Identifying Auditory Biomarkers for ALS

In the status quo, patients with different neurological disorders, such as ALS and Parkinson’s, are usually only diagnosed in the later stages of the disease primarily because the symptoms are often confused with common voice difficulties. Additionally, more than 90% of ALS diagnosed patients have no identifiable cause, generic history, or known diet issues. Doctors typically only diagnose patients when their symptoms persist and become more prevalent. However, one key symptom that starts earlier in patients with neurological disorders is dysarthria, a condition in which the muscles used for speech are weaker and patients have difficulty in controlling them. As a result, patients with neurological disorders are unable to say certain words or phrases or speak with a distinct vocal pattern that varies from people who do not suffer from these neurological disorders, causing issues with communication.  In the United States, someone suffers from a stroke every 40 seconds, and only about 10% of stroke victims recover completely, 25% recover with some minor impairments, but more than 40% have severe impairments needing special and long term care with limited ability to communicate to their loved ones. As the disease progresses, patients are sometimes left unable to do certain daily tasks and increasingly rely on technologies known as automatic speech recognition (ASR) software, such as Siri and Alexa. However, due to their changing vocal patterns, the software does not always recognize the needs and cannot assist the patient with completing daily tasks, making the quality of life for such patients harder.

Amyotrophic lateral sclerosis (ALS) is a neurodegenerative disorder that impacts the nervous system, affecting millions worldwide. However, diagnosing ALS early is difficult since symptoms are often confused for other disorders. ALS patients have difficulty completing simple tasks due to neuro-motor disruptions, making them dependent on automatic speech recognition (ASR) software for day-to-day tasks. Unfortunately, the strained, hypernasal voices of patients result in a 78% failure rate in such software. This project proposes a solution to identify auditory biomarkers for early diagnosis and a novel voice compensation method to improve voice recognition accuracy in ASR software. Auditory features were extracted and also used to generate images through a Mel-Frequency spectrogram. The auditory features and images were processed using an optimized Neural Network algorithm. The algorithm classifies voice files with an accuracy of 91% (auditory features) and 88% (spectrogram images). A new voice compensation algorithm was developed to adjust a word's duration, frequency, and pitch to improve the ASR recognition rate. This method successfully increases the recognition accuracy and allows ASR systems to understand voice commands from ALS patients easily. These methods can easily be applied to other neurological disorders to identify unique auditory biomarkers and create a voice compensation algorithm.

This solution serves several different populations, mostly focusing on the elderly population and people who do not live in close proximity to hospitals. Most people develop ALS between the ages of 40-70 (als.org). However, the process for diagnosing ALS is incredibly time-consuming and very intensive since there are multiple tests to undergo, including spinal taps and electromyography (EMG). The cost of these tests can be very high, and patients may be required to take such invasive and time-consuming tests over a period of months and potentially years, which adds up to time, money, and energy that not all patients may not have. As a result, this solution seeks to make a common type of ALS testing, using one's voice, more accessible. This solution allows patients to save time, money, and energy as they no longer have to commute to the hospital to consult with doctors as frequently and no longer have to take as many expensive medical tests, making ALS testing more accessible and affordable to more people, especially the elderly and people who may not have access to a hospital or cannot afford extensive doctor's appointments. Additionally, patients with ALS have a hard time communicating with others due to changes in their vocal patterns. This hard time communicating with people leads many patients to often suffer from poor quality of life since they are unable to voice their needs. As a result, it is important to provide an accessible tool for patients to communicate and perform daily tasks.

Equality in healthcare is truly an issue that I care about deeply, and I have done research and volunteer work related to my passion. I have done several projects related to using AI and Machine Learning techniques in the healthcare field, such as creating a project to predict Parkinson's Disease as well as another project to predict emotions in non-verbal patients. These projects have taught me many technical skills, such as coding and a basic understanding of machine learning, as well as personal skills, such as resilience and focus.

Additionally, I have done volunteer work with the Tamil Nadu Foundation where I implemented a women’s health project in India for high school students, culminating in the education of 2000+ girls. I educated such girls regarding the importance of menstrual health and donated nearly 3000+ pads to schools in India. I am currently working with the TNF organization to discuss ideas to expand this project all throughout Southern India as well as how to create a sustainable initiative to continue the education of future generations.

I am also currently pursuing degrees in Bioengineering and Business with the hopes of creating more projects dedicated to promoting equal access to healthcare. I want to do everything I can to ensure that no one needs to suffer the consequences of not having accessible and affordable healthcare solutions. With my desire to improve the world and to learn new skills to bolster my desire for change, I want to do my part to make a difference in countless peoples’ lives.

