unpAID

unpAID uses voice-based biomarkers and facial analysis to detect neurodegenerative diseases, specifically Parkinson’s, through a simple voice and video recording from the patient through their smartphone. 

In the United States alone, 90,000 people are diagnosed with Parkinson’s Disease. This diagnosis lasts for a year before confirmation and is expected to double in the U.S. by 2040. 

Many hear of Parkinson’s Disease (PD) to be the second leading neurodegenerative disease in our world, affecting 10 million people worldwide. PD is specifically difficult as it is related to age and there are minimal treatments which are used as the disease progresses. However, the earlier PD is diagnosed, the earlier it can be treated and ultimately the quality of life will be significantly higher. 

In fact, PD is a global problem. It's the main reason for a neurological consultation in Spain. 11% of all PD patients live in Kiribati, which is one of the least developed countries in the world.

But, one of the most mind blowing and obvious problems related to PD is the diagnosis. Symptoms appear a minimum of 1-2 years prior to diagnosis, linguistic changes can be observed as early as 5 years prior to diagnosis and multiple examinations are required for the diagnosis to be confirmed, taking from months up to 1 year long. Simply for a diagnosis. Even then, the diagnosis is done by a trusted neurologist through a 20-minute and 14 page long conversation between the patient and doctor. Meaning, there must be a specialist to make the diagnosis and it is subjective to the neurologist. All in all, this is an inefficient process requiring years and although PD is a slow moving disease, with early detection it can be eradicated with such minimal symptoms and years of lifestyle changes. 

The entire process is very subjective with a low accuracy rate. Specifically, to make an accurate diagnosis, doctors must weigh the symptoms and then see if they meet the clinical criteria of having 1+ symptoms. However, because there’s no conclusive screening or test, patients with very early PD might not even meet the clinical diagnosis criteria. This also means you could have a disease which mimics PD but be diagnosed and treated as PD. More inefficiency. 

The process begins with a neurology appointment where the conduct a physical exam according to the UPDRS (Unified PD Rating Scale - scroll down to see the document). This will lead to further exams and ultimately dopaminergic medications to increase the lack of dopamine associated with PD. However, if there’s a lack of response to medications, it will prompt the doctor to see alternative diagnoses. 

Overall, this is a trial and error method of diagnosing, and this must be changed.

The solution uses AI, specifically machine learning, to analyze an acoustic audio sample of PD patients saying “ahh” to detect PD. It is known that there are many soft signs which could be detected prior to the classical motor dysfunction of PD, including voice changes with frequent pausing, reduced volume and other patterns which can not be analyzed by the basic human ear. Although expert clinicians aim to analyze this, voice changes can occur up to 5 years prior to diagnosis meaning a diverse AI solution is needed to diagnose with high accuracy. 

This is done through a larger subsection known as voice based biomarkers. After analyzing the audio sample, roughly 10-13 features including jitter, pausing, articulation, etc. are analyzed and extracted from the sample. After the values are extracted, they are analyzed and plotting onto a spectrogram which is a plot of frequencies of a particular signal which is further analyzed through a CNN. 

However, after further validation, voice changes will not allow for definite diagnosis of PD and more classifiers are needed to diagnose the disease without the need of specialists. Specifically, a biometric model for detection of facial features through Google MediaPipe Face Mesh (another early soft sign of PD) coupled with the voice system will allow for a two-factor authenticated system for higher accuracy detection. 

Overall, a single-feature modality for accurately detecting early stage PD is less sensitive and can be mistaken for other diseases which mimic PD. We need an integrated multi domain biometric feature which is more sensitive and will result in an overall high accuracy. 

The current target population is anyone with a genetic history of Parkinson's/someone who is beginning to experience minimal symptoms or would like to get tested if they have Parkinson’s in America, Canada and the United States. These patients should have a smartphone to download the app with stable internet and be able to make the sound “ahhh.” As the model has not been trained on extensive data. Typically, voice analysis systems are built on linguistic data and require the patient to be of a specific race and speak a particular language. However, considering this model is acoustic, it does not matter which language the patient speaks or if they have an accent as they simply repeat the sound “ahhh.” 

