Prediction of treatment response with multi-modality MRI in first-episode antipsychotic-naive schizophrenia
Lu Liu1, Yuan Xiao1, Wenjing Zhang1, and Su Lui2

1Huaxi MR Research Center (HMRRC), Department of Radiology, West China Hospital of Sichuan University, Chengdu, China, People's Republic of, 2Huaxi MR Research Center (HMRRC), Department of Radiology, West China Hospital of Sichuan university, Chengdu, China, People's Republic of

Synopsis

Using SVM to characterize structural and functional pattern implicated in antipsychotic-naive schizophrenia to predict the response of anti-psychotic treatment and also to examine the structural neuroanatomy and functional activity . And revealed that anatomical and functional changes revealed by Multi-modality MRI in first-episode schizophrenia patients before treatment showed the potential in predicting the one year treatment response, which could help psychiatrists to make treatment strategy for individual patient in future.

Purpose

Nearly 1/3 schizophrenia patients are not response to the antipsychotic treatment. While it is still a great challenge to characterize the treatment response in these patients before treatment. Previous imaging studies showed anatomical and functional deficits in schizophrenia patients before and after treatment1,2,3 . However, it is still unclear whether these neuroimaging features revealed before treatment have the potential to predict the response of antipsychotics in first-episode schizophrenia patients, which is the aim of this study.

Methods

The study was approved by the local research ethics committee, and written informed consent was obtained from all participants. Diagnoses of schizophrenia were determined by the consensus of two experienced clinical psychiatrists using the Structured Interview for the DSM-IV Axis I Disorder, Patient Edition (SCID). Psychopathology ratings were obtained using the Positive and Negative Syndrome Scale (PANSS). According to the reduction ratio of PANSS scores( (total PANSS scores of drug-naive patients - total PANSS scores of 1 year treatment patients)/ (total PANSS scores of drug-naive patients -30) ) after one year's antipsychotic treatment, we divide 40 patients into response group( 20 patients with reduction ratio>50%) and non-response group(20 patients with reduction ratio<50%). The age, gender and years of education of two groups were well matched, and all the patients recruited and scanned structural and resting-state fMRI at drug-naive states. The MRI examinations were performed on a 3-Telsa GE MRI system with an 8 channel phase array head coil. High resolution T1-weighted images were acquired with a volumetric three-dimensional spoiled gradient recall sequence (TR=8.5msec, echo time=3.4msec, flip angle=12°, slice thickness=1 mm) while the resting-state fMRI sensitized to changes in BOLD signal levels were obtained via a GE-EPI sequence (TR/TE=2000/30msec, flip angle=90°, slice thickness=5mm with no gap, 30 axial slices, 200 volumes in each run). VBM analyses of T1 images were performed using SPM8 (http://www.fil.ion.ucl.ac.uk/spm) and the VBM8 toolbox (http://dbm.neuro.uni-jena.de/vbm). While the ALFF maps were calculated using DPARSFA Version 2.2 (http://www.restfmri.net). Subsequently, SVM implemented in the PROBID software package (http://www.brainmap.co.uk/probid.htm) was used to investigate accuracy of whole brain structural and functional parameters in predicting response for treatment of schizophrenia. The combined classification accuracy was obtained by integrating the above two kernels into one model. Statistical significance of classification accuracy for each modality was set at P < 0.001 after permutation testing (1000 times).

Results

Demographics like age, gender and years of education were not significantly different between the two groups (p>0.05). All the subjects are right-handed. SVM allowed the classification of the two groups with diagnostic accuracy of GMV was 52.5% (P<0.001, sensitivity=55% specificity=50%), while the diagnostic accuracy of ALFF was 82.5% (P<0.001, sensitivity=85% specificity=80%)(Figure 1). The set of regions showed different value between the two groups mainly are right postcentral gyrus, right superior frontal gyrus, left premotor area, left hippocampus, right parahippocampal gyrus, left superior temporal gyrus and right cerebellar hemisphere for GMV(Figure 2) and right superior parietal lobule, right praecuneus, left lingual gyrus and left superior temporal gyrus for ALFF (Figure 3).The combination of the two kernel yielded an accuracy of 65% (P<0.001) with sensitivity and specificity up to 60% and 70% respectively. Receiver Operating Characteristic (ROC) curves of these modalities were also obtained as showed in Figure 4.

Discussion

To our knowledge, this is the first study to investigate the prediction of treatment outcome of schizophrenia using multivariate pattern analysis. Consistent with our hypothesis, both structural and functional MRI show potential value in differentiating baseline schizophrenia from response group and non-response group, though the accuracy is not high enough., The ALFF lead to better discrimination than GMV, supporting the notion that the functional changes are more like to be state-related in schizophrenia. Such findings also provide evidences to support the anatomical and functional deficits mainly involving frontal lobe, basal ganglia and hippocampus could be used to predict outcomes for schizophrenia, though the accuracy is not high enough.

Conclusion

Anatomical and functional changes revealed by Multi-modality MRI in first-episode schizophrenia patients before treatment showed the potential in predicting the one year treatment response, which could help psychiatrists to make treatment strategy for individual patient in future.

Acknowledgements

No acknowledgement found.

References

1. Paola Dazzan. Neuroimaging biomarkers to predict treatment response in schizophrenia: the end of 30 years of solitude? Dialogues Clin Neurosci. 2014 Dec;16(4):491-503.

2. Mourao-Miranda J, Reinders AA, Rocha Rego V, et al. Individualized prediction of illness course at the first psychotic episode: a support vector machine MRI study. Psychol Med. 2012;42:1037-1047.

3. Tiago Reis Marques, Heather Taylor, Chris Chaddock, et al.. White matter integrity as a predictor of response to treatment in first episode psychosis. Brain 2014: 137; 172–182.

Figures

Figure 1. Classification plot obtained from PROBID for the discrimination between response group and non-response group ALFF maps, yielding an accuracy 82.5%(sensitivity=85% and specificity=80%, P≤ 0.001).

Figure 2. The discrimination maps for GMV. These regions were identified by set the threshold to ≥ 30% of the weight vector scores. Warm color(positive value) indicated regions contributing to identifying the response group. White cool color(negative weights) indicates higher parameter values for non-response group than response group.

Figure 3. The discrimination maps for ALFF. These regions were identified by set the threshold to ≥ 30% of the weight vector scores. Warm color(positive value) indicated regions contributing to identifying the response group. White cool color(negative weights) indicates higher parameter values for non-response group than response group.

Figure 4. ROC curves for the comparisons between response and non-response group using the three kernels, which yielded an accuracy of 52.5% for the GMV (55% sensitivity, 50% specificity), 82.5% for the ALFF(85% sensitivity, 80% specificity) and 65% for the combination ( 60% sensitivity, 70% specificity), statistically significant at P<0.001.



Proc. Intl. Soc. Mag. Reson. Med. 24 (2016)
4129