Automatic migraine classification via feature selection committee and machine learning techniques over imaging and questionnaire data

Machine Learning Methods to Estimate Individualized Treatment Effects for Use in Health Technology Assessment

BACKGROUND

Recent developments in causal inference and machine learning (ML) allow for the estimation of individualized treatment effects (ITEs), which reveal whether treatment effectiveness varies according to patients' observed covariates. ITEs can be used to stratify health policy decisions according to individual characteristics and potentially achieve greater population health. Little is known about the appropriateness of available ML methods for use in health technology assessment.

METHODS

In this scoping review, we evaluate ML methods available for estimating ITEs, aiming to help practitioners assess their suitability in health technology assessment. We present a taxonomy of ML approaches, categorized by key challenges in health technology assessment using observational data, including handling time-varying confounding and time-to event data and quantifying uncertainty.

RESULTS

We found a wide range of algorithms for simpler settings with baseline confounding and continuous or binary outcomes. Not many ML algorithms can handle time-varying or unobserved confounding, and at the time of writing, no ML algorithm was capable of estimating ITEs for time-to-event outcomes while accounting for time-varying confounding. Many of the ML algorithms that estimate ITEs in longitudinal settings do not formally quantify uncertainty around the point estimates.

LIMITATIONS

This scoping review may not cover all relevant ML methods and algorithms as they are continuously evolving.

CONCLUSIONS

Existing ML methods available for ITE estimation are limited in handling important challenges posed by observational data when used for cost-effectiveness analysis, such as time-to-event outcomes, time-varying and hidden confounding, or the need to estimate sampling uncertainty around the estimates.

IMPLICATIONS

ML methods are promising but need further development before they can be used to estimate ITEs for health technology assessments.

HIGHLIGHTS

Estimating individualized treatment effects (ITEs) using observational data and machine learning (ML) can support personalized treatment advice and help deliver more customized information on the effectiveness and cost-effectiveness of health technologies.ML methods for ITE estimation are mostly designed for handling confounding at baseline but not time-varying or unobserved confounding. The few models that account for time-varying confounding are designed for continuous or binary outcomes, not time-to-event outcomes.Not all ML methods for estimating ITEs can quantify the uncertainty of their predictions.Future work on developing ML that addresses the concerns summarized in this review is needed before these methods can be widely used in clinical and health technology assessment-like decision making.

Influence of next-generation artificial intelligence on headache research, diagnosis and treatment: the junior editorial board members' vision - part 1

Artificial intelligence (AI) is revolutionizing the field of biomedical research and treatment, leveraging machine learning (ML) and advanced algorithms to analyze extensive health and medical data more efficiently. In headache disorders, particularly migraine, AI has shown promising potential in various applications, such as understanding disease mechanisms and predicting patient responses to therapies. Implementing next-generation AI in headache research and treatment could transform the field by providing precision treatments and augmenting clinical practice, thereby improving patient and public health outcomes and reducing clinician workload. AI-powered tools, such as large language models, could facilitate automated clinical notes and faster identification of effective drug combinations in headache patients, reducing cognitive burdens and physician burnout. AI diagnostic models also could enhance diagnostic accuracy for non-headache specialists, making headache management more accessible in general medical practice. Furthermore, virtual health assistants, digital applications, and wearable devices are pivotal in migraine management, enabling symptom tracking, trigger identification, and preventive measures. AI tools also could offer stress management and pain relief solutions to headache patients through digital applications. However, considerations such as technology literacy, compatibility, privacy, and regulatory standards must be adequately addressed. Overall, AI-driven advancements in headache management hold significant potential for enhancing patient care, clinical practice and research, which should encourage the headache community to adopt AI innovations.

Migraine headache (MH) classification using machine learning methods with data augmentation