The population is currently underserved because as many other patients around the world, PD diagnosing is a global problem and beginning with a group in America who has the necessary resources to speak and record them walking/talking are a great way to continuously train the model. Overall, with early detection, this can be further verified by a neurologist, would require less overall tests with remote diagnosing and would allow for decrease the effects of PD on patients’ lives. 

Many who have has family history of PD or are experiencing mild difficulties but are overlooking the problem are elongating the time in which they will be suffering with harsh symptoms. This solution will address thieir needs as if they are not meeting the clinical criteria to get tested for Parkinson’s or believe they might have something similar to the disease, they can go onto the app and do a quick voice analysis and facial feature analysis. This will also act as further validation for the doctors, knowing that there are soft signs present from vocal cords and the articulatory system. 

It might seem as though a 16 year old isn’t old enough or equipped to tackle a tough problem but its really the opposite! The younger, the more time on our hands and the more we are curious about. It’s currently a one person show for this project however, I have extensive experience in deep tech, specifically nanotechnology project, have been the youngest Innovation Intern at Walmart Blue Labs (Walmart’s Canada’s Innovation Incubator), spoke at WebSummit (the world’s largest tech conference) in Lisbon, Portugal at 16 years old, have had two other internships at Academia labs and startups working on gene therapies, have competitively played sports and continue to develop myself on a daily basis, built a project on anti-counterfeiting using quantum dots which was on route to implementation in Europe. 

I participate in numerous hackathons during my free time away from school or projects. One of the one month long hackathons was when I was the project manager at 15 years old for a consulting team of defining the future of retail for Walmart Canada, ultimately leading me to pitch to the CEO of Walmart alongside 500+ Walmart executives and this led to my internship. I also participated in a moonshot hackathon to create a moonshot company inspired by GoogleX, all about off-shore wind turbines which led to my team receiving 3/4 awards. 

That being said for each project, I was getting my reps in. That means presenting at the world’s largest tech conference as a teenager and interning at the world’s largest retailer while creating QuantumTags, I learned and acquired numerous skills and failed during the journey as well. This means after going through these experiences, I know what works and what doesn’t and have built a network of mentors and experts (which is continuing to grow) to build this project further. Although I know its always difficult to trust a single person, especially someone so young if you close off the fact that I’m 16 years old and look at everything I accomplished while in an IB school in just one year, my commitment and drive is what got me here. The funding is an investment in my project and my drive, and I guarantee it’ll be another to add to the list. 

For all of my projects and talks, you can check out: https://tks.life/profile/pavi.dhiman#portfolio 

Although this project is still being iterated on, the main points of validation have been with doctors and neurologists as well as the Chief Medical Officers in hospitals ranging from Canada to clinics in the U.S. Specifically, this solution will be an add-on to their current process until it has a high enough accuracy to be the process. However, the main point of validation is if the process can be automated with as little bias or subjective opinions, then it will also be a more streamlined and accurate process. In the beginning, the system was solely based on voice analysis however, adding the facial biometric solution will add another layer and more iteration will allow for higher accuracy. 

Once more of the model is built out, the model will gather new data from various hospitals and see the accuracy. That’s when I’ll run the pilot program, continuing to iterate and build the entire app. 

As mentioned earlier, the current diagnosis pathway for PD is not efficient nor accurate and with an aging population in America, the prevalence of this disease is expected to double in the next decade. Additionally, when thinking about the problem on a global scale, there’s a lack of neurologists around the world meaning people live with PD, don’t know they live with it and therefore, cannot make the correct lifestyle changes. 

The closest technology has got to diagnosing Parkinson’s is through wearable sensors. Luckily, physicians can analyze these biometric features remotely, including those in underdeveloped countries without access to a specialist. Apps have been developed for phone or watch but they focus on detecting arm swing movement pattern/related motor symptoms. However, this solution requires active participation and is expensive which hinder the widespread use in a large population. Overall, the global distribution is hindered due to price, however, the use of physicians treating remotely is the exact unique selling point of this company, which will be integrated into our solution. Plus, since our solution is a multi-domain feature detection, it allows for greater accuracy and done all through a smartphone allows for minimal expenses on the user side. 

Compared to the need of having a specialist on hand or having an expensive digital solution, these have been some of the main hindering factors to widespread digital solutions being implemented for PD diagnosis, however, our solution accounts for both of these while also considering the future of the population and Parkinson's.

1. Implement unpAID into 2 clinics centred around neurodegenerative diseases in Canada/the U.S. 

This first impact goal ties in with the second one but is the first one needed for all of the other impact goals to begin. Implementing this solution into 2 clinics will allow for the app to receive more data, find bottlenecks in the solution and iterate quickly while testing the solution in a real world environment. This implementation will begin with simple conversations with doctors currently validating my solution and see if they would be willing to try it, alongside a small incentive to push the product out if there is minimal luck. This would require a working model, website and content surrounding the product. 

2. Have a minimum of 50 (healthy, prospective and already diagnosed) users who have been diagnosed accurately. Run this as a pilot program. 

This second goal of a pilot program goes hand in hand with implementation in 2 clinics. This pilot program will be for families or anyone wishing to participate as there will be healthy patients (can be the family of the PD patient), someone with family history and finally a PD patient for the patients attending these clinics and will test them at home. This will allow for both qualitative and quantitative feedback. In this case, success looks like answering the following questions and going back to iterate:

- Is the UI friendly/gives a caring feeling when getting diagnosed? Is it similar to a doctor?

- For PD patients, what’s the main difference you feel when getting diagnosed by the doctor vs. the AI model?

- If you could add or remove 2 features on the app, what would it be?

- If you received a first time diagnosis from the app, how do you think it would feel?

- How much time did it take you to complete the test?

3. In the next 5 years, implement unpAID in developing countries suffering from neurodegenerative diseases.

When I first began this journey of understanding the broken healthcare system and saw the impact of voice based biomarkers on non-invasive diagnosing, I immediately thought of how this could work in developing countries who have a lack of doctors. However, those specific diseases and implementation would take almost a decade with no guarantee. However, after knowing the prospects of neurodegenerative diseases, specifically where I live: the implementation is more accessible, the use is compatible with the population (using smartphones) and if approved, it can make a global impact. However, neurodegenerative diseases are still affecting those in developing countries, therefore an impact goal is to target them and their specific diseases as well as provide them with next steps for treatment. 

4. After validation of the first product and a trial run, soon add on models to diagnose 2 other neurodegenerative diseases (ie, Alzheimer’s, Huntington’s Disease, etc.) and have the app running as a cohesive system. 

Parkinson’s is not the only neurodegenerative disease out there, in fact there are hundreds of others which have an impact on people’s lives. After successfully diagnosing Parkinson’s, running the pilot program and beginning to implement the product, a new add-on will be diagnosing more diseases so that the app will almost act like a specialized neurologist. However, this is currently a moonshot.

The core technology used in unpAID is AI through voice analysis and audio samples. Voice based biomarkers are a new application of AI, with many academia continuing to research. One of the main and most researched correlation between voice and body is the brain. 

Diseases are unpredictable; they can affect all sorts of organs from different methods and “forms of attack,” each disease might target one primary organ, but it impacts other traditional body functions. The organs might be the heart, lungs, brain, muscles or even the vocal folds, altering someone’s voice. We unlock incredible new opportunities by analyzing these voice changes with AI’s help. From diagnosing to risk prediction to complete remote monitoring, it is incredible.

Voice-based biomarkers (VBBs) work is based on the human voice, our rich medium with a complex array of sounds coming from our vocal cords. Surprisingly enough, our vocal cords contain crucial information for diagnosing diseases. Considering the shift already, all phones or home devices contain some form of a vocal assistant and have allowed considerable use of voice-controlled search; in fact, 31% of smartphone users already use voice technologies once a week. Plus, the evolution of this technology combined with audio-signal analysis, Natural Language Processing (NLP) and understanding allow for this potential new application for diagnosis, classification and remote monitoring through VBBs, which will increase the number of people using their vocal assistants.

This is mainly because the human voice produces multiple frequencies. However, the human ear can only hear and understand a narrow sequence. But through AI, we can register frequencies across the entire spectrum. Vocal biomarkers refer to any signature, feature or combination of features from voice signals associated with clinical outcomes to measure a condition and generate the severity of a disease.

