperable, consistent, and sharable ontology related to AD is a challenge. It requires a comprehensive clinical approach. The patient's clinical characteristics and data must be presented at the point of care, such as disease, complications, laboratory examination, physical examination, symptoms, medical history, medication, and genetics. It includes the ability to design strategies to diagnose AD, detect risk levels (NC, MCI, and AD), and monitor their severity according to the complete medical profile.
This section discusses the construction of ADDO ontology for AD diagnosis. Hopefully, ADDO introduces interesting features for classifying NC, MCI, and AD patients and will play a significant role in developing a standard, intelligent, and interoperable CDSS. We have imported top-level ontology via BFO and OGMS with the proposed ADDO to reduce the possibility of errors and support semantic interoperability between ADDO and other systems. Besides, preserve the reusability, consistency, and formality of the resulting knowledge. ADDO construction has five phases, including the preparation as a first and fundamental stage for acquiring information related to AD knowledge and determining its clinical aspects, and decides the appropriate way of developing the ADDO ontology. The second phase is ADDO initialization, where features are collected, and a description is provided about ontology classes, properties, and instances. In the third phase, BFO and OGMS merged into ADDO as the top-layer ontology. It provides a general top level structure that can support interoperability between existing systems and ADDO. Finally, implementation and evaluation of ADDO check the correctness and consistency. Fig. 3 illustrates the ADDO construction framework.

A. ADDO PREPARATION 1) SCOPE AND PURPOSE DETERMINATION
One of the first and essential steps in ontology construction is defined by its domain and scope. It is a significant step to ensure the validation of the specific domain before building an ontology. In this paper, our purpose is to develop a standard disease diagnosis of a coherent ontology associated with Alzheimer's patients that includes comprehensive information from different sources to improve reasoning accuracy. The physician or medical students are the intended users of the ADDO. ADDO makes decisions in the following manner. First, the ADDO receives the account of patient profile data, including diseases, diagnostic tests, family history, symptoms, medications, patient history, and physical examination. Then, the reasoner performs reasoning on ADDO's knowledge base, and the inferred diagnosis decision is added to the knowledge base.

2) REQUIREMENT SPECIFICATION
As mentioned, it is essential to identify some essential information to build this ontology. There are many medical key VOLUME 9, 2021 aspects related to AD that must cover, as shown in Fig. 4. ADDO represents the significant aspects of patient characteristics under the patient profile and his periodic visits data in a standard way. Each patient profile has patient characteristics and diagnosis. ADDO ontology provides a possible diagnosis for the patient subject to the responsibilities outlined in its chronic diseases and drugs.
where C is the patient characteristics, including: 1) Demographic data: participant ID, age at baseline, birth date, gender, participant education (years), marital status, ethnic category, and racial categories. 2) Vital sign: blood pressure value, height, respiration, temperature value, weight, and seated pulse rate. 3) Symptoms: early symptom of AD, symptom of MILD AD, symptom of moderate AD, and symptom of severe AD). 4) Pain: abdominal pain, acute pain, chest pain, headache, and earache. 5) Medical history: smoking history value, alcohol intake value, drug abuse, major surgical procedures, and allergies or drug sensitivities. 6) Family history: monitoring of patient relatives who have a family history of AD or dementia. 7) Patient Visit: longitudinal progression of the participant visit, such as visit code and visit date. 8) Complications: the likelihood of suffering from other complications, such as acute disease or chronic disease related to cancer, liver, diabetes, hepatic, cardiovascular, and kidney. 9) Medication: define drugs used in treatment as Donepezil and Axura. 10) Diagnostic tests: They categories results. The diagnostic tests are described by a huge number of features. Some of this feature is shown in Table 1.
where D(C) = AD diagnosis. The patient is described by these features, and the ADDO suggests a diagnosis of patient conditions.

