Machine learning approaches in brain imaging analysis
| Supervised algorithms | Unsupervised methods | |
| Techniques | Support vector machines, random forest and sparse partial least squares discriminate analysis. | Hierarchical clustering, principal coordinate analysis and sparse k-mean clustering. |
| Rationale | Reduce dimensionality of multimodal large-scale functional, structural and anatomical neuroimaging data by finding a set of brain signatures comprised by selected set of brain features. These brain signatures form the basis of a classification or predictive algorithms that provide insight into the pathophysiological mechanisms. | Integrate and decipher large amounts of multivariate neuroimaging data to subgroups of patients based on objective biological markers and characterise central nervous system alterations for further pathophysiological investigations targeting treatment of chronic pain and other brain disorders. |
| Examples | Functional dyspepsia145 and IBS.20 | Has been applied successfully to clinical, physiological and microbiota data in IBS51 176 177 but not brain data. |
| Outcomes | Identify patterns that discriminate and predict acute pain state, pain diagnosis and pharmacological and non-pharmacological treatment outcomes including longitudinal symptom trajectories. | Future identification of therapeutic targets and development of tailored patient treatment. In combination with other biological data, results may translate into identification of novel therapeutic targets and development of individualised pain therapies based on brain signatures.178–180 |