Major Depressive Disorder (MDD) is a debilitating mental health condition that affects millions of people globally. Characterized by persistent feelings of sadness, hopelessness, and a loss of interest in daily activities, MDD can have a significant impact on an individual’s ability to function in daily life. Other common symptoms include significant changes in appetite, sleep disturbances, feelings of guilt or worthlessness, and, in some severe cases, thoughts of self-harm or suicide.
As one of the most prevalent mental health disorders, MDD has a profound impact on both the individuals suffering from it and society as a whole. The condition often leads to reduced productivity, impaired relationships, and a decreased quality of life. Thankfully, there are numerous treatment options available for MDD, including medications, psychotherapies, and a combination of both. However, these treatments are not universally effective, and the journey to finding the most suitable treatment for an individual often involves a lengthy trial-and-error process.
This unpredictability in treatment response is one of the key challenges in treating MDD. A medication or therapy that works for one person may not be effective for another, even if they have similar symptoms. As a result, determining the right treatment approach for each patient can take time and requires careful monitoring. Given the high emotional and financial cost of prolonged treatment trials, there is a growing interest in developing methods to predict which treatments will be most effective for an individual early in their treatment journey.
Recent research, particularly a groundbreaking study conducted by scientists at the National University of Singapore, has explored the potential of using functional near-infrared spectroscopy (fNIRS) combined with machine learning techniques to predict how patients with MDD will respond to different treatments. This study represents a promising step forward in personalized medicine for depression, with the goal of helping doctors more efficiently select the optimal treatment for their patients.
The Importance of Predicting Treatment Response in MDD
The effectiveness of treatments for MDD varies widely between individuals, and there is no “one-size-fits-all” approach. While medications such as antidepressants and psychotherapies like cognitive-behavioral therapy (CBT) can be beneficial for many people, they do not work for everyone. Some patients experience minimal improvement, while others may experience adverse side effects or a worsening of symptoms. Therefore, identifying the treatment that is most likely to work for a specific individual is critical to improving the overall success of mental health interventions.
The ability to predict a patient’s response to different treatments could drastically shorten the time needed to find an effective therapy. This would not only improve outcomes for patients but also reduce the emotional and physical toll of trying multiple ineffective treatments. Furthermore, such predictive tools could help healthcare providers optimize their treatment protocols, ensuring that patients receive the most suitable therapies from the start.
The Role of fNIRS and Machine Learning in Predicting Treatment Response
In their study, the researchers from the National University of Singapore and collaborating institutes used a combination of functional near-infrared spectroscopy (fNIRS) and machine learning algorithms to investigate how well these tools could predict treatment responses in individuals with MDD.
fNIRS is a non-invasive brain imaging technique that measures changes in blood oxygenation levels in the brain. By using near-infrared light, fNIRS can detect the concentration of oxygenated and deoxygenated hemoglobin in various brain regions. This information provides insights into brain activity and can help researchers understand how different regions of the brain respond to various stimuli or treatments. The dorsolateral prefrontal cortex (dlPFC), in particular, is a brain region involved in executive functions such as cognitive flexibility, working memory, and decision-making—functions that are often impaired in individuals with depression.
The study combined fNIRS data with clinical information gathered from patients during assessments, such as the Hamilton Depression Rating Scale (HAM-D), a widely used tool for evaluating the severity of depressive symptoms. Using machine learning techniques, the researchers aimed to identify biomarkers that could predict how patients would respond to different treatment options for MDD.
Study Design and Findings
The study involved 70 participants who had been diagnosed with MDD. These participants underwent treatment over a six-month period, with their treatment responses measured using the HAM-D scale, which provides a numerical score based on the severity of symptoms. The goal of the research was to determine if fNIRS data, combined with clinical assessment data, could be used to predict which patients would respond positively to treatment and which would not.
The researchers employed several machine learning models to analyze the data, including support vector machines, random forests, XGBoost, discriminant analysis, Naïve Bayes, and transformers. The results were encouraging: the fNIRS data, when analyzed alone, were able to predict treatment responses with a reasonable degree of accuracy.
The primary finding of the study was that changes in the concentration of hemoglobin in the dorsolateral prefrontal cortex (dlPFC) were significantly correlated with how patients responded to treatment. Specifically, the task-related changes in total hemoglobin (HbT)—the difference in hemoglobin concentration before and after a treatment task—appeared to be a reliable biomarker of treatment response.
One of the key findings was that the Naïve Bayes model, which was used to predict treatment responses, performed well when trained on fNIRS data alone. The model’s accuracy, as measured by balanced accuracy (bAcc) and area under the curve (AUC), was 70% (bAcc) and 0.77 (AUC) for inner cross-validation and 73% (bAcc) and 0.77 (AUC) for outer cross-validation. These findings suggest that fNIRS data alone can provide valuable insights into the likelihood of treatment success for patients with MDD.
Interestingly, when the researchers combined fNIRS data with clinical information, the performance of the predictive model actually decreased. The bimodal model (combining both fNIRS and clinical data) achieved a 68% balanced accuracy and an AUC of 0.70 in outer cross-validation, which was lower than the performance of the fNIRS-only model. This suggests that while clinical data are important, fNIRS may provide stronger predictive power on its own for determining treatment responses.
Implications of the Findings
The findings from this study have significant implications for the treatment of MDD. If future research can replicate these results, it may become possible to use fNIRS and machine learning models as part of clinical practice to predict which treatments will be most effective for a given patient. This would reduce the need for patients to undergo multiple rounds of trial-and-error treatment, ultimately leading to faster recovery and fewer setbacks.
Moreover, the study highlights the potential for biomarkers in predicting treatment response. By identifying specific brain activity patterns associated with therapeutic outcomes, clinicians may be able to more accurately select the appropriate treatment for each patient. This would not only improve the patient’s chances of recovery but also help optimize resource allocation in healthcare systems.
Future Directions and Challenges
While the study represents a promising step toward personalized treatment for MDD, there are still challenges to overcome. One of the primary limitations of the study is its relatively small sample size, which may limit the generalizability of the findings. Additionally, while fNIRS is non-invasive, it may not capture all the complexities of brain activity that contribute to treatment responses. Further research is needed to explore whether other biomarkers—such as those related to inflammation or genetic factors—might further improve predictive accuracy.
Another challenge lies in integrating these findings into clinical practice. While machine learning models show great promise in research settings, incorporating them into everyday clinical workflows will require careful validation, collaboration between clinicians and researchers, and the development of user-friendly tools that can be used by healthcare providers.
Conclusion
The study by researchers at the National University of Singapore represents an exciting advance in the search for more effective, personalized treatments for Major Depressive Disorder. By combining functional near-infrared spectroscopy with machine learning techniques, the researchers have demonstrated the potential for identifying biomarkers that could predict treatment responses. As research in this area progresses, it could lead to more efficient, tailored treatment strategies that help patients recover more quickly and effectively from depression. With continued advancements in this field, the future of mental health care looks promising, with the potential for more personalized and data-driven approaches to treatment.
Reference: Su Hui Ho et al, Application of functional near-infrared spectroscopy and machine learning to predict treatment response after six months in major depressive disorder, Translational Psychiatry (2025). DOI: 10.1038/s41398-025-03224-7