Can machine learning tell bipolar disorder from BPD?

Five studies, 591 participants: models beat traditional methods, but were far better at spotting bipolar than borderline.

This review examined whether machine learning can help distinguish bipolar disorder from borderline personality disorder—two conditions that are often confused in clinical practice. Out of 60 studies screened, only five met the criteria, covering data from 591 participants. Across these studies, machine‑learning models showed moderate to strong accuracy, performing better than traditional diagnostic methods. However, they were far more accurate at identifying bipolar disorder than borderline personality disorder, highlighting the ongoing difficulty of diagnosing BPD. Models using brain‑imaging features reached about 80% accuracy, and smartphone‑based mood data could differentiate BD, BPD, and healthy controls with about 75% accuracy. Overall, the findings suggest that machine learning holds promise for improving diagnostic precision, but more research is needed before these tools can be reliably used in clinical settings.