A Habenula Neural Biomarker Simultaneously Tracks Weekly and Daily Symptom Variations during Deep Brain Stimulation Therapy for Depression

Shi Liu, Yu Qi, Shaohua Hu, Ning Wei, Jianmin Zhang, Junming Zhu, Hemmings Wu, Hailan Hu, Yuxiao Yang, Yueming Wang

This research paper presents a significant step toward "smarter" brain implants for mental health. Researchers identified a personalized neural biomarker in the lateral habenula (LHb) that simultaneously tracks depression symptoms across both daily and weekly timescales. By applying machine learning to brain signals from a patient undergoing Deep Brain Stimulation (DBS), the team developed a model that accurately predicts symptom severity and responds dynamically to treatment adjustments.

TL;DR

Researchers have identified a neural biomarker in the lateral habenula that tracks both the slow, weekly recovery and the rapid, daily mood fluctuations of a patient with treatment-resistant depression. This "dual-scale" biomarker provides a high-accuracy feedback signal that could enable future adaptive DBS systems to automatically adjust therapy in real-time.

What Problem Does This Paper Address?

Major Depressive Disorder affects over 300 million people, and roughly 30% of them are "treatment-resistant," meaning standard drugs and therapy don't work. Deep Brain Stimulation (DBS)—essentially a pacemaker for the brain—is a promising alternative, but its effectiveness varies wildly between patients.

The current "gold standard" for DBS is continuous, static stimulation. However, depression symptoms aren't static; they fluctuate over months (slowly) and hours (quickly). To improve outcomes, researchers are pursuing adaptive DBS (aDBS), which would only "fire" or adjust its parameters when the brain actually needs it. The missing piece of the puzzle is a reliable biomarker: a specific neural signal that acts as a proxy for how the patient is feeling. While biomarkers have been found in other brain regions, we haven't had one that can track both slow and fast symptom changes simultaneously in the lateral habenula (a key "anti-reward" hub in the brain).

Methods

The study followed a single patient with treatment-resistant depression over a 41-week LHb-DBS treatment course.

  • Data Collection: The researchers recorded daily Local Field Potentials (LFP)—the electrical "hum" of a population of neurons—from the LHb for 30 minutes a day. Concurrently, the patient provided daily mood self-reports (Visual Analogue Scale), and psychiatrists performed weekly clinical ratings (HAMD/MADRS).
  • Feature Extraction: They extracted 24 different features from the brain signals, including 10 spectral features (like power in specific frequency bands) and 14 temporal features (which describe how the signal changes over time, such as complexity and "memory").
  • Machine Learning: The team tested six different machine learning models to see which could best classify "high" vs. "low" symptom states. Logistic Regression (LR) emerged as the winner due to its high accuracy and relative simplicity.
  • Validation: They used Dynamic Time Warping (DTW) to measure how well the biomarker's "predictions" matched the actual timing of the patient's symptom variations.

Datasets

The study utilized a unique, longitudinal dataset collected from a single male patient (aged 36-40) over 41 weeks.

  • Neural Data: 122 days of LHb LFP recordings.
  • Clinical Scales: Weekly assessments using the Hamilton Depression Rating Scale (HAMD) and Montgomery-Asberg Depression Rating Scale (MADRS).
  • Self-Reports: Daily scores for depression (VAS-D) and anxiety (VAS-A).

Key Insights and Ideas

  • The Power of Temporal Features: Most previous research focused on spectral features (like "Alpha" or "Beta" waves). This study found that temporal features, specifically the Hurst exponent (a measure of a signal's "long-term memory" or autocorrelation), were the strongest predictors of depression.
  • Dual-Scale Tracking: This is one of the first biomarkers shown to track both the "slow" clinical trajectory of recovery over months and the "fast" daily ups and downs. This suggests the LHb is a critical node for both long-term mood stability and immediate emotional response.
  • Depression vs. Anxiety: Interestingly, the biomarker was highly specific to depression. While it predicted depression scores (VAS-D) with high accuracy, it failed to track anxiety scores (VAS-A), suggesting the LHb's role in this patient was specifically tied to the "anti-reward" depressive circuit rather than general distress.
  • Dose-Response Sensitivity: The biomarker wasn't just a passive observer; it changed when the researchers altered the DBS frequency (moving from 1Hz to 20Hz to 130Hz), proving it can serve as a feedback signal for tuning stimulation parameters.

Key Results

  • Classification Accuracy: The model distinguished between high and low symptom states with a cross-validated accuracy of 0.97.
  • Tracking Performance: The biomarker explained 74% of the variance in weekly clinical ratings and 63% of the variance in daily self-reports.
  • Clinical Outcome: By the end of the 41-week study, the patient achieved full clinical remission (HAMD score dropped from 20 to 7).
  • Frequency Effects: The data suggested that 1Hz stimulation was ineffective, while 20Hz and 130Hz led to significant symptom improvement, which was reflected in the biomarker's value.

Limitations and Future Work

  • N-of-1 Study: As a single-patient study, these results are a "proof of concept." The specific features (like the Hurst exponent) might vary between individuals, requiring personalized models for each patient.
  • Stimulation Artifacts: LFP signals were recorded with the stimulation turned off to avoid electrical noise. For a truly "closed-loop" system, researchers need better ways to filter out noise so they can record and stimulate at the same time.
  • Frequency Range: Due to hardware noise, the study only looked at signals below 30Hz. Higher frequency bands (Gamma) might contain even more useful information.
  • Next Steps: The authors suggest that future work should focus on multi-channel recordings and integrating these biomarkers into real-time adaptive DBS devices.

Summary updated

Abstract

Abstract Deep brain stimulation (DBS) targeting the lateral habenula (LHb) is a promising therapy for treatment-resistant depression (TRD) but its clinical effect has been variable, which can be improved by adaptive DBS (aDBS) guided by a neural biomarker of depression symptoms. A clinically-viable neural biomarker is desired to classify depression symptom states, track both slow and fast symptom variations during the treatment, respond to DBS parameter alterations, and be neurobiologically interpretable, which is currently lacking. Here, we conducted a study on one TRD patient who achieved remission following a 41-week LHb DBS treatment, during which we assessed slow symptom variations using weekly clinical ratings and fast variations using daily self-reports. We recorded daily LHb local field potentials (LFP) concurrently with the reports during the entire treatment process. We then used machine learning methods to identify a personalized depression neural biomarker from spectral and temporal LFP features. The identified neural biomarker classified high and low depression symptom severity states with a cross-validated accuracy of 0.97, with the most contributing spectral and temporal feature being LFP beta band power and Hurst exponent, respectively. It further simultaneously tracked both weekly (slow) and daily (fast) depression symptom variation dynamics, achieving test data explained variance of 0.74 and 0.63, respectively. It also responded to DBS frequency alterations. Finally, it can be neurobiologically interpreted as indicating changes in LHb excitatory and inhibitory balance during DBS treatment. Together, our results hold promise to identify clinically-viable neural biomarkers to facilitate future aDBS for treating TRD.
BibTeX
@article{liu2025habenula,
  author = {Shi Liu and Yu Qi and Shaohua Hu and Ning Wei and Jianmin Zhang and Junming Zhu and Hemmings Wu and Hailan Hu and Yuxiao Yang and Yueming Wang},
  title = {{A Habenula Neural Biomarker Simultaneously Tracks Weekly and Daily Symptom Variations during Deep Brain Stimulation Therapy for Depression}},
  journal = {IEEE J. Biomed. Health Informatics},
  year = {2025},
  doi = {10.1109/jbhi.2025.3566057},
}
Figure