3) REUSE OF ONTOLOGY
To build a standard and coherent ontology, we may exclude the model's development from scratch and support the adoption of some features from the existing ontology. It avoids some problems, including inconsistencies and redundancy problems in the class hierarchy, unstable references, and conflict in the term names. Most of the current ontology construction methodology comprises many terminologies (as SNOMED CT and RxNorm) and standard ontologies (like SYMP and DOID). It increases the quality of patient care by supporting the enrichment of the ontology, improving data analysis, and providing consistent representation. This step produces a mid-level ontology to analyze symptoms, drugs, disease, and other AD concepts. We build symptom classes based on Symptom Ontology (SYMP) [35]. Time classes are based on TIME ontology [36]. Disease classes are based on Human Disease Ontology (DOID) [37]. Adverse effect import from The Drug-Drug Interactions Ontology (DINTO) [38]. The adverse effect can be added to related patient history drugs, complications, and drugs. Drugs classes are based on RxNORM Ontology [39] to collect the most suitable drugs, contraindications, dosages, and other critical features. Finally, the gene classes, brain region, and diagnostic procedures are extracted from ADO [40]. Then connect the mid-level ontology with a top-level ontology to provide integration of medical information and ADDO.

B. INITIALIZATION 1) CLASSES' CONFIRMATION
To build a custom disease diagnosis ontology, the first step involves checking and collecting many medical conditions and features to make the decisions as complete as possible. These features were collected from medical experts, recent literature, and electronic medical record. For example, MCI patients have some symptoms, such as dizziness, memory problem, judgment of the time, accomplishment planning steps, or visual perception problems. It is essential to understand the nature and types of medications, as appropriate medication intake can reduce AD patients' complications. Maintain a timeline of patient visits and determine their severity as NC, MCI, or AD. Hence, classes as patients, relative medications, vital signs, lab tests, and medical history are utilized for ADDO to develop better diagnosis decisions for AD, as shown in Table 2. The illustration of ADDO architecture is shown in Fig. 5.

2) ADDO PROPERTIES
To represent domain knowledge, relying solely on the class hierarchy is not sufficient. It should also focus on the internal structure of these classes that represent relations. In the VOLUME 9, 2021 ontology domain, relations are represented by properties. They are divided into object and data properties. Object properties specify objects and define how links and associations are formed between ontology concepts. For example, the patients' visit can define the object property, has_Visit, which can be defined by P has_visit V = def. For every PatientProfile individuals p of P, there are some individuals Visit v of V such that p has_Visit v. Data properties connect individuals with literals, which are used to define the ontology concepts' values. For example, the Thyroid Stim Hormone (Test AXT117) for a specific patient is 6.26, and we may apply the data properties (has_value) to has_value (AXT117, 6.26). For further clarification, Table 3 provides ADDO object properties. The list of data properties of ADDO is shown in Table 4.

3) ADDO AXIOMS
To achieve precise semantics, we infer additional knowledge and support the computational search in ADDO. It is managed by a group of axioms that formulate the logical definitions of its classes and properties. Some of these axioms are explained as follow: The  medicalHistory (related to allergiesor, drug_sensitivities, smoking, alcoholIntake, and previous surgical), gene, complication, and at least one patient visit. This class is designated as follows: One of the critical classes for ADDO is the disease class whose characteristics specifies (i.e., code, contradicted drugs, and recommended drug). This class is implemented as follows: The neurological and physiological symptom class is used to model patient's symptom (i.e. onset data, chronicity, severity, and continue). We imported the possible AD symptoms from SYMP ontology and implemented it as follows:

