Table 1.
Main articles using AI to improve migraine diagnosis and classification.
| Reference (First Author, Year) |
Type of AI | Main Results | Number of Patients/Data Analysed |
|---|---|---|---|
| Chiang, 2022 [36] | AI-ECG algorithm able to predict the risk of subclinical AF using a convolutional neural network | MwA is associated with an increased risk of subclinical AF | 40,002 patients (17,840 MwA and 22,162 MwoA) |
| Cowan, 2022 [29] | CDE: a decision tree designed to ask questions to diagnose ICHD-3 primary headaches and several secondary ones | A positive CDE result helps rule in migraine or probable migraine diagnoses; a negative result helps rule them out | 202 patients |
| Frid, 2019 [40] | Predictive (classification methods and attribute-selection techniques) and traditional explanatory (statistical) analyses on functional connectivity measures | Functional connectivity metrics of resting-state EEG can be considered a biomarker to differentiate MwA from MwoA; MwoA patients show higher connectivity in the theta band | 52 patients (30 MwA and 22 MwoA) |
| Gálvez-Goicurla, 2022 [42] | Pain episodes clustered and then classified by unsupervised and supervised ML models | Migraine pain types are classified as high- or medium-intense pain episodes, sudden pain episodes, long-lasting pain episodes, and mild- or low-intense pain episodes. | 344 migraine attack data from 51 patients |
| Hsiao, 2022 [41] | SVM algorithms to establish the classification model | Functional connectivity of resting neuromagnetic activity may identify CM; discriminative features may be found from the interactions among salience, sensorimotor, and default mode networks; the classification model exhibited excellent performance in differentiating CM from HC and high performance in distinguishing CM from EM and FM | 240 subjects (70 HS, 100 CM, 35 EM, and 35 FM); data from 56 HS and 80 CM were included in the training dataset, while those of 14 HS, 20 CM, 35 EM, and 35 FM were included in the testing datasets |
| Katsuki, 2023 [26] | AI-based headache diagnosis model | AI model demonstrated high diagnostic performance for migraine | 6058 patients (4829 with migraine, 834 with TTH, 78 with TACs, 38 with other primary headache disorders, and 279 with other headaches) (4240/6058 training and 1818/6058 test datasets) |
| Katsuki, 2023 [27] | AI-based headache diagnosis model | AI model improved the non-specialist diagnostic performance | 4000 headache patients diagnosed by a specialist (2800 training and 1200 test datasets) |
| Kwon, 2020 [31] | Stacked classifier model with four layers of binary XGBoost classifiers (1. migraine vs. non-migraine; 2. TTH vs. non-TTH; 3. TAC vs. non-TAC (i.e., epicranial headaches and TCH); and 4. epicranial headaches vs. TCH) | Excellent performance of the ML approach, but good accuracy just for migraine | 2162 patients who visited the headache clinic (1286 training and 876 test datasets) |
| Liu, 2022 [30] | Decision tree, random forest, gradient boosting algorithm, and SVM models used to build a discriminant model and a confusion matrix used to calculate the evaluation indicators of the models | Applying ML to the decision-making system for primary headaches improves diagnostic accuracy; nausea/vomiting and photophobia/phonophobia are identified as the most important factors for distinguishing migraine from TTH | 173 patients (84 with migraine and 89 with TTH) |
| Perez-Benito, 2019 [37] | Subgrouping based on ML algorithms: nearest neighbours’ algorithm, multisource variability assessment, and random forest model | Based on pain intensity, one group of patients was younger, with lower joint positioning sense error in cervical rotation, greater cervical mobility in rotation and flexion, lower flexion-rotation test scores, positive PAIVMs reproducing migraine, normal PPTs over the tibialis anterior, a shorter migraine history, and lower cranio-vertebral angles while standing than the remaining subgroups. | 67 women affected by migraine |
| Sanchez-Sanchez, 2021 [39] | Supervised learning technique based on an artificial neural network | Artificial neural networks can achieve high precision and accuracy in migraine classification | 400 medical records of users diagnosed with pathologies associated with migraine |
| Sasaki, 2023 [34] | AI-based model using 17 objective items from questionnaires and predicted migraine or non-migraine diagnosis | AI model exhibited high diagnostic performance for paediatric and adolescent migraine | 909 questionnaire sheets (636 training and 273 test datasets) |
| Simic, 2021 [28] | Various mathematical, statistical, and artificial intelligence techniques, including decision-making methodology and clustering methods | The optimal number of clusters is three, representing three classes of headaches: (i) migraine, (ii) TTH, and (iii) other primary headaches; good quality of the system | 1022 subjects |
Abbreviations: AI = artificial intelligence; CDE = Computer-based Diagnostic Engine; CM = chronic migraine; EM = episodic migraine; FM = fibromyalgia; HS = healthy subject; ML = machine learning; MwA = migraine with aura; MwoA = migraine without aura; PAIVMs: passive accessory intervertebral movements; PPTs: pressure pain thresholds; SVM = support vector machine; TAC = trigeminal autonomic cephalalgia; TCH = thunderclap headache; and TTH = tension-type headache.