||Author’s name, year, place of study
||The nature of the study
||Number of participants (N), software/applications used
||Key Elements / Outcome / Learning Points
||Almadhoun et al March 2021, Gaza, Palestine. Engineering Department
||unused subjects; Design an expert system
||Delphi programming language and videos
||Diagnose 11 headache problems
||11 questions to answer, and each question has multiple sub-questions describing several symptoms. Overlapping symptoms can cause misdiagnosis
|No training required before using this expert system
||Inability to diagnose other types of headaches not included in the system
|The accuracy, specificity, or sensitivity of the expert system has not been verified.
||Kwon et al 2020 Seoul, South Korea
||Number = 2162. Divided into two groups: Training = 1286, Test = 876.
||75 detailed examination questions were used
||Other, less prevalent types despite major primary headaches and secondary headaches (other than thunderclap headache causes) were excluded due to the long list of heterogeneous diseases they cause.
|Stacked classifier model used with 4 layers of XGBoost binary classifiers to distinguish: migraine, (tension headache) TTH, (trigeminal autonomic headache) TAC, epicranial headache, and sudden thunderclap headache.
||Stacked XGBoost Classifier Result: Accuracy: 81%
||Only 3 clinical features in each group were used to derive insight into headache types and the clinical symptoms used here differ from the ICHD-3 (International Classification of Headache Disorders) criteria.
|LASSO (Absolute Least Shrinkage and Selection Factor) used for each stacked workbook layer. Comparison of LASSO with SVM-RFE (vector machine recursive advantage removal) and mRMR (minimum max iteration of trapping).
||Sensitivity to: Migraine = 88%, TTH = 69%, TAC = 53%, Epicranial = 51%, Thunderclap = 51%.
||The clinical course cannot be understood from these 75 questionnaires, so an accurate diagnosis is difficult.
|The selected features used the XGBoost classifier that was compared with k-NN (k-nearest neighbour), SVM (support vector machine) and random forest.
||Specificity: Migraine = 95%, TTH = 55%, TAC = 46%, Epiranial = 48%, Thunderclap = 51%.
||Data from a single center.
|The performance report in the Migraine rating was excellent, the rest inferior. This study can be used as a prescreening.
||Traditional machine learning is used here. There is no point in deep learning.
||Krausek et al 2012. Wroclaw, Poland Mixed section: Technology, technical sciences and medicine
||A questionnaire filled out with the subjects that headache patients also included; ML algorithms are developed and tested; future study
||number = 579
||The best observed results with % accuracy: Random Forest = 79.97 ± 3.13,
||It includes only migraine, TTH, and other loosely defined headache types which included all remaining headaches whether primary or secondary.
|Age: 20 – 65 years
||Packing = 78.24 ± 2.98,
|Algorithms used: Naive Bayes (probabilistic classifier), C4.5 (based on “top-down induction of decision tree” (TDIDT), Support Vector Machine (SVM), Packing (or bootstrap aggregation), Boost, Random Forest.
||Boost 76.68 ± 2.43
|Filter selection algorithms used: consistency scaling filter, relief, genetic algorithm shell
||Julian et al , 2019. Study conducted in a hospital, emergency department. Objective: To detect possible secondary headache
||Number = 7972. Primary headache = 7098. Secondary headache = 874.
||Possible secondary headache: sensitivity = 89%, specificity = 73%, negative predictive value = 98.2%.
||Limited to emergency setting
|Records were processed using: Latent Semantic Analysis (LSA). Supporting Vector Machine (SVM) model used for training. Use the Python program.
||The best time in an emergency
||Emerging primary headache needs to be explored.
||Messina and others. , April 2020. Mila, Italy. Department of Neurosciences
||Opinion about machine learning in headache
||Celik et al 2009.
||Retrospective collection of records
||Artificial Immune System (Computational Artificial Intelligence)
||They were working on a headache classification project and they were creating a database from the neurology department of a private hospital
||Claims that the results will be published after the completion of the project.
||Details are unknown.
||According to Tezel et al. And the
||Topics not used. Designed and developed an artificial intelligence system
||Clonal selection algorithm (artificial immune approach)
||Work on a headache diagnosis. It included 250 different training set performances. 150 symptoms related to headache.
||Categorized into Migraine, TTH, and Constant Headache.
|On the basis of the principle of clonal selection
||Test set: correctly categorized symptom set: 96.74%
|Inspired by biological immunology.
||Incorrectly classified symptom group: 3.26%
||Katsuki and others , 2020. Department of Neurosurgery. Objective: For the automated diagnosis of primary headache
||Retrospectively check headache database and develop DL نظام
||number = 848
||The sample size is small.
|Age: 40-74 years old
||Classified into: Migraine, TTH, TAC, and other primary headache disorders.
||They conducted the study in one hospital.
|It uses a deep learning framework – one prediction.
||No external validation
|The use of an artificial neural network (ANN) with internal cross validation.
||There is no separation between chronic and recurrent episodic headache from >=15 days per month to >15 days per month for migraine or TTH headache.
|The model’s confusion matrix is also used
|Japanese language used with onomatopoeia and therefore uses Natural Japanese Language Processing (NLP)
||Keight et al. , . Department of Engineering, Medicine and Neurosurgery
||Retrospective headache data collection from two medical facilities
||number = 836
||Headache categorized into tension-type headache, chronic tension-type headache, migraine with aura, migraine without aura, trigeminal headache.
