Could Artificial Intelligence Help With Early Diagnosis of Schizophrenia?

Schizophrenia is a notoriously challenging mental illness to diagnose, as its symptoms vary widely from patient to patient and are often subtle. Clinicians rely on their expertise and observation of patients' speech patterns to make a diagnosis, but this subjective approach leads to significant delays in treatment, with an average of 18 months between the onset of symptoms and a proper diagnosis.

Recently, researchers have begun exploring whether artificial intelligence (A.I.) could improve early diagnosis and care for schizophrenia by analyzing speech patterns that are difficult for clinicians to detect.

A team of researchers in the Netherlands trained an A.I. program to analyze 88 different audio features, such as volume, pauses, pronunciation, and intonation, in recordings of patients diagnosed with schizophrenia. When tested on new audio files, the A.I. correctly identified patients with schizophrenia from healthy controls with 86.2% accuracy, even distinguishing between different subtypes of the disorder.

While this approach may not be useful for clear-cut cases, it shows promise for detecting early-stage or high-risk patients and predicting relapse.

Another group of researchers took a different approach by examining the content of patients' speech rather than its sounds. They built a machine learning model that assigned a unique "address" to each word in a patient's conversation transcript, allowing the A.I. to measure how meaning flows throughout a patient's speech. When applied to transcripts of patients with and without schizophrenia, the A.I. achieved 87% accuracy in distinguishing between the two groups, significantly outperforming clinical raters, who were only 68% accurate.

Beyond diagnosis, A.I. could also play a role in routine monitoring of patients' symptoms over time, providing more precise and responsive care at a lower cost, and even enabling remote monitoring. As psychiatrist Thomas Insel notes, "We have the tools to do that with the kind of precision that we have never had before."