Migraine headache, a prevalent and intricate neurovascular disease, presents significant challenges in its clinical identification. Existing techniques that use subjective pain intensity measures are insufficiently accurate to make a reliable diagnosis. Even though headaches are a common condition with poor diagnostic specificity, they have a significant negative influence on the brain, body, and general human function. In this era of deeply intertwined health and technology, machine learning (ML) has emerged as a crucial force in transforming every aspect of healthcare, utilizing advanced facilities ML has shown groundbreaking achievements related to developing classification and automatic predictors. With this, deep learning models, in particular, have proven effective in solving complex problems spanning computer vision and data analytics. Consequently, the integration of ML in healthcare has become vital, especially in developing countries where limited medical resources and lack of awareness prevail, the urgent need to forecast and categorize migraines using artificial intelligence (AI) becomes even more crucial. By training these models on a publicly available dataset, with and without data augmentation. This study focuses on leveraging state-of-the-art ML algorithms, including support vector machine (SVM), K-nearest neighbors (KNN), random forest (RF), decision tree (DST), and deep neural networks (DNN), to predict and classify various types of migraines. The proposed models with data augmentations were trained to classify seven various types of migraine. The proposed models with data augmentations were trained to classify seven various types of migraine. The revealed results show that DNN, SVM, KNN, DST, and RF achieved an accuracy of 99.66%, 94.60%, 97.10%, 88.20%, and 98.50% respectively with data augmentation highlighting the transformative potential of AI in enhancing migraine diagnosis.

Brain connectome-based imaging markers for identifiable signature of migraine with and without aura

Background

Cortical spreading depression (CSD) has been considered the prominent theory for migraine with aura (MwA). However, it is also argued that CSD can exist in patients in a silent state, and not manifest as aura. Thus, the MwA classification based on aura may be questionable. This study aimed to capture whole-brain connectome-based imaging markers with identifiable signatures for MwA and migraine without aura (MwoA).

Methods

A total of 88 migraine patients (32 MwA) and 49 healthy controls (HC) underwent a diffusion tensor imaging and resting-state functional magnetic resonance imaging scan. The whole-brain structural connectivity (SC) and functional connectivity (FC) analysis was employed to extract imaging features. The extracted features were subjected to an all-relevant feature selection process within cross-validation loops to pinpoint attributes demonstrating substantial efficacy for patient categorization. Based on the identified features, the predictive ability of the random forest classifiers constructed with the 88 migraine patients' sample was tested using an independent sample of 32 migraine patients (eight MwA).

Results

Compared to MwoA and HC, MwA showed two reduced SC and six FC (five increased and one reduced) features [all P<0.01, after false discovery rate (FDR) correction], involving frontal areas, temporal areas, visual areas, amygdala, and thalamus. A total of four imaging features were significantly correlated with clinical rating scales in all patients (r=-0.38 to 0.47, P<0.01, after FDR correction). The predictive ability of the random forest classifiers achieved an accuracy of 78.1% in the external sample to identify MwA.

Conclusions

The whole-brain connectivity features in our results may serve as connectome-based imaging markers for MwA identification. The alterations of SC and FC strength provide possible evidence in further understanding the heterogeneity and mechanism of MwA which may help for patient-specific decision-making.

Searching for the Predictors of Response to BoNT-A in Migraine Using Machine Learning Approaches

OnabotulinumtoxinA (BonT-A) reduces migraine frequency in a considerable portion of patients with migraine. So far, predictive characteristics of response are lacking. Here, we applied machine learning (ML) algorithms to identify clinical characteristics able to predict treatment response. We collected demographic and clinical data of patients with chronic migraine (CM) or high-frequency episodic migraine (HFEM) treated with BoNT-A at our clinic in the last 5 years. Patients received BoNT-A according to the PREEMPT (Phase III Research Evaluating Migraine Prophylaxis Therapy) paradigm and were classified according to the monthly migraine days reduction in the 12 weeks after the fourth BoNT-A cycle, as compared to baseline. Data were used as input features to run ML algorithms. Of the 212 patients enrolled, 35 qualified as excellent responders to BoNT-A administration and 38 as nonresponders. None of the anamnestic characteristics were able to discriminate responders from nonresponders in the CM group. Nevertheless, a pattern of four features (age at onset of migraine, opioid use, anxiety subscore at the hospital anxiety and depression scale (HADS-a) and Migraine Disability Assessment (MIDAS) score correctly predicted response in HFEM. Our findings suggest that routine anamnestic features acquired in real-life settings cannot accurately predict BoNT-A response in migraine and call for a more complex modality of patient profiling.

