Limits...
First steps in using machine learning on fMRI data to predict intrusive memories of traumatic film footage.

Clark IA, Niehaus KE, Duff EP, Di Simplicio MC, Clifford GD, Smith SM, Mackay CE, Woolrich MW, Holmes EA - Behav Res Ther (2014)

Bottom Line: While traditional mass univariate regression analysis highlighted an association between brain processing and symptomatology, this is not the same as prediction.Using MVPA and a machine learning classifier, it was possible to predict later intrusive memories across participants with 68% accuracy, and within a participant with 97% accuracy; i.e. the classifier could identify out of multiple scenes those that would later return as an intrusive memory.MVPA opens the possibility of decoding brain activity to reconstruct idiosyncratic cognitive events with relevance to understanding and predicting mental health symptoms.

View Article: PubMed Central - PubMed

Affiliation: University Department of Psychiatry, Warneford Hospital, University of Oxford, United Kingdom.

Show MeSH

Related in: MedlinePlus

Illustration of the prediction aspect of the machine learning analysis. a. Shows the training element of the machine learning approach. The classifier was provided with information concerning the timing of the Flashback scenes (emotional scenes that returned as a intrusive memory for that individual) and Potential scenes (emotional scenes that did not return as a intrusive memory for that individual, but did in other participants) from which to learn the patterns of brain activation for each scene type. Training was performed on all but 1 participant. b. Shows the predictive element of the machine learning approach. For the 1 participant not included in training the machine learning classifier goes through the brain activation data and attempts to identify the Flashback and Potential scenes.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4222599&req=5

fig2: Illustration of the prediction aspect of the machine learning analysis. a. Shows the training element of the machine learning approach. The classifier was provided with information concerning the timing of the Flashback scenes (emotional scenes that returned as a intrusive memory for that individual) and Potential scenes (emotional scenes that did not return as a intrusive memory for that individual, but did in other participants) from which to learn the patterns of brain activation for each scene type. Training was performed on all but 1 participant. b. Shows the predictive element of the machine learning approach. For the 1 participant not included in training the machine learning classifier goes through the brain activation data and attempts to identify the Flashback and Potential scenes.

Mentions: The support vector machine (SVM) classifier was first optimised on the larger of the 2 data sets (Clark et al., submitted for publication; 35 participants). A labelled sequence of Flashback and Potential scene time points in the film was created from the diaries for each individual participant (as each person may have different intrusions). The input features detailed above, reflecting activation across the brain, were extracted from the fMRI data during these Flashback and Potential time points (see Niehaus et al., 2014 for details). The SVM was then trained on this data to learn the patterns for both scene types, using a leave-one-out methodology to provide a test case: for 1 participant brain activation was not included in the training. Based upon the learned patterns of activity from all other participants, the classifier then attempted to identify the film scenes that later induced intrusive memories for the left-out participant. Identification based on brain activation patterns was the checked against the participant's diary entries (see Fig. 2). This leave-one-out ‘cross-validation loop’ was conducted 35 times, each one with a different participant left out of the training set. Results were averaged over the performance of the SVM on the left-out participant.


First steps in using machine learning on fMRI data to predict intrusive memories of traumatic film footage.

Clark IA, Niehaus KE, Duff EP, Di Simplicio MC, Clifford GD, Smith SM, Mackay CE, Woolrich MW, Holmes EA - Behav Res Ther (2014)

Illustration of the prediction aspect of the machine learning analysis. a. Shows the training element of the machine learning approach. The classifier was provided with information concerning the timing of the Flashback scenes (emotional scenes that returned as a intrusive memory for that individual) and Potential scenes (emotional scenes that did not return as a intrusive memory for that individual, but did in other participants) from which to learn the patterns of brain activation for each scene type. Training was performed on all but 1 participant. b. Shows the predictive element of the machine learning approach. For the 1 participant not included in training the machine learning classifier goes through the brain activation data and attempts to identify the Flashback and Potential scenes.
© Copyright Policy
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC4222599&req=5

fig2: Illustration of the prediction aspect of the machine learning analysis. a. Shows the training element of the machine learning approach. The classifier was provided with information concerning the timing of the Flashback scenes (emotional scenes that returned as a intrusive memory for that individual) and Potential scenes (emotional scenes that did not return as a intrusive memory for that individual, but did in other participants) from which to learn the patterns of brain activation for each scene type. Training was performed on all but 1 participant. b. Shows the predictive element of the machine learning approach. For the 1 participant not included in training the machine learning classifier goes through the brain activation data and attempts to identify the Flashback and Potential scenes.
Mentions: The support vector machine (SVM) classifier was first optimised on the larger of the 2 data sets (Clark et al., submitted for publication; 35 participants). A labelled sequence of Flashback and Potential scene time points in the film was created from the diaries for each individual participant (as each person may have different intrusions). The input features detailed above, reflecting activation across the brain, were extracted from the fMRI data during these Flashback and Potential time points (see Niehaus et al., 2014 for details). The SVM was then trained on this data to learn the patterns for both scene types, using a leave-one-out methodology to provide a test case: for 1 participant brain activation was not included in the training. Based upon the learned patterns of activity from all other participants, the classifier then attempted to identify the film scenes that later induced intrusive memories for the left-out participant. Identification based on brain activation patterns was the checked against the participant's diary entries (see Fig. 2). This leave-one-out ‘cross-validation loop’ was conducted 35 times, each one with a different participant left out of the training set. Results were averaged over the performance of the SVM on the left-out participant.

Bottom Line: While traditional mass univariate regression analysis highlighted an association between brain processing and symptomatology, this is not the same as prediction.Using MVPA and a machine learning classifier, it was possible to predict later intrusive memories across participants with 68% accuracy, and within a participant with 97% accuracy; i.e. the classifier could identify out of multiple scenes those that would later return as an intrusive memory.MVPA opens the possibility of decoding brain activity to reconstruct idiosyncratic cognitive events with relevance to understanding and predicting mental health symptoms.

View Article: PubMed Central - PubMed

Affiliation: University Department of Psychiatry, Warneford Hospital, University of Oxford, United Kingdom.

Show MeSH
Related in: MedlinePlus