Algorithmic methods helps make brain scanning for schizophrenia more accurate

Schizophrenia is a serious mental disorder that causes hallucinations, delusions and disordered thought. Brain-scanning methods have revealed differences in healthy controls from people with the condition, but they are not yet accurate enough to inform diagnosis. Professor Joseph Kambeitz from Ludwig-Maximillians University introduces us to new algorithmic methods that could help change this.

Although it is often mistakenly associated with multiple-personality disorders, schizophrenia is in fact linked with abnormal social behaviour and a failure to recognise reality. Common symptoms include false beliefs and auditory hallucinations. WHO estimates that schizophrenic disorders affect around 26 million people worldwide and result in moderate or severe disability in 60% of cases. However, after more than a century of research on the disease, much still remains unknown.

"Hallucinations can affect any modality, but it's usually auditory ones in the form of voices commenting on subjects' behaviour or even giving orders. People can also suffer from delusional ideas; they feel persecuted, that people are observing them or trying to harm them," explains Professor Joseph Kambeitz from the Department of Psychiatry at Ludwig-Maximillians University in Munich, Germany.

"Subjects tell us they have a feeling that their surroundings seem somehow not real, that the people around them are not genuine people but actors in a movie."

Schizophrenia affects men and women with the same prevalence, and usually strikes in late adolescence. Some patients only have a single episode of the condition throughout their lives while others suffer from reoccurring events of psychotic symptoms.

The condition's cause remains largely debated in the scientific community. Many neurobiologists assume there is an environmental component at play, but there is also some evidence for genetic factors having a role in the pathology. One major challenge is that many patients with schizophrenia show very different symptoms from each other, so it is difficult to know if the condition is actually a constellation of different disorders that should be sub-classified - it's consequently quite hard to diagnose unless the patient shows obvious symptoms.

Modes of investigation
Kambeitz splits himself professionally between taking care of patients in the psychiatric ward and also conducting research into psychosis - a broad term that can refer to anything from relatively normal aberrant experiences through to complex and catatonic expressions of schizophrenia and type 1 bipolar disorder.

He reveals that over the past couple of decades, several studies have indicated significant differences in brain structure between healthy controls and schizophrenia patients.

"There is very little doubt that brain abnormalities in patients with schizophrenia exist compared with healthy controls," he says. "For instance, when you use imaging techniques like PET to investigate different neurotransmitter systems, you see a difference in patients with schizophrenia and people without."

He explains there is good evidence that people with schizophrenia produce too much dopamine - a chemical released by nerve cells that plays a major role in reward-motivated behaviour. Typical treatment includes antipsychotics, which block the action of this chemical. The main effect of such drugs is to reduce the so-called 'positive' symptoms of the disorder: the delusions and hallucinations.

Other imaging techniques, like MRI, have also shown structural abnormalities between the two groups. In some parts of the brain, grey matter is increased, but it is also found to be lacking in others.

"There is probably not one individual region that is affected in schizophrenia but rather a network of different regions," says Kambeitz.
He points out that another MRI technology called resting-state functional imaging (rsfMRI) has shown interesting findings when comparing patients with schizophrenia and healthy controls.

"Subjects are put in a scanner, and instructed to just lie still, have their eyes open and daydream. At the same time, we measure with functional MRI sequence the blood contrast in the brain. What you usually find as well in those studies is that people with schizophrenia show abnormalities in this resting state," he explains.

From group to individual
Even though imaging results do indicate differences in brain structure between healthy controls and schizophrenia patients, a substantial overlap is usually observed between the groups. Consequently, as yet, such procedures cannot be used for diagnosing the condition.

"I think that our general approach of looking at group-level statistics does not allow us to make claims on the basis of one individual patient, so that is the problem so far," explains Kambeitz. "As most studies show, it is not one individual brain region that is affected in patients with schizophrenia, so if you just look at one, this is probably not sufficient information to guide the diagnostics in the everyday clinic.