Explainable AI (XAI) Applied in Machine Learning for Pain Modeling: A Review

Pain is a complex term that describes various sensations that create discomfort in various ways or types inside the human body. Generally, pain has consequences that range from mild to severe in different organs of the body and will depend on the way it is caused, which could be an injury, illness or medical procedures including testing, surgeries or therapies, etc. With recent advances in artificial-intelligence (AI) systems associated in biomedical and healthcare settings, the contiguity of physician, clinician and patient has shortened. AI, however, has more scope to interpret the pain associated in patients with various conditions by using any physiological or behavioral changes. Facial expressions are considered to give much information that relates with emotions and pain, so clinicians consider these changes with high importance for assessing pain. This has been achieved in recent times with different machine-learning and deep-learning models. To accentuate the future scope and importance of AI in medical field, this study reviews the explainable AI (XAI) as increased attention is given to an automatic assessment of pain. This review discusses how these approaches are applied for different pain types.

Quantitative prediction model for affinity of drug-target interactions based on molecular vibrations and overall system of ligand-receptor

Background

The study of drug–target interactions (DTIs) affinity plays an important role in safety assessment and pharmacology. Currently, quantitative structure–activity relationship (QSAR) and molecular docking (MD) are most common methods in research of DTIs affinity. However, they often built for a specific target or several targets, and most QSAR and MD methods were based either on structure of drug molecules or on structure of receptors with low accuracy and small scope of application. How to construct quantitative prediction models with high accuracy and wide applicability remains a challenge. To this end, this paper screened molecular descriptors based on molecular vibrations and took molecule-target as a whole system to construct prediction models with high accuracy-wide applicability based on dissociation constant (Kd) and concentration for 50% of maximal effect (EC50), and to provide reference for quantifying affinity of DTIs.

Results

After comprehensive comparison, the results showed that RF models are optimal models to analyze and predict DTIs affinity with coefficients of determination (R²) are all greater than 0.94. Compared to the quantitative models reported in literatures, the RF models developed in this paper have higher accuracy and wide applicability. In addition, E-state molecular descriptors associated with molecular vibrations and normalized Moreau-Broto autocorrelation (G3), Moran autocorrelation (G4), transition-distribution (G7) protein descriptors are of higher importance in the quantification of DTIs.

Conclusion

Through screening molecular descriptors based on molecular vibrations and taking molecule-target as whole system, we obtained optimal models based on RF with more accurate-widely applicable, which indicated that selection of molecular descriptors associated with molecular vibrations and the use of molecular-target as whole system are reliable methods for improving performance of models. It can provide reference for quantifying affinity of DTIs.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12859-021-04389-w.

Prioritizing Pain-Associated Targets with Machine Learning

While hundreds of genes have been associated with pain, much of the molecular mechanisms of pain remain unknown. As a result, current analgesics are limited to few clinically validated targets. Here, we trained a machine learning (ML) ensemble model to predict new targets for 17 categories of pain. The model utilizes features from transcriptomics, proteomics, and gene ontology to prioritize targets for modulating pain. We focused on identifying novel G-protein-coupled receptors (GPCRs), ion channels, and protein kinases because these proteins represent the most successful drug target families. The performance of the model to predict novel pain targets is 0.839 on average based on AUROC, while the predictions for arthritis had the highest accuracy (AUROC = 0.929). The model predicts hundreds of novel targets for pain; for example, GPR132 and GPR109B are highly ranked GPCRs for rheumatoid arthritis. Overall, gene-pain association predictions cluster into three groups that are enriched for cytokine, calcium, and GABA-related cell signaling pathways. These predictions can serve as a foundation for future experimental exploration to advance the development of safer and more effective analgesics.

A common model for the breathlessness experience across cardiorespiratory disease

Chronic breathlessness occurs across many different conditions, often independently of disease severity. Yet, despite being strongly linked to adverse outcomes, the consideration of chronic breathlessness as a stand-alone therapeutic target remains limited. Here we use data-driven techniques to identify and confirm the stability of underlying features (factors) driving breathlessness across different cardiorespiratory diseases. Questionnaire data on 182 participants with main diagnoses of asthma (21.4%), COPD (24.7%), heart failure (19.2%), idiopathic pulmonary fibrosis (18.7%), other interstitial lung disease (2.7%), and "other diagnoses" (13.2%) were entered into an exploratory factor analysis (EFA). Participants were stratified based on their EFA factor scores. We then examined model stability using 6-month follow-up data and established the most compact set of measures describing the breathlessness experience. In this dataset, we have identified four stable factors that underlie the experience of breathlessness. These factors were assigned the following descriptive labels: 1) body burden, 2) affect/mood, 3) breathing burden and 4) anger/frustration. Stratifying patients by their scores across the four factors revealed two groups corresponding to high and low burden. These two groups were not related to the primary disease diagnosis and remained stable after 6 months. In this work, we identified and confirmed the stability of underlying features of breathlessness. Previous work in this domain has been largely limited to single-diagnosis patient groups without subsequent re-testing of model stability. This work provides further evidence supporting disease independent approaches to assess breathlessness.

