Demirci_EU Neurology 10/03/2010 09:56 Page 105
Functional Magnetic Resonance Imaging – Implications for Detection of Schizophrenia
and sensitivity in a two-class prediction model. Reporting only the Georgopoulos et al.
16
presented a classification method using MEG
overall prediction accuracy will not explain the overall performance of and assigned group memberships to subjects with various illnesses
the technique. (Alzheimers disease, schizophrenia, multiple sclerosis, Sjogren’s
syndrome, chronic alcoholism, facial pain). They used 248 axial
Ford et al.
13
combined structural and functional MRI data for gradiometers on 142 human subjects and obtained 30,628 partial zero-
classification purposes. They extracted hippocampal formation by lag cross-correlations between sensors for all sensor pairs and used
applying a mask and then extracted the functional and structural data them as the predictor set. They looked for subsets of this predictor set
within the mask. The high-dimensional data were then projected onto and investigated whether any such predictor subsets correctly classified
a lower-dimensional space, and Fisher’s linear discriminant (FLD) subjects into their respective groups. This was a dimensionality
analysis was used to maximise the ratio of between-class and within- reduction problem. They indicated that a subset of 12 predictors
class variability considering the training set. The prediction accuracy (correlations) gave a prediction accuracy of 86.6% and assigned 86.6%
of the classifier was tested using a total of 23 subjects (15 of the subjects to their respective groups correctly. They used this same
schizophrenia patients and eight healthy controls) with a leave-one- set of 12 predictors and presented cross-validation results around
out method. One of the subjects was removed from the whole set for 77–79% with two different jackknifed methods: k-fold and leave-one-out.
validation purposes (K=1, one-fold cross-validation), and the rest of Although these results are encouraging, especially given the specificity
the subjects were used as training data. A maximum classification of the approach to multiple different groups, they appear to be biased to
accuracy of 83–87% was presented, which is reasonable. However, it the data at hand because the same set of 12 predictors was used for
would also be informative to know the prediction performances of each different training set, and a different set of predictors was not
both classes separately, especially in this case where the number of obtained for each training set separately.
subjects in the two groups differs. For example, for this particular set,
85% overall prediction performance could possibly be obtained with Fan et al.
17
applied a multivariate classification approach combining data
100% detection performance on schizophrenia patients and only a from both a functional feature map (cerebral blood flow) and structural
57% detection performance on healthy controls (43% false alarm), MRI data to detect brain abnormality associated with pre-natal cocaine
which would indicate a poor performance on healthy controls. exposure in adolescents. Regions with voxels of similar correlation to
the disease were obtained using a Pearson correlation coefficient for
In a similar study, Ford et al. also proposed to use principal component three different feature maps separately. A leave-one-out method
analysis (PCA) to represent subjects in a lower-dimensional space with was employed and an effective cross-validation strategy was followed
maximal variance and uncorrelated samples, based on the idea that to measure the overall correlation of a feature to class label. Then,
fMRI activation patterns show differentiations among healthy controls, statistical regional features (histograms) and a PCA were used to
patients with schizophrenia, Alzheimer’s disease and mild traumatic represent each region with a feature vector. Subjects were represented
brain injury. The FLD classifier was applied to fMRI brain activation with the vectors from three different feature maps. Promising results
maps in this lower-dimensional space to differentiate patients from were obtained on 49 subjects (25 pre-natal cocaine-exposed subjects
healthy controls.
14
The prediction accuracy of the schizophrenia and 24 normal controls). Fan et al.
17
mention the possibility that obtained
patients varied between 60 and 80% for different numbers of principal classification accuracy might be an indication of overfitting based on the
components on a set of 25 subjects (10 healthy controls and 15 random permutation tests they performed.
patients with schizophrenia). Specificity and sensitivity performances
were not reported separately. The authors appropriately pointed Parameter Selection (Optimisation Bias)
out that their results should be interpreted cautiously because of the Selecting a set of parameters based on the prediction accuracy obtained
small data set. and using the set of parameters with the best performance in the final
classifier is an example of parameter selection or optimisation bias.
