Separation of patients with schizophrenia and bipolar disorder based on MRI scans: Can machine learning aid in clinical diagnosis?
-
1
Brain Center Rudolf Magnus, University Medical Center Utrecht, Psychiatry, Netherlands
Background: The question of whether brain changes in schizophrenia and bipolar disorder are similar is not only relevant regarding their possible genetic and biological overlap; it also has clinical and diagnostic implications. Here we address the issue of whether structural MRI brain scans can be used to differentiate the two disorders. In a recent study we built a structural-MRI-based schizophrenia classification model and tested its predictive capacity in an independent test sample [1]. Using two large data sets, we confirmed the feasibility to use structural MRI for individualized prediction whether a subject is a schizophrenia patient or a healthy control, with an accuracy of 70.4%. Although scientifically interesting, the clinical use is limited: these classification models become really useful if they can predict a subject’s future status, or its current status if this cannot be determined by other means.
Methods: Structural magnetic resonance 1.5 T whole brain images of 66 patients with schizophrenia, 66 with bipolar disorder, and 66 healthy controls were segmented and further processed to create gray matter density (GMD) maps, reflecting local gray matter tissue presence [2]. These GMD maps were used to train linear Support Vector Machines (SVM) [3]) to separate the three groups (Fig. 1). The validity, or generalizibility, of the models was first tested by leave-one-out cross-validation. Secondly, the models were applied without change to an independent data set acquired on a 3 T scanner that included 46 schizophrenia patients, 47 patients with bipolar disorder and 43 healthy control subjects.
Results: Schizophrenia patients could be correctly classified versus healthy subjects with an accuracy of 90%; they could be differentiated from bipolar patient with the same level of accuracy, i.e. also 88%. The model separating bipolar patients from healthy control subjects performed worse: 67% of the healthy subjects were correctly classified and only 53% of the bipolar patients (Fig. 1). Application of the 1.5 T models on the 3 T validation set yielded average classification accuracies of 76% (healthy vs schizophrenia), 66% (bipolar vs schizophrenia) and 61% (healthy vs bipolar).
Conclusion: We demonstrated the feasibility to use structural MRI for individualized prediction whether a subject is a schizophrenia patient or not, with an accuracy of 88%. In an independent validation set acquired on a scanner with different field strength the unaltered models performed significantly above chance level, with accuracies of 76% (schizophrenia vs healthy) and 66% (schizophrenia vs bipolar). While the use of MRI to separate schizophrenia patients from healthy subjects has limited clinical value, the accurate separation of schizophrenia patients from bipolar patients could become a diagnostic aid for psychiatrists. The results also indicate that the gray matter pathology differs between schizophrenia and bipolar disorder to such an extent that they can be reliably differentiated using machine learning paradigms.
References
1. Nieuwenhuis et al (2012), Neuroimage 61:606-612.
2. Ashburner and Friston (2000), NeuroImage 11:805-821.
3. Vapnik (1999), IEEE Trans Neural Netw 10:988–999.
Keywords:
machine learning,
MRI,
Schizophrenia,
Bipolar Disorder,
Diagnosis, Computer-Assisted
Conference:
Neuroinformatics 2014, Leiden, Netherlands, 25 Aug - 27 Aug, 2014.
Presentation Type:
Poster, to be considered for oral presentation
Topic:
Neuroimaging
Citation:
Schnack
H,
Nieuwenhuis
M,
Van Haren
NE and
Kahn
R
(2014). Separation of patients with schizophrenia and bipolar disorder based on MRI scans: Can machine learning aid in clinical diagnosis?.
Front. Neuroinform.
Conference Abstract:
Neuroinformatics 2014.
doi: 10.3389/conf.fninf.2014.18.00064
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
25 Apr 2014;
Published Online:
04 Jun 2014.
*
Correspondence:
Dr. Hugo Schnack, Brain Center Rudolf Magnus, University Medical Center Utrecht, Psychiatry, Utrecht, Netherlands, h.schnack@gmail.com