The future of psychiatry is here!
In 1973, David Rosenhan published the results of his famous work “On Being Sane in Insane Places”, criticizing the reliability psychiatric diagnosis, in the journal Science.
It is clear that we cannot distinguish the sane from the insane in psychiatric hospitals. The hospital itself imposes a special environment in which the meanings of behavior can easily be misunderstood. The consequences to patients hospitalized in such an environment — the powerlessness, depersonalization, segregation, mortification, and self-labeling — seem undoubtedly counter-therapeutic.
In one part of his study, Rosenhan asked neurotypical individuals to fake auditory hallucinations in order to get admitted to different psychiatric hospitals. Twelve hospitals were involved in five different states. Despite them faking their symptoms, all these pseudopatients were admitted and diagnosed with schizophrenia. Once they became in-patients, they started to act as if they didn’t have any symptoms and they informed the staff that they no longer have auditory hallucination. Nonetheless, the hospital forced them to admit to having psychiatric illness, diagnosed them with schizophrenia “in remission”, and they were allowed to be discharged only if they agreed to take anti-psychotics. What Rosenhan wanted to communicate is that relying on subjective criteria to diagnose mental illness is problematic.
The current diagnostic model in psychiatry is not highly reliable due to several factors including the patient’s ability to provide reliable information, psychological state, and clinical presentation. Physicians also play a role in this, as different physicians might have varying opinions on the same case. All of these factors, which the diagnostic decision is based on, are quite subjective, and thus can’t pinpoint the diagnosis consistently and reliably. As nosology is a key aspect of psychiatry on which patient assessment and treatment options are based, it is essential to have a layer of appraisal centered around objective evaluation of the patients to establish a more reliable diagnostic decision.
What if instead of relying on patients’ self-report we can get this information directly by scanning their brains?
This is what our latest paper published in Translational Psychiatry titled “Machine learning classification using neuroimaging data in schizophrenia, autism, ultra-high risk and first episode psychosis” investigated.
What did we want to know?
•Can we distinguish between individuals with autism spectrum disorder (ASD), schizophrenia and typically developing (TD) individuals based solely on their MRI scans?
•What are the most informative brain feature groups that contributed to the classification?
•Are we able to assess the brain patterns of the “intermediate phenotype” (Ultra-high risk for psychosis (UHR) and first episode psychosis (FEP) subjects)?
•Do UHR or FEP individuals carry brain neural signatures similar to those observed in a more severe patient population such as schizophrenia?
•Are the classifiers used clinically relevant?
How we did it -I
We collected brain MRI scans from ASD, schizophrenia and TD individuals. We then extracted brain features from those scans and made the classifiers learn to identify each of the three groups based on those features. Once the classifiers learned, we asked them to look at new brain scans that they haven't been exposed to before, to try to classify them into those three groups.
What did we find out?
The algorithms were able to correctly classify those new scans with high accuracy.
How we did it -II
After examining the above results, we chose the algorithms that had the highest accuracy of classification, and we asked those trained classifiers to look at yet another batch of scans which belonged to FEP or UHR individuals. We wanted to know which brain signatures are these individuals closer to, TD, ASD or schizophrenia?
What did we find out?
As expected, the majority of these scans (FEP & UHR) were classified as schizophrenia. This potentially means that the brain signature of schizophrenia exists in early stage subjects long before symptom onset.
Is it possible to have a neural pattern similar to schizophrenia, but not show the same symptoms?
Yes! Schizophrenia Spectrum Disorder (SSD) includes schizophrenia, schizophreniform disorder, schizoaffective disorder, delusional disorder, brief psychotic disorder, and schizotypal personality disorder. These have overlaps, more so than differences and an individual who is at high risk might have neural correlates that are more similar to SSD than a typically developing brain.
Interestingly, despite some overlap between SSD and ASD, our classifiers did not classify any of the UHR or FEP individuals into ASD, at least based on the features we used.
There is much to be done, and it will take a village for us to get to where we hope to be. Luckily, the National Institute of Mental Health understands the challenge, and has recently launched an endeavor titled Accelerating Medicines Partnership — Schizophrenia (AMP SCZ) that could help take us there.
To learn more about schizophrenia, psychosis, autism and other neuropsychiatric or neurodevelopmental disorders visit NIMH’s Mental Health Information.
Paper reference
Yassin, W., Nakatani, H., Zhu, Y. et al. Machine-learning classification using neuroimaging data in schizophrenia, autism, ultra-high risk and first-episode psychosis. Transl Psychiatry 10, 278 (2020). https://doi.org/10.1038/s41398-020-00965-5