Personalized Body Constitution Inquiry Based on Machine Learning

Background

Body constitution (BC) is the abstract concept indicating the state of a person's health in Traditional Chinese Medicine (TCM). The doctor identifies the body constitution of the patient through inspection and inquiry. Previous research simulates doctors to identify BC types according to a patient's objective physical indicators. However, the lack of subjective feeling information can reduce the accuracy of the machine to imitate the doctor's diagnosis. The Constitution in Chinese Medicine Questionnaire (CCMQ) is used to collect subjective information but suffers from low acquisition efficiency.

Methods

This paper presents a personalized body constitution inquiry method based on a machine learning technique. It employs a random generator, a feature extractor, and a classifier to simulate the doctor inquiry and generate a personalized questionnaire. Specifically, the feature extractor evaluates and sorts the question of the constitution in the CCMQ based on the recognition results of the tongue coating image of patients. The sorted questions and relevant BC label are inputted into the classifier; the best questions are screened out for patients.

Results

The experimental results show that our method can select personalized questions from the CCMQ for the patients, significantly reducing the time and the number of questions to answer. It also improves the accuracy of recognizing BC. Compared with the CCMQ, patients had 68.3% fewer questions to answer and the time occupied by answering is reduced by 80.3%.

Conclusions

The proposed method can simulate the doctor's inquiry and pick out personalized questions for patients. It can act as auxiliary diagnosis tools to collect subjective patient feelings and help make further judgments on the patient's BC types.

Machine learning-based automated classification of headache disorders using patient-reported questionnaires

Classification of headache disorders is dependent on a subjective self-report from patients and its interpretation by physicians. We aimed to apply objective data-driven machine learning approaches to analyze patient-reported symptoms and test the feasibility of the automated classification of headache disorders. The self-report data of 2162 patients were analyzed. Headache disorders were merged into five major entities. The patients were divided into training (n = 1286) and test (n = 876) cohorts. We trained a stacked classifier model with four layers of XGBoost classifiers. The first layer classified between migraine and others, the second layer classified between tension-type headache (TTH) and others, and the third layer classified between trigeminal autonomic cephalalgia (TAC) and others, and the fourth layer classified between epicranial and thunderclap headaches. Each layer selected different features from the self-reports by using least absolute shrinkage and selection operator. In the test cohort, our stacked classifier obtained accuracy of 81%, sensitivity of 88%, 69%, 65%, 53%, and 51%, and specificity of 95%, 55%, 46%, 48%, and 51% for migraine, TTH, TAC, epicranial headache, and thunderclap headaches, respectively. We showed that a machine-learning based approach is applicable in analyzing patient-reported questionnaires. Our result could serve as a baseline for future studies in headache research.

Headache and Neuroimaging: Why We Continue to Do It

The appropriate imaging of patients with headache presents a number of important and vexing challenges for clinicians. Despite a number of guidelines and studies demonstrating a lack of cost-effectiveness, clinicians continue to image patients with chronic nonfocal headaches, and the trend toward imaging is increasing. The reasons are complex and include the fear of missing a clinically significant lesion and litigation, habitual and standard of care practices, lack of tort reform, regulatory penalties and potential impact on one's professional reputation, patient pressures, and financial motivation. Regulatory and legislative reforms are needed to encourage best practices without fear of professional sanctions when following the guidelines. The value of negative findings on imaging tests requires better understanding because they appear to provide some measure of societal value. Clinical decision support tools and machine intelligence may offer additional guidance and improve quality and cost-efficient management of this challenging patient population.