4) ADDO INSTANCES
An ontology represents data in the form of triples (subjectpredicate-object) and is connected in the form of a directed labeled graph (link-by-link). Mapping data to ontology (e.g., RDF, OWL) has several advantages [41], such as data becomes self-describing when semantics are added, facilitate data integration, and support data source discovery and sharing. ADDO is a general ontology and can be applied with any data sources related to AD diagnosis. This paper used data as a case study that originated from the ADNI (adni.loni.usc.edu). It was initiated in 2003 to pose challenges to increasing public awareness of AD and other dementias. It provides a multi-categorical data nature, including MRI image data, PET image data, biospecimen data, clinical data, and genetics data. ADNI data are made available in spreadsheets (CSV).
In this work, we employed Cellfile [42], which is the Protégé-based plugin to perform a comprehensive mapping of the ADNI spreadsheet content to OWL entities and axioms. Cellfile provides an automated process to generate the mapping process. It converts every CSV table into a class and each table column into a data property. This approach did not yield satisfactory results. It increases the number of data properties and decreases the number of object properties. It makes ontology shallowness and related only to ADNI structure.
For example, ADNI contains an ADSXLIST.csv file that describes the patient symptoms checklist. It contains many features like RID (Participant roster ID), VISCODE, USER-DATE, EXAMDATE, and 28 symptoms only related to AD as AXNAUSEA(Nausea), AXVOMIT(Vomiting), AXDI-ARRH(Diarrhea), etc. Each symptom feature has an Absent or Present value. For mapping this table in a traditional way using Cellfile. We have patient symptoms restricted with 28 data properties only. That is further from reality. Since the patient symptoms should be more comprehensive and not defined by a limited number of data properties, there are certainly cases of patients who may suffer from other symptoms.
To avoid the previous problem, ADDO has comprehensively built that supports interoperability in the following manner. First, we import symptom classes from standard SYMP ontology and place them according to BFO and OGMS principles. Then, we add an object property called has_symptom (Domain ''PatientVisit'', Range ''symptom''). For every PatientVisit individuals v of V, there are some individuals symptom s of S such that v has_symptom s. Then, we add many object properties (Domain ''Symptom'', Range ''primitive Type'') to describe symptom data as SYMP.CEASE (Cease Date), SYMP.CHRON (Chronicity), SYMP.CONTD (Is the event ongoing?), SYMP.ONSET (Onset Date), SYMP.SEVER (Severity). The primitive type has data property as hasValue, has_severit, where is the patient symptom value added (connect individuals with literals). So, data are mapped into more closely mimic reality categories. It makes the mapping of ADNI or other related data to OWL classes more straightforward and yields better results.

C. MERGING TOP-LEVEL ONTOLOGY
To create a complete class hierarchy of ADDO, some classes are built from scratch, such as patient, patient profile, and patient visits. Other classes are imported from standard ontologies, such as DOID and SYMP. ADDO is fully developed as an extension of BFO and OGMS. They are ontological structures to support semantic interoperability by offering reuse top-level ontology terms between a significant number of domain-specific ontologies. OGMS is responsible for utilizing entities in a clinical domain. It recognizes the disease of healthcare. It contains diagnostic procedures, which consist of diseases and their causes and symptoms. It provides high-level terms as 'disease', 'disease course', 'diagnosis', and 'disorder'.
ADDO classes architecture follows the top-down approach and the is_a hierarchy, which is the backbone of every ontology. These classes were added under the most relevant concepts of BFO and OGMS. For example, the ADDO's 'disease' class is added as a subclass of BFO's disposition, and the ADDO's 'neurological and physiological symptom' class is implemented as a subclass of OGMS's symptom.
As shown in Fig. 6, ADDO is constructed by merging three levels of abstraction. The first level is the upper level that uses global ontologies as BFO and OGMS to create a standard ontology. The second level is the mid-level that contains the patient profile, patient visit, and time trying to build the AD diagnosis according to the specific patient profile. The third level is the ADDO subclasses, which are detailed parts for each of the presented mid-level.

D. ADDO IMPLEMENTATION
The ADDO implementation is done by using ontology editors. We used Protégé 5.5.0 editor to develop ADDO. OWL 2 language is utilized to express it. ADDO reuses other standard ontologies by importing the needed classes by using Onto-Fox [43]. It is a web-based ontology tool that supports ontology reuse by fetching ontology axioms and terms. ADNI data is uploaded to ADDO using the Cellfie plugin. Table 5 shows the metric data collected from Protégé.