There are several reasons why I wanted to apply to Solve:

• Network of creators and innovators:
  • Solve would be a way for me to be connected to other creators and innovators who are also incredibly passionate about creating change in the world. Having a community of like-minded individuals is something I find valuable and Solve would help me connect with other individuals who would inspire and push me to continue innovating.
• Advice from professionals:
  • Creating meaningful change in our world is a long journey filled with struggles, and having a network of professionals and mentors would help me on my journey. Solve would be an opportunity to get to know other changemakers and get helpful advice from people that are following a similar path as me.
• Implementation of the solution
  • One of the primary barriers that currently exists is figuring out how I can implement this model in the real world through either an app or a device. I would first need to learn the technical skills needed to build this solution or find something with these skills before partnering with hospitals and organizations.
• Finding a team
  • Another main barrier is the current lack of a team working on this project. I want to start working with a team to offer different perspectives and have different skills to ensure that this final project is beneficial for everyone. I would want a team that could help me work on the technical side of things as well as do market research to understand what else is currently happening as well as find partner hospitals and organizations to work with.
• Liabilities
  • There potentially may be some liabilities when working with hospitals and organizations that I may need to take into consideration, ranging from HIPAA to the consent of patients. I want to ensure that there are no legal issues or liability issues when working with hospitals and organizations, so I want to have the legal expertise to ensure that there are no issues.

This solution is a new, revolutionary way for people to get diagnosed with ALS quickly and safely compared to the status quo. People will no longer have to spend countless hours waiting for the doctor and up to years to be diagnosed with ALS. By combining utilizing the power of machine learning to solve an unmet need, thousands, and potentially millions of people, could have access to a solution that can help them get the care they need. Utilizing machine learning to diagnose diseases beyond just ALS can allow millions of people to gain access to affordable and accessible healthcare solutions.

• Improve access to ALS testing for all people all over the world and quality of life for ALS patients
  • The primary objective of this project is to create systems to allow for the early diagnosis of ALS and to improve the quality of life for patients with ALS who primarily rely on ASR software to complete daily tasks. By creating tools to assist doctors in the early diagnosis of ALS, patients and doctors can take more measures to slow down the onset of ALS while maintaining independence by using ASRs to complete basic tasks

• Integrate this solution in communities around the world by creating an app or device with this technology
  • The primary way to integrate this solution in communities is by creating an app or device for people to connect to and use the test I created. With more and more people connected to the internet in recent years, more people can use tools that rely on software and the internet as long as they are provided with such tools.

• Partner with hospitals and organizations to provide more people with access to such technology
  • Hospitals and organizations that promote health equality have established connections with patients and other networks meant to support all patients with neurological disorders, including ALS. By connecting with hospitals and organizations, this solution can easily be integrated into existing networks where more people can be helped.

One primary metric I plan to use to measure my progress towards these goals is by looking at the number of users this solution has. Once partnering with local hospitals and organizations, this solution will have more users that have access to a simple app/device that they can then use to take these tests from the comfort of their home while being monitored by doctors and other healthcare professionals. Additionally, another metric I would like to use is the number of patients that have been diagnosed through this platform. The primary objective of this solution is to make ALS testing more accessible and affordable for everyone, and one such way to quantify this objective is by looking at how many diagnoses have taken place through this platform. Another metric I would also like to look at is the amount of time and money saved as a result of this solution. By gathering information about a user's location and the nearest hospital to them, the model can then use this information to estimate how much time and money was saved as a result of taking such tests at home rather than commuting to and from a hospital that may be inaccessible for people, especially the elderly. 

One of the primary activities my organization and team would work on is finding and partnering with local organizations and hospitals to plan which communities and patients to work with. This way, we can systematically work with communities that need this service in ways that benefit everyone. Working with local organizations and local hospitals would be the best plan of action since such groups would best know the needs of the community and would be able to provide recommendations for how to best go about the implementation of this service. By reaching out to local organizations, we can build upon this relationship by organizing workshops and events for people to get tested over a period of time in ways that minimally interfere with their daily activities. This way, we can continue to collect data and test the algorithm with more data to allow for real-time fine-tuning, but also provide more people with the service. This initial run can then serve as a proof-of-concept that applying machine learning can aid doctors in the diagnosis of diseases, and as a result, we can further develop additional tools to aid in the diagnosis of other diseases, such as Parkinson's Disease or Alzheimer’s. All in all, this solution can easily be used as a tool to create equitable and affordable healthcare solutions for all.

