It is important to invest resources in research on mental health as it will help us to understand the best ways to treat and prevent mental ill health and also promote good mental health. Mental illnesses can range from anxiety, depression, bipolar disorder, Alzheimer, schizophrenia, autism and epilepsy etc. Given today’s powerful tools and technology, it is worthwhile to invest in using them for complex problems such as investigation of neuroimaging data for a better understanding of how our brain works and its underlying mechanisms responsible for our normal functioning. This can aid in creating better health solutions and techniques for diagnosis and therapeutic treatment thereby improving the quality of life. We can then offer people the very best treatment they deserve and make sure they have the best start in life.

What it does

This project generates brain subnetworks characteristics that are common across 6 subjects. It is achieved by processing EEG (Electroencephalography) data by machine learning techniques such as Principle Component Analysis (PCA) and Nonnegative Matrix Factorization (NMF) for deciphering brain mechanisms with penalty method. Technologies used: Python and MATLAB.

How I built it

Coherence between brain regions was first calculated in MATLAB. Then, by using PCA (in MATLAB) I learned how many subgraphs are possible (k). Finally, I applied NMF (in python) to the coherence data with k as the parameter to get k subnetwork/subgraph signature factor and temporal signature factor for each subgraph. These factors give information to help in analyzing brain subnetwork. Finally, I dealt with optimization and regularization parameter.

Challenges I ran into

The challenge was regularization parameter determination. In my mathematical model, one of the penalty parameter (beta that controls temporal signature factor) did not impact the frobenius norm error as compared to the second parameter (k that plays a key role in defining spatial signature factor) while searching for the minima with the cost function for optimization. I had to plot and analyze the error for varied values of k and beta during NMF analysis. Also, during PCA, it was hard to determine the number of components for which I had to do variance analysis with my application’s physics in mind.

Accomplishments that I'm proud of

I am proud to have manage and dealt with the regularization parameter determination complexity. I am glad that I was able to apply machine learning in this unique and creative way. I am grateful to have the opportunity to network and learn so much from this hackathon event. Most important of all, I am glad that I can contribute towards better mental health cause by characterizing neural signatures.

What I learned

I learned to deal with analysis and machine learning challenges as described in the “Challenges I ran into” section. Such analysis was new for me and good to learn. These findings helped me better understand a new facet about mathematical analysis and played a key role in helping me summarize my results. Also, I learned more about python.

What's next for Brain Learning with Machine Learning## Inspiration

I look forward to discussing my results with medical experts in further details for potential clinical applications.

Built With

Share this project: