Multi-task multiple kernel machines for personalized pain recognition from functional near-infrared spectroscopy brain signals

Abstract

Currently there is no validated objective measure of pain. Recent neuroimaging studies have explored the feasibility of using functional near-infrared spectroscopy (fNIRS) to measure alterations in brain function in evoked and ongoing pain. In this study, we applied multi-task machine learning methods to derive a practical algorithm for pain detection derived from fNIRS signals in healthy volunteers exposed to a painful stimulus. Especially, we employed multi-task multiple kernel learning to account for the inter-subject variability in pain response. Our results support the use of fNIRS and machine learning techniques in developing objective pain detection, and also highlight the importance of adopting personalized analysis in the process.

Pain Detection with fNIRS-Measured Brain Signals: A Personalized Machine Learning Approach Using the Wavelet Transform and Bayesian Hierarchical Modeling with Dirichlet Process Priors

“…After processing the fNIRS signals as described in Sec.II-B, windows of duration 20 seconds were extracted from the HbO signals. The choice of signal modality (that is, HbO) and window size was informed by previous work done on pain detection from fNIRS [14], [33]. From these windows, we extracted the D = 80 features described in Sec.II-C from the prefrontal fNIRS channels depicted in Fig.1.…”

“…Whereas previous studies (e.g. [14]) have included other brain regions such as the sensorimotor cortex, focusing only on prefrontal cortical area has many advantages including its ease of access (especially in surgical settings where the patient is kept in a supine position) and the ability to avoid hair contamination in the recorded fNIRS signals. Moreover, the need of only sampling from the prefrontal cortex greatly reduces the required number of fNIRS optodes and optical fibres to be mounted on the participant's head, leading to an easier installation process and more patient comfort.…”

“…1} is a binary label and N is the number of samples in the dataset. Many classification tasks can be viewed as consisting of multiple correlated subtasks, such as pain detection in a particular patient subpopulation that responds to noxious stimuli in a characteristic way, similar to yet different from other patient subpopulations [7], [8], [14], [15]. Therefore, pooling all subtasks and treating them as a single task may not be appropriate.…”

Detection of Real-World Driving-Induced Affective State Using Physiological Signals and Multi-View Multi-Task Machine Learning

“…Drive profiling and task assignment with spectral clustering All three datasets described in Sec.II contain data from multiple drives. Following [19], [20], we group E drives into different profiles based on the unique physiological responses of the drivers corresponding to driving-induced affective state, for each drive d in the dataset. These profiles are then used to define the tasks of our multi-view multi-task machine learning model (see Sec.III-C), such that each task in the model corresponds to a distinct profile.…”

“…Following [20], [23], we consider two types of regularizers Ω(·), the 1 -norm and 2 -norm regularizers respectively: where the coefficient ν controls the influence of the regularizer on Eq. 2, with larger values of ν enforcing similar kernel combination parameters across the tasks r, which correspond to each of the clusters in Sec.III-B.…”