Limits...
Investigating the use of support vector machine classification on structural brain images of preterm-born teenagers as a biological marker.

Chu C, Lagercrantz H, Forssberg H, Nagy Z - PLoS ONE (2015)

Bottom Line: Separately, a random half of the available data were used for training twice and each time the other, unseen, half of the data was classified, resulting 86% and 91% accurate classifications.Statistically significant correlations were also found between IQ (R = -0.30, p < 0.001) and the distance to decision boundary.The long-term goal is to automatically and non-invasively predict the outcome of preterm-born individuals on an individual basis using as early a scan as possible.

View Article: PubMed Central - PubMed

Affiliation: DeepMind Technologies Ltd., London, United Kingdom; Wellcome Trust Centre for Neuroimaging, UCL Institute of Neurology, London, United Kingdom.

ABSTRACT
Preterm birth has been shown to induce an altered developmental trajectory of brain structure and function. With the aid support vector machine (SVM) classification methods we aimed to investigate whether MRI data, collected in adolescence, could be used to predict whether an individual had been born preterm or at term. To this end we collected T1-weighted anatomical MRI data from 143 individuals (69 controls, mean age 14.6y). The inclusion criteria for those born preterm were birth weight ≤ 1500g and gestational age < 37w. A linear SVM was trained on the grey matter segment of MR images in two different ways. First, all the individuals were used for training and classification was performed by the leave-one-out method, yielding 93% correct classification (sensitivity = 0.905, specificity = 0.942). Separately, a random half of the available data were used for training twice and each time the other, unseen, half of the data was classified, resulting 86% and 91% accurate classifications. Both gestational age (R = -0.24, p<0.04) and birth weight (R = -0.51, p < 0.001) correlated with the distance to decision boundary within the group of individuals born preterm. Statistically significant correlations were also found between IQ (R = -0.30, p < 0.001) and the distance to decision boundary. Those born small for gestational age did not form a separate subgroup in these analyses. The high rate of correct classification by the SVM motivates further investigation. The long-term goal is to automatically and non-invasively predict the outcome of preterm-born individuals on an individual basis using as early a scan as possible.

No MeSH data available.


Related in: MedlinePlus

Gray Matter segment discriminative power for classification and p value map.Mean gray matter segment of all subjects is shown in the middle column for anatomical orientation. The color code on the images of the left column is in arbitrary units and indicates the weight (i.e. discriminative power) in each voxel. This value in each voxel is relative to the weight in all other voxels containing brain GM. The plots in the right column depict the corresponding p value maps. The color bar on the far right belongs only to the p value maps.
© Copyright Policy
Related In: Results  -  Collection

License
getmorefigures.php?uid=PMC4383582&req=5

pone.0123108.g002: Gray Matter segment discriminative power for classification and p value map.Mean gray matter segment of all subjects is shown in the middle column for anatomical orientation. The color code on the images of the left column is in arbitrary units and indicates the weight (i.e. discriminative power) in each voxel. This value in each voxel is relative to the weight in all other voxels containing brain GM. The plots in the right column depict the corresponding p value maps. The color bar on the far right belongs only to the p value maps.

Mentions: All MRI data were collected at the Karolinska Hospital using a 1.5 T Signa scanner (General Electric, Waukesha, WI). The protocol included a T1–weighted gradient echo 3D anatomical image with echo time = 6 ms, repetition time = 24 ms, flip angle = 30° and voxel size 0.98 mm x 0.98 mm x 1.5 mm. Using the freely available SPM software (Wellcome Trust Centre for Neuroimaging, London, UK) the T1–weighted images were segmented into gray matter (GM), white matter and cerebro-spinal fluid [25]. Using DARTEL [26] in SPM, which is a robust non–linear registration method [27], the GM segments were normalized to the common space of the mean GM segment of all the subjects, then subsequently modulated by the Jacobian determinants, re–sampled to 1.5 mm isotropic voxels and then smoothed isotropically using a filter with 6 mm at full width at half maximum. Subsequently, the above GM images were used, without feature selection or feature reduction, as input features to train a linear, hard margin SVM (http://www.csie.ntu.edu.tw/~cjlin/libsvm/) in Matlab (Mathworks, Natick, MA, USA). The groups of control and preterm born individuals were labeled -1 and +1 respectively [12]. Leave–one–out cross validation was applied to estimate accuracies [15]. The prediction scores (i.e. distance to the decision boundary including sign—see Fig 1), which is the dot–product between a test image and the weight map (Fig 2) plus the bias, were further used for correlation analyses. To assess the validity of results that the above leave–one–out procedure yielded, we made two additional training/classification cycles. In each case, we randomly selected 37 out of the 74 cases and 35 of the 69 controls to train the SVM. Subsequently the remaining, unseen individuals were given to the SVM for classification. In addition to the simple classification, receive operating characteristic (ROC) curves and the area under the ROC cureves (AUC) were calculated.


