Congrats to Aarush Mehta from our lab on his first first-author publication! He started working with us as a high school student and is now a senior at
@stonybrooku.bsky.social. Thanks to coauthors:
Amir Nikzad, Yan Cong, Sunghye Cho, Sameer Pradhan
4/4

A negative relationship between Happiness and Thought Disturbance and a positive relationship between Anger and Hostile Suspiciousness were replicated across samples... 3/4

A negative relationship between Dominance and Anxious Depression symptoms and a positive relationship between Surprise and Hostile Suspiciousness were replicated across timepoints.... 2/4

🔍 By analyzing both linguistic and paralinguistic speech characteristics, our approach demonstrated robust disorder-specific learning across multiple groups, including those with sub-clinical psychotic experiences & major depressive disorder. 📊 #PrecisionMedicine 3/3

The machine learning model achieved 86% accuracy distinguishing conditions on the schizophrenia-bipolar spectrum using speech features. 2/3

Correcting the tag here ;) @dpn-journal.bsky.social. Thanks for publishing our work! What a great venue for research on digital psychiatry and neuroscience.

Many thanks to wonderful collaborators as always!! 6/6

@mspilka.bsky.social, Majnu John, @michaelbirnbaum.bsky.social, Ema Saito, @sarahberretta.bsky.social, @leilyb.bsky.social, ‪@myl1947.bsky.social‬, Anil Malhotra, Bill Simpson, John Kane @northwellhealth.bsky.social

Thought Disorder was particularly related to features describing subordinated constructions, efficient identification of picture elements, and decreased semantic distance between sentences. Negative Symptoms was related to features describing decreased speech complexity. 5/6

Examining one feature at-a-time, we can identify quite a few which predict each of the psychosis symptoms domains at a significance level that survives corrections for multiple comparisons. These relationships were largely robust to interactions with gender andrace. 4/6

A single component score representing the 357 speech and language features significantly predicted psychosis symptoms over time for all three domains. 3/6

We collected 160 sessions for 66 participants x 4 timepoints: 1) around the time of admission, 2) around discharge, 3) and 4) at 3 & 6 month follow up. The final feature set included 357 task-specific features representing a range of speech and language analyses. 2/6

Amazing!! Congrats!!