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: After psychological trauma, why do some only some parts of the traumatic event return as intrusive memories while others do not?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

The top weighted input features compromising 8 ICA components (a–h) and their corresponding time points (in brackets) involved in the prediction of a Flashback scene at the time of viewing traumatic footage. The ICA components are presented in the weighted order of the features used in the classifier. Features could be involved at 1 or all of 3 time points; i) the initial 6 s of the Flashback scene, ii) the remainder of the Flashback scene or iii) the 12 s post Flashback scene. Proposed functions of networks within the feature are included to provide a guide to their potential role in intrusive memory formation with names taken from Smith et al. (2009). 6 images are taken for each ICA component and are shown in the axial plane with their corresponding z coordinate. The underlying image is the Montreal Neurological Institute (MNI) 152 template, z-statistic images are thresholded at z > 2.3. z-Statistic range is represented by the change in colour.
© Copyright Policy
Related In: Results  -  Collection


getmorefigures.php?uid=PMC4222599&req=5

fig3: The top weighted input features compromising 8 ICA components (a–h) and their corresponding time points (in brackets) involved in the prediction of a Flashback scene at the time of viewing traumatic footage. The ICA components are presented in the weighted order of the features used in the classifier. Features could be involved at 1 or all of 3 time points; i) the initial 6 s of the Flashback scene, ii) the remainder of the Flashback scene or iii) the 12 s post Flashback scene. Proposed functions of networks within the feature are included to provide a guide to their potential role in intrusive memory formation with names taken from Smith et al. (2009). 6 images are taken for each ICA component and are shown in the axial plane with their corresponding z coordinate. The underlying image is the Montreal Neurological Institute (MNI) 152 template, z-statistic images are thresholded at z > 2.3. z-Statistic range is represented by the change in colour.

Mentions: A total of 117 input features (i.e. averaged activation across the 39 ICA brain networks during the 3 defined time points of the scenes; the initial 6 s of the scene, the remaining duration of the scene after the initial 6 s, and the 12 s post scene) contributed to intrusive memory scene prediction. Below we describe the top weighted input features of the classifier for predicting Flashback versus Potential events (i.e. the features contributing most strongly towards prediction in terms of their weighting within the classifier). We also note their possible cognitive function. While these networks are those top weighted by the classifier, this is not a statistical measure and can only provide a guide towards their predictive contribution. There are 2 components of each feature; the location in the brain (i.e. the ICA component) and the timing of activation. The top weighted input features comprise 8 ICA components, 3 of which were important for intrusive memory prediction at 2 time points (see Fig. 3; ICA components (a–h) are displayed according to their weighting, activation time points are displayed in brackets).


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)

The top weighted input features compromising 8 ICA components (a–h) and their corresponding time points (in brackets) involved in the prediction of a Flashback scene at the time of viewing traumatic footage. The ICA components are presented in the weighted order of the features used in the classifier. Features could be involved at 1 or all of 3 time points; i) the initial 6 s of the Flashback scene, ii) the remainder of the Flashback scene or iii) the 12 s post Flashback scene. Proposed functions of networks within the feature are included to provide a guide to their potential role in intrusive memory formation with names taken from Smith et al. (2009). 6 images are taken for each ICA component and are shown in the axial plane with their corresponding z coordinate. The underlying image is the Montreal Neurological Institute (MNI) 152 template, z-statistic images are thresholded at z > 2.3. z-Statistic range is represented by the change in colour.
© Copyright Policy
Related In: Results  -  Collection

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

fig3: The top weighted input features compromising 8 ICA components (a–h) and their corresponding time points (in brackets) involved in the prediction of a Flashback scene at the time of viewing traumatic footage. The ICA components are presented in the weighted order of the features used in the classifier. Features could be involved at 1 or all of 3 time points; i) the initial 6 s of the Flashback scene, ii) the remainder of the Flashback scene or iii) the 12 s post Flashback scene. Proposed functions of networks within the feature are included to provide a guide to their potential role in intrusive memory formation with names taken from Smith et al. (2009). 6 images are taken for each ICA component and are shown in the axial plane with their corresponding z coordinate. The underlying image is the Montreal Neurological Institute (MNI) 152 template, z-statistic images are thresholded at z > 2.3. z-Statistic range is represented by the change in colour.
Mentions: A total of 117 input features (i.e. averaged activation across the 39 ICA brain networks during the 3 defined time points of the scenes; the initial 6 s of the scene, the remaining duration of the scene after the initial 6 s, and the 12 s post scene) contributed to intrusive memory scene prediction. Below we describe the top weighted input features of the classifier for predicting Flashback versus Potential events (i.e. the features contributing most strongly towards prediction in terms of their weighting within the classifier). We also note their possible cognitive function. While these networks are those top weighted by the classifier, this is not a statistical measure and can only provide a guide towards their predictive contribution. There are 2 components of each feature; the location in the brain (i.e. the ICA component) and the timing of activation. The top weighted input features comprise 8 ICA components, 3 of which were important for intrusive memory prediction at 2 time points (see Fig. 3; ICA components (a–h) are displayed according to their weighting, activation time points are displayed in brackets).

Bottom Line: After psychological trauma, why do some only some parts of the traumatic event return as intrusive memories while others do not?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