|The study was conducted in two medical centers in Turkey.
||Area under the curve (AUC): 0.985
|9 machine learning workbooks used in a supervised learning environment
||Sensitivity: 1 Specificity: 0.966
||Yin et al , 2015. China. Objective: To diagnose two types of headaches, probable migraine and probable migraine
||This comprehensive study worked on 3 steps. Obtaining data through clinical interviews, building a case library, and finally developing a case-based reasoning system
||Clinical decision support systems (CDSSs) are based on case-based reasoning (CBR).
||It can be a diagnostic tool for the general practitioner.
||Insufficient case library due to complex headache
|The K-Nearest Neighbor (KNN) method was implemented.
||The accuracy is very high in recognizing these two headaches.
||Needs a multicenter study and validation
|Number = 676 cases
||CBR has been used previously: (1) CASEY: to diagnose cardiac complications
|Likely migraine (PM) 56.95% TTH (PTTH) Likely: 43.05%
||(2) Decision-based support system for the diagnosis of COPD (COPD)
|Test group: N = 222. PM: 76.1%, PTTH: 23.9%
||(3) A hybrid case-based thinking approach to the diagnosis of breast cancer and thyroid disease.
||Qawasmeh et al 2020. Jordan
||Developed an ML-based system where prediction accuracy is checked by answering a web-based questionnaire
||n = 614 patient records. public hospital. males = 199; Female = 415. Different age group.
||Hybrid model (clustering and classification): K-median pool integrated with Random Forest . classifier
||Migraine with aura was excluded from this study as the difference could be stroke.
|A High Performance Headache Prediction Support System (HPSS) based on a hybrid machine learning model was used.
||Migraine prediction accuracy = 99.1%
|19 questions related to headache symptoms were used according to the ICHD-3 criteria.
||Overall accuracy = 93% (random set)
|26 classification algorithms were applied to 614 patients.
||HPSS claimed good positive feedback from patients, medical students, and physicians. Its easy to use interface saves time and effort.
||Woldeamanuel et al. , 2021. Headache and facial pain section. Stanford, California, USA
||A meta-analysis of 41 studies
||Total = 41 studies. Median age 43 years, 77% women. The mean sample size was 288.
||used case-based logic, DL, classifier set, ant colony, artificial immunity, random forest, black and white box combination, fuzzy mixed expert system
||60% of digital tools are based on ICHD standards.
|4 studies based on a questionnaire
||10 studies (25%) compared multiple ML programs
||12% of the instruments were evaluated in non-clinical centers
|Telephone interviews in two studies
||Diagnostic accuracy = 89%, sensitivity = 87%, specificity = 90%
|Face-to-face interview: 82% (strong advantage)
||There was no appropriate patient selection method in 39% of the included studies
|No description of age or sex ratio in 25 studies
||Sah et al 2017. Bhopal, India
||Database created from headache diaries and selection technique used for analysis
||Work on the classification of migraine headaches. Used: K-NN Data Mining Classifiers, Support Vector Machine (SVM), Random Forest, Naïve Bays.
||The best result was obtained from the Naïve Bays rating. AUC 0.475, precision 0.905
||Data was collected from headache diary
||Liu and others , 2022. Shanghai, China. medical school
||ML is used to identify primary headache. This is a cross-sectional study design.
||The logistic regression model was relatively better.
||Only two types of headache were made.
|Number = 173 patients (84: migraine, 89: TTH), information collected in neurology clinics using a questionnaire (19 questions)
||The logistic regression has an accuracy of 0.84 and an area under the receiver operating characteristic (ROC) curve of 0.90
||Mild headaches were not included in this study because they did not come for medical advice.
|Used: Decision Tree, Random Forest, Gradient Boosting Algorithm, Logistic Regression, Support Vector Machine (SVM) Algorithms
||Help distinguish between migraine and TTH and their distinct characteristics
||Small sample size
||Sanchez and others 2020. Colombia
||The study was designed to test the classification system to distinguish between migraine types
||It aims to classify migraines based on symptoms
||ANN delivered excellent results with 97.5% accuracy and 97% accuracy.
|N = 400 retrospective medical records
|Uses a set of 23 variables/questionnaires for symptoms or signs
|Implemented Artificial Neural Network (ANN), Logistic Regression Models, SVM, Nearest Neighbor, Decision Tree
||Celik et al 2017.
||A cross-sectional study to evaluate the accuracy of a classifier algorithm for diagnosing primary headache type using a web-based questionnaire
||It aims to diagnose primary headache based on an ant colony improvement algorithm.
||Classification accuracy = 96.9412%
||26 patients were misdiagnosed by classifying an ant colony
|It uses the web-based questionnaire system www.migbase.com. Used MySQL database and Hypertext Preprocessor PHP (PHP) programming language, 40 themes/questions included
||The accuracy for migraine, TTH, and cluster headache was 98.2%, 92.4%, and 98.2%, respectively.
||A similar study was conducted in Turkey using the same group of patients and the same questionnaire website but applied artificial immune algorithms for primary headache (2015) which had an accuracy of 99.6471% (using AIRS2-Parallel algorithm) 
|n = 850 headache patients from 3 cities who visited a neurologist
|Age group = 15 to 65 years.
|70% female and 30% male.