Although bipolar disorder (BD) is a relatively frequent, severe and disabling disease, we still lack reliable diagnostic tools. The existence of anatomo-functional brain abnormalities in BD is now well established by MRI studies [1Phillips M.L., Swartz H.A. A critical appraisal of neuroimaging studies of bipolar disorder: toward a new conceptualization of underlying neural circuitry and a road map for future research Am J Psychiatry 2014 ;  171 : 829-843 [cross-ref]

Cliquez ici pour aller à la section Références]. Machine learning allows individual level analysis and could potentially lead to diagnostic applications [2Rajpurkar P., et al. CheXNet: radiologist-level pneumonia detection on chest X-rays with deep learning 2017 ;

Cliquez ici pour aller à la section Références]. This is the first international multi-site ML study on BD aggregating raw MRI data. We trained two support vector classifier (SVC), with linear kernel and with radial basis functions kernel, using as input features T1 MRI regions of interest volumes from 419 patients BD and 551 healthy controls (HC) from 13 international sites. We assessed those classifiers using two strategies, 5 fold cross validation and leave-one-site-out. Both methods reached similar results, respectively AUC=61.70%/61.65%, sensitivity=57.52%/60.14% and specificity=65.88%/63.16%. The regions that contributed to identification of BD participants were highly correlated between both analyses (r2=0.9479). These results point to robust, generalizable models. Classification was affected by antipsychotics treatment, but this does not seem to drive classification performances of BD vs HC. Using voxel-based morphometry features may increase the classifiers performances [3de Pierrefeu A., et al. Identifying a neuroanatomical signature of schizophrenia, reproducible across sites and stages, using machine learning with structured sparsity Acta Psychiatr Scand 2018 ;  138 : 571-580 [cross-ref]

Cliquez ici pour aller à la section Références]. To reach clinical meaningful performance (i.e. 80%), the continuation of collaborative work to get larger multicentric databases is essential and must be extended to multimodal data [4Besga A., et al. Discrimination between Alzheimer's disease and late onset bipolar disorder using multivariate analysis Front Aging Neurosci 2015 ;  7 : 231

Cliquez ici pour aller à la section Références].