The visual system as target of non-invasive brain stimulation for migraine treatment: Current insights and future challenges

The visual network is crucially implicated in the pathophysiology of migraine. Several lines of evidence indicate that migraine is characterized by an altered visual cortex excitability both during and between attacks. Visual symptoms, the most common clinical manifestation of migraine aura, are likely the result of cortical spreading depression originating from the extrastriate area V3A. Photophobia, a clinical hallmark of migraine, is linked to an abnormal sensory processing of the thalamus which is converged with the non-image forming visual pathway. Finally, visual snow is an increasingly recognized persistent visual phenomenon in migraine, possibly caused by increased perception of subthreshold visual stimuli. Emerging research in non-invasive brain stimulation (NIBS) has vastly developed into a diversity of areas with promising potential. One of its clinical applications is the single-pulse transcranial magnetic stimulation (sTMS) applied over the occipital cortex which has been approved for treating migraine with aura, albeit limited evidence. Studies have also investigated other NIBS techniques, such as repetitive TMS (rTMS) and transcranial direct current stimulation (tDCS), for migraine prophylaxis but with conflicting results. As a dynamic brain disorder with widespread pathophysiology, targeting migraine with NIBS is challenging. Furthermore, unlike the motor cortex, evidence suggests that the visual cortex may be less plastic. Controversy exists as to whether the same fundamental principles of NIBS, based mainly on findings in the motor cortex, can be applied to the visual cortex. This review aims to explore existing literature surrounding NIBS studies on the visual system of migraine. We will first provide an overview highlighting the direct implication of the visual network in migraine. Next, we will focus on the rationale behind using NIBS for migraine treatment, including its effects on the visual cortex, and the shortcomings of currently available evidence. Finally, we propose a broader perspective of how novel approaches, the concept of brain networks and the integration of multimodal imaging with computational modeling, can help refine current NIBS methods, with the ultimate goal of optimizing a more individualized treatment for migraine.

Machine learning approach to predict medication overuse in migraine patients

Machine learning (ML) is largely used to develop automatic predictors in migraine classification but automatic predictors for medication overuse (MO) in migraine are still in their infancy. Thus, to understand the benefits of ML in MO prediction, we explored an automated predictor to estimate MO risk in migraine. To achieve this objective, a study was designed to analyze the performance of a customized ML-based decision support system that combines support vector machines and Random Optimization (RO-MO). We used RO-MO to extract prognostic information from demographic, clinical and biochemical data. Using a dataset of 777 consecutive migraine patients we derived a set of predictors with discriminatory power for MO higher than that observed for baseline SVM. The best four were incorporated into the final RO-MO decision support system and risk evaluation on a five-level stratification was performed. ROC analysis resulted in a c-statistic of 0.83 with a sensitivity and specificity of 0.69 and 0.87, respectively, and an accuracy of 0.87 when MO was predicted by at least three RO-MO models. Logistic regression analysis confirmed that the derived RO-MO system could effectively predict MO with ORs of 5.7 and 21.0 for patients classified as probably (3 predictors positive), or definitely at risk of MO (4 predictors positive), respectively. In conclusion, a combination of ML and RO - taking into consideration clinical/biochemical features, drug exposure and lifestyle - might represent a valuable approach to MO prediction in migraine and holds the potential for improving model precision through weighting the relative importance of attributes.

Subgrouping Factors Influencing Migraine Intensity in Women: A Semi-automatic Methodology Based on Machine Learning and Information Geometry

BACKGROUND

Migraine is a heterogeneous condition with multiple clinical manifestations. Machine learning algorithms permit the identification of population groups, providing analytical advantages over other modeling techniques.

OBJECTIVE

The aim of this study was to analyze critical features that permit the differentiation of subgroups of patients with migraine according to the intensity and frequency of attacks by using machine learning algorithms.

METHODS

Sixty-seven women with migraine participated. Clinical features of migraine, related disability (Migraine Disability Assessment Scale), anxiety/depressive levels (Hospital Anxiety and Depression Scale), anxiety state/trait levels (State-Trait Anxiety Inventory), and pressure pain thresholds (PPTs) over the temporalis, neck, second metacarpal, and tibialis anterior were collected. Physical examination included the flexion-rotation test, cervical range of cervical motion, forward head position while sitting and standing, passive accessory intervertebral movements (PAIVMs) with headache reproduction, and joint positioning sense error. Subgrouping was based on machine learning algorithms by using the nearest neighbors algorithm, multisource variability assessment, and random forest model.

RESULTS

For migraine intensity, group 2 (women with a regular migraine headache intensity score of 7 on an 11-point Numeric Pain Rating Scale [where 0 = no pain and 10 = maximum pain]) were younger and had 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, shorter migraine history, and lower cranio-vertebral angles while standing than the remaining migraine intensity subgroups. The most discriminative variable was the flexion-rotation test score of the symptomatic side. For migraine frequency, no model was able to identify differences between groups (ie, patients with episodic or chronic migraine).