E. ADDO EVALUATION
It is significant to ensure the correctness and quality of the ontology. We used HermiT (version 1.4.3.456) reasonser and SPARQL semantic queries in Protégé to check the consistency of ADDO ontology. The retrieving information process from ADDO is shown in Fig. 7.
SPARQL is a particularly powerful language used to query data corresponding to the RDF data model. A query about VOLUME 9, 2021 retrieving information from ADDO is carried out in the following manner. First, ADNI data, which are stored in CSV files, will be uploaded to ADDO using the Cellfie plugin. Second, the SPARQL query engine can access the knowledge base to retrieve information about the patient's profile. The ontology engineering is not able to express all relations. Ontology can be extended by adding SWRL rules. Moreover, using the SWRL rule reasoner provides the opportunity to infer new facts that can be added to the knowledge base. The next section is showing some questions and their corresponding SPARQL queries. Q1: Are there any patients with ages 50 to less than 70 years at their baseline visit, with mild to moderate AD, without the APOE4 allele as these would be good candidates for the clinical trial? ADAS COG helps in evaluating cognition that can assess which stage of AD person is going through. Its score is ranging from 0 to 70. Score 0 is the least impairment, while score 70 is the most impaired [45] filter((?RID=1018)&&((?fearture="WholeBrain")|| (?fearture="Hippocampus")||(?fearture="Entorhinal")|| (?fearture="Ventricles")||(?fearture="Fusiform")|| (?fearture="MidTemp")||(?fearture="ICV"))).} Magnetic resonance imaging (MRI) [48] is an essential component of the diagnosis of AD. MRI uses a magnetic field and radiofrequency pulses to collect data about brain abnormalities and decrease the volume in some areas of the brain, such as the Hippocampus, Entorhinal, Fusiform, MidTemp, Ventricles, and whole-brain volume. For example, the presence of distortions in the Hippocampus means an inability to form and retain new memories. Distortions in the Middle temporal area lead to recognizing known faces, contemplating distance, and accessing word meaning while reading. Problems with facial recognition are a result of fusiform distortions.
Q8: How many patients are contradicted with Axura? filter regex(?dlabel, "Axura", "i").} } Axura [49] drug used in patients with moderate to severe AD. Its active substance contains memantine hydrochloride. Axura should not be used in two cases. In the first case, the patient is allergic, contradict with memantine hydrochloride. In the second case, the patient is tacking drugs contradict with Axura drug. The sparql query gets a list of the Neuropsychological test associated with AD diagnosis.
Q11: List all patients with their diagnoses at baseline visits who have a history of smoking or alcohol Intake and suffering from a specific disease with the disorder of cardiac function or disorder of coronary artery?
In this section, many competency questions are implemented in the SPARQL query. We have carefully selected these questions to cover many important clinical aspects of diagnosing AD. The SPARQL query answers were evaluated by the domain experts.
As a result, ADDO is understood clearly by the domain expert. It covered all medical concepts of AD. Their classes and properties are defined consistently. Finally, ADDO is a complete and accurate clinical knowledge of AD diagnosis.

V. ADDO FUZZY EXTENSION
ADDO ontology is a crisp ontology in which the membership degrees of all properties are equal to 1. So, it is limited to classifying precise information about AD. In contrast, most patient data is ambiguous and contains different ranges of linguistic values, especially diagnostic tests. For example, we reviewed some SPARQL queries in the previous section, such as Q2 required patients with low ADAS COG scores, and we used a crisp value of less than 15. Besides, Q12 required AD patients with moderate MMSE COG scores, and we used a crisp interval value [21], [26].
To avoid these problems, we have to extend ADDO to fuzzy ontology. Hopefully, our proposed model copes with vague values. It tries to solve inconsistency. It accommodates the linguistic variables that assist the fuzzy query formulation and enhance diagnosis results' efficiency.
In the fuzzy ontology domain, fuzzification is implemented using the fuzzy OWL 2 protégé's plugin in which Fuzzy aspects are formulated as annotations in the crisp ontology [51]. There are some requirements that must be taken into the crisp ontology, such as fuzzy logic annotation. ADDO uses Fuzzy logic to perform approximate reasoning. Fig. 8 shows a snapshot of the ADDO fuzzy logic annotation. Fuzzy data types are created for each of the ADDO Fuzzy terms. Fuzzy modifiers allow using real-world linguistic, such as very and slightly, in which fuzzy data types can get degrees of membership. Concepts are fuzzified, where the instance belongs to a certain degree. Fuzzy role assertion connects instances at a membership degree.
Instance belongs to a Fuzzy concept in a certain membership degree. For example, 'moderate_education' (education_ level) is a fuzzy concept, where 'moderate' is a linguistic term. Therefore, 'moderate_education' is a fuzzy concept, such as 'RID_139 is an instance moderate_education at 0.9 membership degree'. An object property connects instances at a membership degree as 'RID_1200 hasADASTest serve at a degree 0.7'. The property 'hasADASTest' connects instances' RID_1200' and 'serve' at a degree of 0.7. Data property specifies a literal value to individuals at a certain degree, such as 'RID_3_diagnosis_value has_value MCI at 0.8 membership degree'. Here, 'has_value' is a data property that connects instances' RID_3_diagnosis_value' and 'MCI'.

