A Machine Learning Framework for Predicting Cognitive Response to Prefrontal Transcranial Photobiomodulation in Bipolar Disorder Using FNIRS Data

Abstract

A \"smart\" framework utilizing machine learning to predict problems related to transcranial photobiomodulation (tPBM) therapy was developed to aid physicians in evaluating the likelihood that a patient diagnosed with bipolar disorder will respond favorably to tPBM based on functional near-infrared spectroscopy (fNIRS) (i.e., brain activity biomarkers). To aid in predicting whether or not a patient will experience a positive response to treatment, machine learning algorithms, such as XGBoost, were utilized. Additionally, a Fisher Score algorithm identifies key brain signal patterns from the huge amount of data collected during fNIRS tests for proper feature selection. Other data preprocessing techniques were performed to enhance accuracy through cleaning and normalization of the datasets before utilizing the appropriate algorithms for mechanical learning (e.g., machine learning management). A web-based interface was also created to allow access to clinicians by providing prediction outcomes and reasonably simple explanations for aiding clinician decision-making and developing personalized treatment plans and limiting unnecessary medical interventions. This machine learning-based predictive framework enhances the efficiency of management of individuals diagnosed with bipolar disorder by integrating the analysis of brain signals with XGBoost and automation methods.

Introduction

Bipolar disorder is a mental health condition characterized by alternating depressive and manic phases, which affect a person’s mood, energy, and daily functioning. Managing the disorder is challenging, as treatment responses vary among individuals, often requiring trial-and-error approaches that can delay effective care.

To address this, the study proposes a machine learning–based system to predict patient response to transcranial photobiomodulation (tPBM), a non-invasive light-based brain therapy. The system uses functional near-infrared spectroscopy (fNIRS) data, which measures brain activity through oxygenated and deoxygenated hemoglobin signals.

The methodology involves several stages: data collection, preprocessing (cleaning and normalization), feature extraction (e.g., mean, standard deviation, skewness, kurtosis), and feature selection using the Fisher Score method to retain only the most relevant features. These features are then used to train an XGBoost machine learning model, which predicts whether a patient will respond to tPBM therapy.

The system is implemented as a web-based platform with a user-friendly interface, allowing clinicians to upload patient data and receive quick, interpretable predictions. It also ensures secure data handling and supports visualization of results for easier decision-making.

Conclusion

The purpose of this study is to demonstrate the application of machine learning as a method of assisting with the treatment of individuals suffering from BPD via various methods: fNIRS brain signal data, pre-processing, extracting features from that data, and finally selecting important features for predicting response to therapy using the Fisher Score. Then, using the XGBoost machine learning algorithm to make predictions about whether or not an individual who has BPD is going to respond to tPBM therapy. The entire system has been developed to be able to use a web interface so that physicians can upload data to the system and obtain predictions without any reservations. Finally, this project will enable more effective and efficient clinical decisions to be made. Furthermore, this study presents evidence of how technology aids in providing better healthcare services through the provision of individualized care. It allows for predicting the treatment outcome before administering it, rather than just using a trial-and-error process, which saves time and avoids providing unnecessary treatment. The visualization and storage of the data enhances its utility and ease of understanding. Technology will be improved upon with continued data accumulation and development of improved algorithms; however, as seen in this project, there are already many indications that an improvement in performance will occur over time. The future of the approach will include being adapted into real medical environments. Ultimately, this project has positively impacted the effectiveness and quality of care provided to people with bipolar disorder.

References

[1] John D. Kelleher, Fundamentals of Machine Learning for Predictive Data Analytics, MIT Press, 2015. [2] Tianqi Chen and Carlos Guestrin, “XGBoost: A Scalable Tree Boosting System,” Proceedings of the ACM SIGKDD, 2016. [3] Artur M. R. Gomes et al., “Functional Near-Infrared Spectroscopy in Mental Health Research,” NeuroImage, 2019. [4] World Health Organization, “Bipolar Disorder,” 2023. [5] George B. Moody, “Biomedical Signal Processing and Analysis,” IEEE Press, 2017. [6] Isabelle Guyon, “An Introduction to Variable and Feature Selection,” Journal of Machine Learning Research, 2003. [7] Leo Breiman, “Statistical Modeling: The Two Cultures,” Statistical Science, 2001. [8] Hinton Geoffrey, “Reducing the Dimensionality of Data with Neural Networks,” Science, 2006. [9] Pinti Paola et al., “The Present and Future Use of Functional Near-Infrared Spectroscopy,” Neurophotonics, 2018. [10] IEEE, “Machine Learning in Healthcare Applications,” IEEE Journals, 2020.

Copyright

Copyright © 2026 Mrs. G. Rajeswari , P. Keerthika, K. Padma , T.T.R. Ranganayagi. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.