This project primarily relies on the use of different computer models and audio files to use audio data to make predictions on whether a patient has ALS or not. Using the Praat software, distinct audio features, such as Shimmer, Jitter, Pitch, Number of Pulses, Tempo, and Harmonicity, were extracted using a command-line script and observed for patients with and without ALS. Once this software filtered through the data, a Mel-Frequency Spectrogram process was used. The Mel-Frequency Spectrogram method involves using a Fourier Transformation on the raw voice file signals to convert it to several sets of images.

Another key component used in this solution is the Machine Learning component. Once the data has been cleaned through a series of filtering different features of a voice file (pitch, frequency, volume, etc.), the data is then fed through a machine learning algorithm, namely a stacked neural network model, that makes the final prediction of whether a patient has ALS or not. Additionally, the second aspect of this project relies on the creation of a "voice compensation" processing model that modifies an ALS patient's voice file by altering speed, pitch, and amplitude before sending this audio to an Automated Speech Recognition (ASR) software, such as Siri or Alexa. Once this modified audio file serves as an input for the ASR software, the software then better recognizes what a patient is asking for and can better suit their needs.

Growing up in a predominantly white neighborhood was hard. I would always get questions about my lunches or would be asked to say something in my “native” language despite English being my first language. It was hard for me to feel included at school when I was younger, but eventually, I found a group of friends who were supportive and inclusive. Finding this community of people who were willing to listen and celebrate cultural differences was such an important part of my development and growth, and as a result, I have become committed to providing the same sense of community for others. My personal experiences growing up as well as seeing major equity gaps in my communities have inspired me to take action to make society more inclusive and equitable. I am constantly learning how to create inclusive communities and promote unique voices. Bringing in unique and underrepresented voices is crucial to creating meaningful solutions and change in our society, and I am and will continue to be supportive of diverse and inclusive communities.

My main approach to creating an inclusive platform is to first create a team of diverse individuals who have worked in healthcare and are committed to creating an equitable and inclusive platform for all. Promoting diversity first begins with having a diverse team filled with people from all different backgrounds who are ready and committed to sharing their ideas to create an inclusive platform. Another crucial step, especially in technology-related projects, is acknowledging and correcting biases within the algorithm by collecting more data and training the algorithm to recognize and account for differences in language patterns and pronunciations. We can better train the algorithm with various data types by having a diverse team of people working on this project too. Another crucial step I want to take is listening to people and implementing their feedback into making a better product. This comes from making a transparent workplace and process to allow for all feedback to be processed and implemented into the final product.

The primary product that our business would provide is the application that allows patients to use voice-based tests to help diagnose ALS faster and more accurately. Along with this app, we would provide partnerships and workshops for organizations, hospitals, and patients to better understand the diagnosis process and get additional resources to navigate this process. We would also work with hospitals and organizations in real-time to provide them with resources to help analyze this information and also use such data (with patient approval) to improve the algorithm and continue improving accuracy. We would mostly provide this resource through creating mobile applications but may choose to expand to a website depending on user access and availability. With more and more people connecting to the internet now, we want to do our part in making sure that the internet provides people with tools to make their lives better.

Our primary revenue source will come from donations and grants in addition to selling this technology to organizations and hospitals. By creating partnerships with organizations and hospitals, we want to maximize the number of people who benefit from this technology.

This business is incredibly important as it helps doctors diagnose ALS, and potentially other diseases, much faster than they currently can. This process is less time-consuming and resource-intensive for both the doctors and patients. In the end, this technology allows doctors and patients to take early measures to treat and prevent the further onset of ALS, thus improving the quality of life for patients.

In the short term, we hope to fund our work through donations and grants received from various organizations and competitions/networking events. Such donations and funds should be enough to keep us operating and continuing to develop new innovative features in the application. In the long term, we hope to partner with organizations to help offset some of the costs associated with implementing it in communities and providing the technology people need to access this application. We also hope to sell this technology to hospitals and make some money by providing training to hospital staff and doctors to ensure that all patients that use this application can have their results and diagnoses verified. We also hope to raise investment capital from different sources and firms to ensure that all people can access this technology without any barriers and issues.

Currently, to fund the project, I have received various grants through science fairs and scholarships. Namely, the Society for Science has provided monetary support through the International Science and Engineering Fair and the Regeneron Science Talent Search. Additionally, The Biotechnology Institute has provided some monetary support through the BioGENEius Challenge competition.