A. DEFINITION OF FUZZY SETS
ADDO contains many diagnostic test categories, which are numerical features, such as cognitive tests, brain imaging parameters, and blood tests. For each of the numerical features, three Fuzzy types are defined, which are an abstract role, Fuzzy data type, and Fuzzy concrete role. We used the Fuzzy OWL plugin to define a Fuzzy data type for each Fuzzy value. This plugin contains four Fuzzy datatypes, such as leftshoulder, rightshoulder, trapezoidal, and triangular with k 1 , k 2 , a, b, c, and d parameters.
For instance, Mini-Mental State Examination (MMSE) [52] is useful for cognitive evaluation of the diagnosis and longitudinal assessment of AD. It is a 30-point questionnaire. It consists of various categories, like an orientation to time or place, recall, and language. Normal scores can indicate 28 points or more out of 30. Any score of 9 or less indicates a severe. Fig. 9 shows Fuzzy data type representation for MMSE. Fuzzy types are defined as follows. To discover behavioral changes and assess the patient's mood, the original Alzheimer's Disease Assessment Scale-Cognitive Subscale (ADAS-Cog) [53] includes 11 tasks (70-point). It consists of some questions like following commands, naming fingers and objects, and a word recognition task. Normal scores can refer to 12 or less, for more than 20 point estimate AD.
Higher glucose levels [55] is a risk factor for AD. So, it's a critical blood test. ADDO contains five Fuzzy sets for blood glucose range from 0 to 300 milligrams per deciliter. Low scores equal to 45.0 or less, for more than 180 point estimate severe.
MATLAB is used to fuzzify all numerical features with the help of AD clinical guidelines. The designed Membership Functions (MFs) for each input variable of MMSE, ADAS-Cog11, blood sugar, and patient's risk factor are shown in Figs. 10, 11, and 12, respectively. For more Fuzzy sets and their MFs, parameters of different variables have been shown in Table 6.

B. DEFINITION OF FUZZY MODIFIERS
Certainly, using fuzzy modifier values, such as very, little, slightly, and recently, to describe the Fuzzy concepts can improve semantic queries in the ontology. For example, the Fuzzy ADDO could be more expressive by adding the Fuzzy modifier 'very,' making it easier to get an axiom.

C. ADDO INSTANCES (ABOX)
The ADDO instances have been collected from ADNI CSV files as a case study. We have created a Fuzzy database for the Fuzzy ADDO ontology and filled it with 30 cases of NC, MCI, and AD patients to be store in the same ontology structure. Fig. 13 shows a fragment of the AD patient instances in the Fuzzy data set related to brain region volumes, Cerebospinalfluid test, and cognitive test. The mapping VOLUME 9, 2021 FIGURE 13. A fragment of the AD Patient instances in the fuzzy dataset related to brain regions volumes, cerebospinalfluid test, and cognitive test. process between the Fuzzy database and ADDO is shown in Fig. 14. Each fuzzy object stored in our fuzzy database mapped to ADDO individual identifier and many properties (object property assertion and ADDO data property assertion) that implemented to ADDO by asserting some axioms. Data property assertion for the patient case illustrated in Fig. 14 is defined as follows: <fuzzyOwl2~fuzzyType="concept"> <Concept type="nominal" value="0.4" individual="patient_3_ADAS11_severe" /> </fuzzyOwl2> <fuzzyOwl2~fuzzyType="concept"> <Concept type="nominal" value="0.6" individual="patient_3_ADAS11_Moderate" /> </fuzzyOwl2>

D. QUERYING THE FUZZY ADDO
We designed many queries to ensure the efficiency of retrieving the instances. Q1: Extract the patient RID with very old age, where MMSE test scores are severe along with severe ADAS-Cog11 score? has_age only(Age.value only (hasValue value VeryOld )) and has_ MMSETest some(MMSET.value some (hasValue value MMSESevere)) and has_ ADAS-Cog11~some(ADAS-Cog11. value some (hasValue value ADAS-Cog11Severe)) and Patient Q2: Find the drugs for an MCI patient whose HIV test is negative and whose blood sugar is very high?
Has_BloodSugar some(BloodSugar.value some (hasValue value BloodSugar veryHigh)) and Has_HIV some(HIV.value some(hasValue value HIVpositive)) and medication The Human Immunodeficiency Viruses (HIV) [56] may increase neurodegenerative disorders, such as AD. It is done due to neuronal damage from toxic viral products. The ontology used negative and BloodSugarVHigh as linguistic terms of the Fuzzy variable HIV test and blood sugar and extract the needed information.