CONCLUSIONS

A subgroup of women with migraine who had common migraine intensity was identified with machine learning algorithms.

PROAFTN Classifier for Feature Selection with Application to Alzheimer Metabolomics Data Analysis

Early and accurate Alzheimer’s disease (AD) diagnosis remains a challenge. Recently, increasing efforts have been focused towards utilization of metabolomics data for the discovery of biomarkers for screening and diagnosis of AD. Several machine learning approaches were explored for classifying the blood metabolomics profiles of cognitively healthy and AD patients. Differentiation between AD, mild cognitive impairment (MCI) and cognitively healthy subjects remains difficult. In this paper, we propose a new machine learning approach for the selection of a subset of features that provide an improvement in classification rates between these three levels of cognitive disorders. Our experimental results demonstrate that utilization of these selected metabolic markers improves the performance of several classifiers in comparison to the classification accuracy obtained for the complete metabolomics dataset. The obtained results indicate that our algorithms are effective in discovering a panel of biomarkers of AD and MCI from metabolomics data suggesting the possibility to develop a noninvasive blood diagnostic technique of AD and MCI.

Delineating conditions and subtypes in chronic pain using neuroimaging

Differentiating subtypes of chronic pain still remains a challenge—both from a subjective and objective point of view. Personalized medicine is the current goal of modern medical care and is limited by the subjective nature of patient self-reporting of symptoms and behavioral evaluation. Physiology-focused techniques such as genome and epigenetic analyses inform the delineation of pain groups; however, except under rare circumstances, they have diluted effects that again, share a common reliance on behavioral evaluation. The application of structural neuroimaging towards distinguishing pain subtypes is a growing field and may inform pain-group classification through the analysis of brain regions showing hypertrophic and atrophic changes in the presence of pain. Analytical techniques such as machine-learning classifiers have the capacity to process large volumes of data and delineate diagnostically relevant information from neuroimaging analysis. The issue of defining a “brain type” is an emerging field aimed at interpreting observed brain changes and delineating their clinical identity/significance. In this review, 2 chronic pain conditions (migraine and irritable bowel syndrome) with similar clinical phenotypes are compared in terms of their structural neuroimaging findings. Independent investigations are compared with findings from application of machine-learning algorithms. Findings are discussed in terms of differentiating patient subgroups using neuroimaging data in patients with chronic pain and how they may be applied towards defining a personalized pain signature that helps segregate patient subgroups (eg, migraine with and without aura, with or without nausea; irritable bowel syndrome vs other functional gastrointestinal disorders).

A decision support system to follow up and diagnose primary headache patients using semantically enriched data

BACKGROUND

Headache disorders are an important health burden, having a large health-economic impact worldwide. Current treatment & follow-up processes are often archaic, creating opportunities for computer-aided and decision support systems to increase their efficiency. Existing systems are mostly completely data-driven, and the underlying models are a black-box, deteriorating interpretability and transparency, which are key factors in order to be deployed in a clinical setting.

METHODS

In this paper, a decision support system is proposed, composed of three components: (i) a cross-platform mobile application to capture the required data from patients to formulate a diagnosis, (ii) an automated diagnosis support module that generates an interpretable decision tree, based on data semantically annotated with expert knowledge, in order to support physicians in formulating the correct diagnosis and (iii) a web application such that the physician can efficiently interpret captured data and learned insights by means of visualizations.

RESULTS

We show that decision tree induction techniques achieve competitive accuracy rates, compared to other black- and white-box techniques, on a publicly available dataset, referred to as migbase. Migbase contains aggregated information of headache attacks from 849 patients. Each sample is labeled with one of three possible primary headache disorders. We demonstrate that we are able to reduce the classification error, statistically significant (ρ≤0.05), with more than 10% by balancing the dataset using prior expert knowledge. Furthermore, we achieve high accuracy rates by using features extracted using the Weisfeiler-Lehman kernel, which is completely unsupervised. This makes it an ideal approach to solve a potential cold start problem.

CONCLUSION

Decision trees are the perfect candidate for the automated diagnosis support module. They achieve predictive performances competitive to other techniques on the migbase dataset and are, foremost, completely interpretable. Moreover, the incorporation of prior knowledge increases both predictive performance as well as transparency of the resulting predictive model on the studied dataset.