Selection Bias (Overfitting) Even if cross-validation is applied in every step of the algorithm, we
Application of cross-validation tools appropriately during class might cause optimisation bias if we run our experiment multiple times
prediction and diagnosis studies is crucial and even more important with the same data and select the parameters accordingly to get the
than the choice of algorithmic methods. K-fold cross-validation best prediction performance. The set of parameters obtained using
techniques should be applied in all steps of designing a classifier, the best classification performance might not necessarily give the best
including feature selection. For generalisable conclusions and results, performance for a different set and might not be the best projection that
cross-validation should be applied at every stage of the classification could be applied for the best diagnosis method.
algorithms, not only during performance evaluation. Reproducibility of
the obtained classification accuracies also requires careful selection of Data Fusion of Functional Magnetic Resonance
data, and the selection process should be explained in detail in studies. Imaging with Other Data Sets
Various types of data, such as structural imaging,
15,17
functional
Job et al.
15
extracted three brain areas in a comparison between eight imaging, neuropsychological score and genetics, have been used
schizophrenia subjects and 57 control subjects. The same subjects either individually or combined to develop biomarkers for prognostic
used in region selection were also used in classification. Such an or diagnostic purposes. The use of combinations of multiple measures
approach tends to bias the results as the information on the classes to improve diagnostic results and obtain more dependable
has been used to select the brain areas. Areas showing possible conclusions is attracting increasing interest.
differentiations between schizophrenia and control subjects can be
selected with minimal to no bias by determining the regions without Environmental factors play an important role in the development of
using the test subject in each iteration and then performing schizophrenia, but studies have also consistently shown that relatives
classification of the left-out subject only. of schizophrenia patients have a higher risk of having the impairment
EUROPEAN NEUROLOGICAL REVIEW 105
Page 1 |
Page 2 |
Page 3 |
Page 4 |
Page 5 |
Page 6 |
Page 7 |
Page 8 |
Page 9 |
Page 10 |
Page 11 |
Page 12 |
Page 13 |
Page 14 |
Page 15 |
Page 16 |
Page 17 |
Page 18 |
Page 19 |
Page 20 |
Page 21 |
Page 22 |
Page 23 |
Page 24 |
Page 25 |
Page 26 |
Page 27 |
Page 28 |
Page 29 |
Page 30 |
Page 31 |
Page 32 |
Page 33 |
Page 34 |
Page 35 |
Page 36 |
Page 37 |
Page 38 |
Page 39 |
Page 40 |
Page 41 |
Page 42 |
Page 43 |
Page 44 |
Page 45 |
Page 46 |
Page 47 |
Page 48 |
Page 49 |
Page 50 |
Page 51 |
Page 52 |
Page 53 |
Page 54 |
Page 55 |
Page 56 |
Page 57 |
Page 58 |
Page 59 |
Page 60 |
Page 61 |
Page 62 |
Page 63 |
Page 64 |
Page 65 |
Page 66 |
Page 67 |
Page 68 |
Page 69 |
Page 70 |
Page 71 |
Page 72 |
Page 73 |
Page 74 |
Page 75 |
Page 76 |
Page 77 |
Page 78 |
Page 79 |
Page 80 |
Page 81 |
Page 82 |
Page 83 |
Page 84 |
Page 85 |
Page 86 |
Page 87 |
Page 88 |
Page 89 |
Page 90 |
Page 91 |
Page 92 |
Page 93 |
Page 94 |
Page 95 |
Page 96 |
Page 97 |
Page 98 |
Page 99 |
Page 100 |
Page 101 |
Page 102 |
Page 103 |
Page 104 |
Page 105 |
Page 106 |
Page 107 |
Page 108 |
Page 109 |
Page 110 |
Page 111 |
Page 112 |
Page 113 |
Page 114 |
Page 115 |
Page 116 |
Page 117 |
Page 118 |
Page 119 |
Page 120 |
Page 121 |
Page 122 |
Page 123 |
Page 124 |
Page 125 |
Page 126 |
Page 127 |
Page 128 |
Page 129 |
Page 130 |
Page 131 |
Page 132