Investigating the use of support vector machine classification on structural brain images of preterm-born teenagers as a biological marker.

Chu C, Lagercrantz H, Forssberg H, Nagy Z - PLoS ONE (2015)

Gray Matter segment discriminative power for classification and p value map.Mean gray matter segment of all subjects is shown in the middle column for anatomical orientation. The color code on the images of the left column is in arbitrary units and indicates the weight (i.e. discriminative power) in each voxel. This value in each voxel is relative to the weight in all other voxels containing brain GM. The plots in the right column depict the corresponding p value maps. The color bar on the far right belongs only to the p value maps.
© Copyright Policy
Related In: Results  -  Collection

License
Show All Figures
getmorefigures.php?uid=PMC4383582&req=5

pone.0123108.g002: Gray Matter segment discriminative power for classification and p value map.Mean gray matter segment of all subjects is shown in the middle column for anatomical orientation. The color code on the images of the left column is in arbitrary units and indicates the weight (i.e. discriminative power) in each voxel. This value in each voxel is relative to the weight in all other voxels containing brain GM. The plots in the right column depict the corresponding p value maps. The color bar on the far right belongs only to the p value maps.
Mentions: All MRI data were collected at the Karolinska Hospital using a 1.5 T Signa scanner (General Electric, Waukesha, WI). The protocol included a T1–weighted gradient echo 3D anatomical image with echo time = 6 ms, repetition time = 24 ms, flip angle = 30° and voxel size 0.98 mm x 0.98 mm x 1.5 mm. Using the freely available SPM software (Wellcome Trust Centre for Neuroimaging, London, UK) the T1–weighted images were segmented into gray matter (GM), white matter and cerebro-spinal fluid [25]. Using DARTEL [26] in SPM, which is a robust non–linear registration method [27], the GM segments were normalized to the common space of the mean GM segment of all the subjects, then subsequently modulated by the Jacobian determinants, re–sampled to 1.5 mm isotropic voxels and then smoothed isotropically using a filter with 6 mm at full width at half maximum. Subsequently, the above GM images were used, without feature selection or feature reduction, as input features to train a linear, hard margin SVM (http://www.csie.ntu.edu.tw/~cjlin/libsvm/) in Matlab (Mathworks, Natick, MA, USA). The groups of control and preterm born individuals were labeled -1 and +1 respectively [12]. Leave–one–out cross validation was applied to estimate accuracies [15]. The prediction scores (i.e. distance to the decision boundary including sign—see Fig 1), which is the dot–product between a test image and the weight map (Fig 2) plus the bias, were further used for correlation analyses. To assess the validity of results that the above leave–one–out procedure yielded, we made two additional training/classification cycles. In each case, we randomly selected 37 out of the 74 cases and 35 of the 69 controls to train the SVM. Subsequently the remaining, unseen individuals were given to the SVM for classification. In addition to the simple classification, receive operating characteristic (ROC) curves and the area under the ROC cureves (AUC) were calculated.

Bottom Line: Separately, a random half of the available data were used for training twice and each time the other, unseen, half of the data was classified, resulting 86% and 91% accurate classifications.Statistically significant correlations were also found between IQ (R = -0.30, p < 0.001) and the distance to decision boundary.The long-term goal is to automatically and non-invasively predict the outcome of preterm-born individuals on an individual basis using as early a scan as possible.

View Article: PubMed Central - PubMed

Affiliation: DeepMind Technologies Ltd., London, United Kingdom; Wellcome Trust Centre for Neuroimaging, UCL Institute of Neurology, London, United Kingdom.

ABSTRACT
Preterm birth has been shown to induce an altered developmental trajectory of brain structure and function. With the aid support vector machine (SVM) classification methods we aimed to investigate whether MRI data, collected in adolescence, could be used to predict whether an individual had been born preterm or at term. To this end we collected T1-weighted anatomical MRI data from 143 individuals (69 controls, mean age 14.6y). The inclusion criteria for those born preterm were birth weight ≤ 1500g and gestational age < 37w. A linear SVM was trained on the grey matter segment of MR images in two different ways. First, all the individuals were used for training and classification was performed by the leave-one-out method, yielding 93% correct classification (sensitivity = 0.905, specificity = 0.942). Separately, a random half of the available data were used for training twice and each time the other, unseen, half of the data was classified, resulting 86% and 91% accurate classifications. Both gestational age (R = -0.24, p<0.04) and birth weight (R = -0.51, p < 0.001) correlated with the distance to decision boundary within the group of individuals born preterm. Statistically significant correlations were also found between IQ (R = -0.30, p < 0.001) and the distance to decision boundary. Those born small for gestational age did not form a separate subgroup in these analyses. The high rate of correct classification by the SVM motivates further investigation. The long-term goal is to automatically and non-invasively predict the outcome of preterm-born individuals on an individual basis using as early a scan as possible.

No MeSH data available.


Related in: MedlinePlus