VI. EXPERIMENT AND RESULTS
We introduced a methodology to develop Fuzzy semantic knowledge for decision-making. It is expected to be good practice in AD diagnosis to solve the linguistic variables and reasoning problems. Protégé was used to build semantic knowledge for decision-making. Fuzzy ADDO classes, objects, fuzzy data types, and data properties are shown in Fig.15. The extended Fuzzy ADDO uses OWL2 and Fuzzy annotation properties. The FuzzyOWL2 plug-in enables defining fuzzy elements to the ADDO ontology and uses fuzzyDL to reason query. The fuzzy ADDO evaluation process is based on two phases: ontology evaluation and a comparison between ADDO and other existing AD ontologies.

A. FUZZY ADDO EVALUATION
ADDO evaluation is an important and essential final step to measure the ontology's performance and define the instances that might not be identical. In general, ontology evaluation is done based on two steps. The first one is running reasoners, such as Pellet and Hermit. The second step is the query execution in questions that required answers, such as the SPARQL and DL queries, to extract the required individuals or instances. The Hermit reasoner was used to evaluate ADDO.

B. COMPARISON WITH EXISTING AD ONTOLOGIES
Regarding existing AD studies, it is a lack of building a complete ontology model for AD patients as many of these systems rely on a limited number of concept categories in their knowledge as Ontology-Driven Decision, ADMO, OntoAD. Also, some ontologies have not been encoded with standard medical terminologies, not recognized as top-level ontology. ADO, MIND, and AlzFuzzyOnto have not supported the reusing of already existed ontologies. Furthermore, some ontologies' experimentation and instantiation have not been carried out, like CADRO and AlzFuzzyOnto. For publicly available, such as ontology SWAN, Multiagent, AlzFuzzy-Onto, and Ontology-Driven Decision, are not publicly available. Such ontology does not provide a real and efficient AD diagnosis. They do not incorporate the other chronic diseases and symptoms of age-related changes besides AD that may affect the doctors' judgment.
Our proposed ADDO is fully developed as an extension of BFO and OGMS. ADDO supports AD clinical diagnosis and offers their risk level (NC, MCI, AD) based on representing the critical aspects of patients including disease, complications, medical history, physical examination, drug, symptoms, lab examination, MRI, and genes are considered. Finally, ADDO is extended to a fuzzy ontology, which makes it able to pass the difficulties of dealing with vague and medical linguistic terms. Table 7 shows A comparison between ADDO and other AD ontologies.

VII. CONCLUSION
AD is characterized as a chronic degenerative disease that involves a group of neurological disorders resulting from the accumulation of amyloid plaques that appear in the brain, affecting essential body functions. In this study, ADDO is developed as a standard fuzzy ontology-based semantic knowledge that aims to provide a warning to high-risk patients who have a high chance of having AD. A detailed analysis of patients and a timeline of patient visits is efficiently considered, including patient's demographic data, medical history, disease history, complications, medication, and covers many diagnostic tests. ADDO supports the interoperability by adhering to BFO and OGMS top-level ontologies. Hopefully, ADDO has greater significance in the AD clinical environment. From the experimental results, ADDO provides a standard ontology and supports interoperability by integrating ADDO and heterogeneous AD data. We used ADNI to mapping a set of real instances. ADDO is evaluated by answering many SPARQL semantic queries. As an evaluation result, ADDO is consistent and reliable. In the future, ADDO will evolve to include rule-based implementation using SWRL rules to build rule-based reasoning for AD diagnosis. With the deterioration of the cases of Alzheimer's patients and their confinement to bed with the progress of the disease. Many cases remain with their relatives at home. We expect to make many additions as provides remote healthcare to AD patients with the help of their caregivers to monitor the progression of patients' disease. To help physicians automatically retrieve patient data required for diagnoses and improve accuracy hospitalized not only patients but also remote available patients.