But before we can predict the future of voice-based biomarkers, let’s understand each aspect, starting with the indicator, your vocal cords.Consider your vocal cords to be a ballet dancer. We all know dancing is one of the human activities requiring immense training, skill and expertise. After hours of practice and numerous recitals, the top ballet dancers must constantly perfect their technique. With its upsides, dancing can have its downsides; dancers frequently get injured.From broken bones to neurological disorders, these can stop the dancers’ careers in a split second. Neurological disorders like Parkinson’s especially can also destroy this extraordinary talent. With millions of people worldwide suffering from this incurable weakness, tremors and terrible disease, we haven’t progressed in the right direction.

The issue we face with Parkinsons’ is the lack of early-stage biomarkers detectable by humans. The closest thing we have to this is a 20-minute neurologist test which is $300. But what if we could do this at home? Now, it’s the same concept with your voice. Again, consider your vocal cords to be a ballet dancer. A “vocal cord dancer” must coordinate all of their vocal cords to make sounds. But we take lots of training, and from sound, we can track the vocal fold position as it vibrates, specifically for Parkinson’s. Because just as the limbs are affected in Parkinson’s, so are the vocal cords. And this is all because speaking is a byproduct of a very complex system. We use our lungs, vocal cords, tongue, lips, nasal passages, and brain every time we speak. And so, any disease, injury or medical event with these systems leaves diagnostic clues and biomarkers.

However, combining any digital microphone, precision voice analysis software and the latest machine learning advancements can quantify exactly where someone lies between health and disease. For more, read my article on voice based biomarkers: https://pavidhiman22.medium.com/hey-alexa-whats-my-diagnosis-def9484a22bb 

Currently, our solution is serving zero people as I have created systems to detect diseases but there is limited data out there when it comes to Parkinson’s patients and their characteristics. 

In the next year, I aim to directly work with 50 users (healthy, prospective and affected) for a pilot program across two clinics through roughly 10 doctors. This will be a new step in the diagnosing process meaning doctors, patients and families are all affected with doctors and patients being the most directly affected. 

The main existing barrier is a technical problem of a lack of data. Before running a pilot program, I need a working model for both the biometric system and the voice system and the only way to do this is through pre-existing data online. However, this is a new type of system and does not have data for both parts. However, my current plan is to create similar systems on similar data types to show doctors and prospective users that “this is what the system will be able to do.” After this, I’ll be able to get data directly from their hospitals, clinics, etc. and go from there. 

With that comes the financial problem. Being able to implement and even receive data requires some financial incentive. Further paying for building the app and working prototype also requires money I don’t have as I’m currently still in high school. 

A possible cultural barrier is my age. Trusting the youth for medical-related concerns is difficult due to the trust needed in the field. However, I have worked on this solution for a few months and have extensive experience which I continue to bring to this project. That being said, I also need more validation through companies and patients which will help clinics see the impact this solution can make. I plan to get through this barrier by building such a model with a high accuracy rate but also get high quality validation to support the project.

Currently do not partner with any organization but am in talks with/aiming to talk with the following companies to find an intersection to work within. Companies/instituations: MIT AI, Mayo Clinic, Cleveland Clinic, UHN, Harvard Voice Group.

Although the project is still underway and am making updates and continuing to build the models out, in terms of value impact for the customers, this will act as a sense of accuracy in their lives. As a PD patient, diagnosis is subjective and there’s a guaranteed blood test or a long MRI scan you can use to check if you have the disease or what stage it’s at. So instead, being able to make the diagnosis through your cellphone is accuracy and will allow you to check updates with your progression. This has been a needed solution for years with hundreds of people working on curing the disease, however the patients currently suffering or getting diagnosed, they require a form of confirmation and alongside their doctor, this tool will act as that. 

As per revenue, clinics will first be able to pay to use the models to better diagnose and run check ups on patients. This will allow for a double layer for them. In return, they will pay us and a small portion will be them providing us with data to further improve the models and add more classifiers and features. Eventually when enough money has generated and there is a well running cash flow; alongside added classifiers, the goal is to add this to existing phones like Samsung and Apple as their health app and further spread remote diagnosis through partnerships.

The model and path for financial sustainability is through an embedded and low income client market. The ability to provide access to neurodegenerative disease diagnosis to those who typically cannot afford it is the end goal. The app will begin with being used by institutions and gain funding through organizations helping innovators and entrepreneurs, and then further expanding into a common use for everyone with access to a stable internet connection and a camera.