"You'd have to take into account the whole pattern of functional and structural changes in the brain of one patient to really make a statement that is clinically meaningful."

To overcome such drawbacks, an increasing number of studies have applied new statistical approaches to the analysis of brain alterations in patients with schizophrenia. Such trials indicate that patterns of subtle structural and functional changes can be highly distinctive of the condition, even though each individual component within these patterns might not be.

So Kambeitz and his team conducted a meta-analysis of similar investigations in order to elucidate the performance of neuroimaging in distinguishing patients with schizophrenia from healthy controls. They analysed 38 studies including a total of 1,602 patients with schizophrenia and 1,637 people without the condition - 20 studies used structural MRI, 11 used rsfMRI, four implemented MRI, three used PET and one used diffusion tensor imaging.

"This approach has been introduced in recent years to neuroimaging data analysis, and it comes from computer science," explains Kambeitz. "And it has certain advantages. First, it is not residual analysis to a certain brain region but allows you to take into account the whole pattern of abnormality across the brain."

This way, all the data can be made use of, and the algorithms can be applied to any data modality from PET to MRI. The algorithms are flexible, allowing you to identify brain patterns that can inform a diagnostic decision on the individual subject's level.

"What you usually do is take a group of patients and healthy controls, and try to extract the brain pattern that is informative for your question. You use this statistical model and apply it to an individual, and you get a diagnostic probability, which tells you the likelihood of the patient having schizophrenia," explains Kambeitz.

In response
Results indicate the models had a sensitivity and specificity of 80.3%. Interestingly, the tools were more sensitive in older people and more specific for patients on medication. Sensitivity was also heightened in subjects with chronic schizophrenia compared with those who had only experienced one episode. Studies using rsfMRI were found to be the most effective.

"It shows that the older the patients are, the more pronounced their brain structural or functional abnormalities are, and the more likely such models classify the patients correctly," Kambeitz clarifies. "So this probably means that the models we're investigating in the study perform better in older patients because the brain changes are more pronounced than younger ones."

He says it also indicates that medication probably affects those brain patterns, and that patients who receive the most therapeutics have the most structural and functional brain abnormalities.

The real question though, of course, is whether the results Kambeitz found are sufficient for implementing such tools in the clinic.
"Terribly, the answer is no," he says. "So we still need to improve those models to really help us, and it is also important to note that it depends in which context you use these tools."

He points out that if you work in a clinical psychiatric setting, the prevalence of schizophrenia might be very high, so the models could work well there but probably not for screening a large population where the condition is fairly rare.

"The patients investigated in our meta-analysis would have had quite a severe form of schizophrenia," emphasises Kambeitz, "and a clinician would probably argue that, just by talking to a patient and exploring them clinically, it is easy to differentiate such a patient from a healthy control."

But he also believes the application for such models might be in determining the type of schizophrenia and predicting disease outcome, something that's much trickier clinically.

"If you have a patient in front of you and you know that they're suffering because of some symptoms, but you're not sure whether it is a depression with some schizophrenic symptoms or if it's actually schizophrenia - we see a clinical application here and, also, the models might be a good prediction of the outcome for the patient," explains Kambeitz.

"We so often have a patient but we are not sure how it's going to develop in the future, whether it's going to get better spontaneously or whether they are going to get worse."

The tools could also be useful for determining the correct treatment for patients. A common phenomenon in psychiatry is heterogeneity in response to medication: some people respond well and recover while others experience side effects.

Psychiatrists end up having to try lots of different types of tablets for their patients, which is frustrating for both parties. Kambeitz hopes the algorithmic method of investigation will one day be used to predict who is going to respond best to a particular drug and remove the guessing-game scenario.

The next step for the team is to complete a large multicentre European study that is looking into extending the algorithms and combining different neuroimaging modalities to make them more specific for patients with schizophrenia. Kambeitz is optimistic that imaging will one day prove to be an effective diagnostic tool for the condition.

"I think it will be helpful - I'm sure about that. My feeling is that it will probably have the most impact in predicting treatment response," he concludes.